Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Community-based stream and groundwater monitoring and future change impact modelling of a socio-ecohydrological… Hund, Silja Verena 2018

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2019_february_hund_siljaverena.pdf [ 126.13MB ]
Metadata
JSON: 24-1.0374285.json
JSON-LD: 24-1.0374285-ld.json
RDF/XML (Pretty): 24-1.0374285-rdf.xml
RDF/JSON: 24-1.0374285-rdf.json
Turtle: 24-1.0374285-turtle.txt
N-Triples: 24-1.0374285-rdf-ntriples.txt
Original Record: 24-1.0374285-source.json
Full Text
24-1.0374285-fulltext.txt
Citation
24-1.0374285.ris

Full Text

Community-based stream andgroundwater monitoring and futurechange impact modelling of asocio-ecohydrological system to informdrought adaptation in theseasonally-dry tropicsbySilja Verena HundDiplom (M.Sc.), University of Potsdam, 2011A THESIS SUBMITTED IN PARTIAL FULFILLMENT OFTHE REQUIREMENTS FOR THE DEGREE OFDOCTOR OF PHILOSOPHYinThe Faculty of Graduate and Postdoctoral Studies(Geological Sciences)THE UNIVERSITY OF BRITISH COLUMBIA(Vancouver)November 2018© Silja Verena Hund 2018The following individuals certify that they have read, and recommend to the Faculty of Graduate andPostdoctoral Studies for acceptance, the dissertation entitled:Community-based stream and groundwater monitoring and future change impact modelling of asocio-ecohydrological system to inform drought adaptation in the seasonally-dry tropicssubmitted by Silja Verena Hund in partial fulfillment of the requirements forthe degree of Doctor of Philosophyin Geological SciencesExamining Committee:Mark S. Johnson, Earth, Ocean and Atmospheric Sciences, Institute for Resources, Environment andSustainabilitySupervisorDouw G. Steyn, Earth, Ocean and Atmospheric SciencesSupervisory Committee MemberDiana M. Allen, Earth Sciences, Simon Fraser UniversitySupervisory Committee MemberJeanine M. Rhemtulla, Forest and Conservation SciencesUniversity ExaminerJordi Honey-Rosés, Community and Regional PlanningUniversity ExamineriiAbstractWith a changing climate and a growing population, droughts are becoming more frequent in manywatersheds across the world, necessitating new approaches to improve water security in communities.Drought is typically caused by a combination of hydrological and social drivers. In this thesis, I apply theemerging framework of socio-hydrology (coupled human-water systems). Specifically, the objective ofthis thesis is to assess current and future socio- and hydrological dynamics and impacts on surface waterand groundwater supplies with the goal of informing drought adaptation and improving water security.I focused hereby on two drought-prone rural watersheds in the seasonally-dry tropics of Costa Rica.Using a community-based approach, I implemented a hydrological monitoring network of streams andgroundwater with open-source data loggers, and worked with local communities to assemble societalwater use data. I then synthesized the watersheds in a hydrological model and assessed current social(water use) and hydrological vulnerabilities to drought. Results showed that communities dominantlyrelied on groundwater supplies, and that a temporal mismatch between water availability and needs, highdomestic water use, and increasing extraction rates are increasing pressure on groundwater. Resultsalso indicated high streamflows during the wet season, and thus a potential to increase surface wateruse while streamflows are high. Next, I explored the impacts of the El Niño Southern Oscillation (ENSO),future climate change and water use change on water resources in the study watersheds. During anextreme El Niño, groundwater recharge and streamflow decreased by 60% relative to ENSO Neutral. Ialso found that future climate change may lead to groundwater storage decline, especially if combinedwith high population growth. In the seasonally-dry tropics, wet season rainfall is essential for recharginggroundwater that serves as primary water supply during long dry seasons. Therefore, I developed anovel ‘groundwater recharge indicator’ as a tool to support water managers to respond adaptively toreduced wet season rainfall and increase socio-hydrological resilience to seasonal drought. Overall,this thesis contributes to the field of socio-hydrology and provides novel approaches to improve watersecurity under drought in the seasonally-dry tropics.iiiLay SummaryDrought is a problem in many watersheds and it is often caused by a combination of reduced rainfalland high human water extraction. In the seasonally-dry tropics of Costa Rica, recent droughts have ledto water conflicts and emergency declarations. In this thesis, I focused on two watersheds of this region,and investigated water supplies (streams and groundwater), water use and climate change impacts withthe goal to increase the resilience to drought. I measured streamflows and groundwater levels in thefield and worked with local communities. I also used a virtual watershed software to explore potentialclimate change impacts, and found that groundwater levels may decline with climate change. Lastly, Ideveloped a novel tool for water managers to support adaptation to droughts.ivPrefaceThis thesis presents the original work of the author, with contributions and research publications asindicated below.Throughout the research presented here, Drs. Mark Johnson, Douw Steyn, and Diana Allen providedsupport, and feedback on the chapters of this thesis.The data logger and monitoring station design in Chapter 2 were developed with support from Dr. MarkJohnson and Tom Keddie (who designed the micro-electronic setup), and this work has been publishedin:Hund, S.V., Johnson, M.S., Keddie, T., 2016. Developing a Hydrologic Monitoring Network in Data-Scarce Regions Using Open-Source Arduino Data loggers. Agric. Environ. Res. Lett. 1, 1–5.Field monitoring and data processing of the Eddy Covariance station in the Potrero and Caimital water-sheds was conducted by Dr. LauraMorillas. Long-term rainfall data was obtained by Dr. DouwSteyn andJennifer Romero Valpreda through the FuturAgua project. Collaboration within the FuturAgua projectprovided contact to local stakeholders and water agencies and organization of stakeholder workshops.The field monitoring was completed with help from Grethel Rojas Hernandez.Climate change scenarios in Chapter 5 were developed with support from Dr. Iris Grossmann, andENSO scenarios with support from Dr. Douw Steyn, and Drs. Diana Allen and Mark Johnson alsoprovided feedback on the research of this chapter.The groundwater recharge analysis (Chapter 6) was developed in discussion with Drs. Mark Johnsonand Diana Allen, and Dr. Laura Morillas also provided feedback on the chapter, which has been pub-lished as:Hund, S.V., Allen, D. M., Morillas, L., and Johnson, M.J., 2018. Groundwater recharge indicator as toolfor decision makers to increase socio-hydrological resilience to seasonal drought. Journal of Hydrology.563 (2018) 1119-1134.The above publication also contains parts of Chapter 3 on the hydrological model setup.vTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiiLay Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvList of Acronyms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxList of Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxiiAcknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxiiiDedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxv1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1 Water, humans, and droughts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2 Objective and research questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41.3 Drought-prone Guanacaste in Costa Rica . . . . . . . . . . . . . . . . . . . . . . . . . . . 61.4 Research approach and significance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71.5 Overview of thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8vi2 Community-based monitoring of streams and groundwater using open-source dataloggers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102.1 Field site . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102.2 Community-based monitoring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172.3 Open-source data logger development . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192.4 Hydrological monitoring network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232.4.1 Stream monitoring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232.4.2 Groundwater level monitoring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252.4.3 Meteorological monitoring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262.5 Water use data assembly . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263 Hydrological model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283.1 Hydrological modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283.2 Conceptual model overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293.3 Software selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313.4 Process conceptualization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353.5 Model setup and input parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393.5.1 Model time steps and initial modelling period . . . . . . . . . . . . . . . . . . . . . 393.5.2 Spatial distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403.5.3 Meteorological input data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453.5.4 Evapotranspiration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 463.5.5 Soil input parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 483.5.6 River reaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 523.5.7 Groundwater . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 523.5.8 Water demands by dierent sectors . . . . . . . . . . . . . . . . . . . . . . . . . . 543.6 Model calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 633.7 Model evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67vii4 Assessing current socio-ecohydrological vulnerabilities to drought to improve watersecurity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 744.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 744.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 784.3 Socio-ecohydrological dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 784.3.1 Water use dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 784.3.2 Hydrological dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 874.3.3 Emerging vulnerabilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 944.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 975 Impacts of ENSO-driven climate variability, future climate change and growing waterdemands on water resources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 985.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 985.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1025.2.1 Climate context of Guanacaste . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1025.2.2 Overview of scenarios and modelling . . . . . . . . . . . . . . . . . . . . . . . . . . 1065.2.3 Baseline data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1095.2.4 ENSO scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1095.2.5 Climate change scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1115.2.6 Weather generator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1195.2.7 Population scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1215.2.8 Water demand scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1225.2.9 Hydrological modelling & post-processing . . . . . . . . . . . . . . . . . . . . . . . 1235.2.10 Uncertainties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1245.3 Climate and water use impacts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1265.3.1 ENSO impacts on water resources . . . . . . . . . . . . . . . . . . . . . . . . . . . 1265.3.2 Climate change impacts on water resources . . . . . . . . . . . . . . . . . . . . . 1295.3.3 Multiple interacting drivers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1375.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143viii6 Groundwater recharge indicator as tool for decisionmakers to increase socio-hydrologicalresilience to drought . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1456.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1456.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1476.3 Groundwater recharge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1486.3.1 Observed groundwater levels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1486.3.2 Modelled groundwater recharge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1496.3.3 Socio-hydrological resilience . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1516.3.4 Groundwater recharge indicator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1546.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1597 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166AppendicesA Chapter 2 supporting information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187A.1 Field site photos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187A.2 Stream monitoring stations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198A.3 Recorded stage data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199A.4 Rating curve development . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203A.5 Groundwater monitoring stations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213B Chapter 3 supporting information (model) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216C Chapter 3 supporting information (water use) . . . . . . . . . . . . . . . . . . . . . . . . . . 220D Chapter 5 supporting information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225D.1 The El Niño Southern Oscillation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225D.2 ENSO scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225ixD.3 Climate change scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232D.4 Population scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237D.5 Weather generator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238xList of TablesTable 2.1: Land use/land cover of the Potrero and Caimital watersheds. . . . . . . . . . . . . . 16Table 2.2: Hydrological and social considerations for selecting monitoring site locations. . . . 18Table 2.3: Location, elevation and upslope area of stream monitoring sites. . . . . . . . . . . . 24Table 2.4: Location and elevation at groundwater monitoring sites. . . . . . . . . . . . . . . . . . 26Table 3.1: Criteria for hydrological model selection. . . . . . . . . . . . . . . . . . . . . . . . . . . 32Table 3.2: Advantages and disadvantages of potential modelling software. . . . . . . . . . . . . 34Table 3.3: Change in land use/land cover between 2005, 2010, and 2016. . . . . . . . . . . . 42Table 3.4: HRU classifications. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44Table 3.5: Sources for meteorological data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46Table 3.6: List of soil input parameters. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49Table 3.7: Groundwater-related parameters. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53Table 3.8: Annual population estimates for rural villages. . . . . . . . . . . . . . . . . . . . . . . . 55Table 3.9: Geometric annual growth rates for population in rural villages. . . . . . . . . . . . . . 55Table 3.10: Annual mean of Runo Resistance Factor (RRF). . . . . . . . . . . . . . . . . . . . . 64Table 3.11: Soil layer 1 (top layer) depth. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64Table 3.12: Saturated hydraulic conductivity of soil layer 1. . . . . . . . . . . . . . . . . . . . . . . 65Table 3.13: Preferred flow direction of soil layer 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65Table 3.14: Parameters for soil layer 2 in basalt zone adjusted during calibration. . . . . . . . . . 66Table 3.15: Groundwater related parameters adjusted during calibration. . . . . . . . . . . . . . 66Table 3.16: Total annual evapotranspiration for dierent Kc scaling factors. . . . . . . . . . . . . 67Table 3.17: Goodness-of-fit measures between observed and modelled daily mean streamflow. 69xiTable 4.1: Annual per person water consumption rates. . . . . . . . . . . . . . . . . . . . . . . . 79Table 4.2: International comparison of annual domestic water use per person . . . . . . . . . . 80Table 4.3: Annual water extraction volumes from the Potrero and Caimital watersheds. . . . . 80Table 4.4: Percentages of water supply sources in 2005 and 2016. . . . . . . . . . . . . . . . . 81Table 4.5: Increase of water use in dry season relative to wet season. . . . . . . . . . . . . . . . 84Table 4.6: Ratio of total annual evapotranspiration relative to total annual rainfall. . . . . . . . . 86Table 4.7: Annual modelled mean and range of water flows from 2005 to 2016. . . . . . . . . . 88Table 4.8: Mean and range of annual runo coecients. . . . . . . . . . . . . . . . . . . . . . . . 93Table 4.9: Dierence between annual groundwater recharge and extraction from 2005 to 2016. 94Table 5.1: Factors of monthly change for climate change scenarios. . . . . . . . . . . . . . . . . 115Table 5.2: Climate change scenarios selected for hydrological modelling. . . . . . . . . . . . . . 115Table 5.3: Population estimates from Shared Socio-economic Pathways (SSP) for Costa Rica. 121Table 5.4: Percentage change of water flows for future climate scenarios relative to baseline. 130Table 5.5: Annual total domestic water demand for baseline and population scenarios. . . . . 137Table 6.1: Observed total annual rainfall and modelled groundwater recharge. . . . . . . . . . 150Table A.1: Start and end dates of stream monitoring stations. . . . . . . . . . . . . . . . . . . . 198Table A.2: Start and end dates of groundwater monitoring stations. . . . . . . . . . . . . . . . . 198Table A.3: Latitude and longitude of monitoring stations. . . . . . . . . . . . . . . . . . . . . . . . 198Table A.4: Downstream Potrero River site (SW1) stage and discharge measurements. . . . . . 206Table A.5: Downstream Potrero River site (SW1) rating curve parameter estimates. . . . . . . 206Table A.6: Upstream Potrero River site (SW3) stage and discharge measurements. . . . . . . 208Table A.7: Upstream Potrero River Site (SW3) power law rating curve parameter estimates. . 208Table A.8: Upstream Potrero River Site (SW3) linear model parameter estimates. . . . . . . . 209Table A.9: Upstream Caimital River site (SW4) stage and discharge measurements. . . . . . . 210Table A.10: Upstream Caimital River site (SW4) power law rating curve parameter estimates. . 210Table A.11: Upstream Caimital River site (SW4) linear model parameter estimates. . . . . . . . 211xiiTable A.12: Downstream Caimital River site (SW5) stage and discharge measurements. . . . . 212Table A.13: Downstream Caimital River site (SW5) rating curve parameter estimates. . . . . . 212Table A.14: Lithology logs nearby the Varillal Groundwater site (GW1). . . . . . . . . . . . . . . . 214Table A.15: Lithology logs nearby the Dulce Nombre Groundwater site (GW2). . . . . . . . . . . 214Table B.1: Mean eective water holding capacity for dierent geology and land use cover. . . 216Table C.1: Annual domestic water use rates in Caimital. . . . . . . . . . . . . . . . . . . . . . . . 220Table C.2: Annual municipal/business water use rates in Caimital. . . . . . . . . . . . . . . . . . 220Table C.3: Annual population and geometric growth rates for Nicoya and Hojancha. . . . . . . 221Table C.4: Water use per household and per person for AyA Nicoya. . . . . . . . . . . . . . . . . 221Table C.5: Annual household water use in Nicoya from 2005 - 2016. . . . . . . . . . . . . . . . . 221Table C.6: Municipal and business water use in Nicoya from 2005 - 2016. . . . . . . . . . . . . 222Table C.7: Water demand and supply sources in Nicoya from 2005 - 2016. . . . . . . . . . . . . 222Table C.8: Water use and supply for Hojancha, as serviced by AyA Hojancha. . . . . . . . . . 223Table C.9: Population and annual water use for Hojancha in 2016. . . . . . . . . . . . . . . . . . 223Table C.10: Volumes of water needed for washing of melons after harvesting. . . . . . . . . . . 223Table C.11: Melon irrigation in the Potrero and Caimital watersheds. . . . . . . . . . . . . . . . . . 224Table C.12: Estimated water use for cattle. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224Table C.13: Additional MINAE water licenses in the Potrero and Caimital watersheds. . . . . . . 224Table D.1: Model fitting parameters for double Gaussian model (Steyn et al. 2016). . . . . . . . 226Table D.2: Annual ENSO classification (Steyn et al. 2016). . . . . . . . . . . . . . . . . . . . . . 227Table D.3: Mean monthly total rainfall for baseline and ENSO scenarios. . . . . . . . . . . . . . 228Table D.4: Monthly rainfall change factors for ENSO scenarios. . . . . . . . . . . . . . . . . . . 229Table D.5: Mean monthly air temperature for baseline and La Niña/El Niño. . . . . . . . . . . . 229Table D.6: Mean monthly solar radiation for baseline and La Niña/El Niño. . . . . . . . . . . . . 230Table D.7: Mean monthly wind speed for baseline and La Niña/El Niño. . . . . . . . . . . . . . . 230Table D.8: Mean monthly relative humidity for baseline and La Niña/El Niño. . . . . . . . . . . . 231xiiiTable D.9: The six selected general circulation models (GCM). . . . . . . . . . . . . . . . . . . . 232Table D.10: Monthly rainfall for baseline, climate change scenarios and GCM6 mean. . . . . . . 235Table D.11: Monthly climate change scenario for air temperature (2074 - 2100) . . . . . . . . . . 236Table D.12: Monthly climate change scenario for relative humidity (2074 - 2100). . . . . . . . . . 236Table D.13: Climate change scenario for incoming solar radiation (2074 - 2100). . . . . . . . . . 237Table D.14: Population predictions based on socio-economic pathways (SSPs) and INEC data. 237Table D.15: Population scenarios for rural villages. . . . . . . . . . . . . . . . . . . . . . . . . . . . 238Table D.16: Municipal and business scenarios for rural villages. . . . . . . . . . . . . . . . . . . . 238Table D.17: Population and municipal/business sites for Nicoya and Hojancha. . . . . . . . . . . 238Table D.18: Comparison for observed and LARS-WG generated mean of monthly total rainfall. 238xivList of FiguresFigure 1.1: The Province of Guanacaste, Costa Rica, Central America. . . . . . . . . . . . . . . 6Figure 1.2: Overview of thesis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9Figure 2.1: Location of the Potrero and Caimital watersheds in Guanacaste, Costa Rica. . . . . 11Figure 2.2: Topography, rivers and aquifer in the Potrero and Caimital watersheds. . . . . . . . 12Figure 2.3: Monthly total rainfall for Nicoya from 1980 to 2016. . . . . . . . . . . . . . . . . . . . 13Figure 2.4: Geology of the Potrero and Caimital watersheds. . . . . . . . . . . . . . . . . . . . . . 14Figure 2.5: Land use/land cover of the Potrero and Caimital watersheds. . . . . . . . . . . . . . 15Figure 2.6: Water demand of the Potrero and Caimital watersheds. . . . . . . . . . . . . . . . . . 16Figure 2.7: Monitoring setup at the Potrero river. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21Figure 2.8: Time series of stream stage. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22Figure 2.9: Monitoring site locations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23Figure 2.10: Upslope area for stream monitoring stations. . . . . . . . . . . . . . . . . . . . . . . . 24Figure 3.1: (Overview) conceptual model of dominant processes. . . . . . . . . . . . . . . . . . . 30Figure 3.2: Landscape-based conceptual model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31Figure 3.3: Simplified WEAP flow structure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36Figure 3.4: WEAP rainfall-runo conceptualization. . . . . . . . . . . . . . . . . . . . . . . . . . . 37Figure 3.5: Modelled stream reaches and sub-catchments. . . . . . . . . . . . . . . . . . . . . . . 41Figure 3.6: Slope categories for HRUs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43Figure 3.7: Spatial distribution of sub-catchments (SCs) and hydrological response units (HRUs). 44Figure 3.8: MODIS ET pixels in Potrero and Caimital watersheds. . . . . . . . . . . . . . . . . . . 47Figure 3.9: Daily Kc values for melon, rice, forest and pasture. . . . . . . . . . . . . . . . . . . . 48xvFigure 3.10: Soil depth at lithology logs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50Figure 3.11:Mean leaf area index (LAI) for forest, pasture and crops. . . . . . . . . . . . . . . . . 51Figure 3.12:Water use at ASADA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56Figure 3.13:Observed water depth at artisanal well GW1 in Varillal. . . . . . . . . . . . . . . . . . 57Figure 3.14: Area of melons under production for each day of the 2015 to 2016 melon season. . 60Figure 3.15:Observed and modelled daily mean discharge at SW1. . . . . . . . . . . . . . . . . . 69Figure 3.16:Observed and modelled daily mean discharge at SW3. . . . . . . . . . . . . . . . . . 70Figure 3.17:Observed and modelled daily mean discharge at SW4. . . . . . . . . . . . . . . . . . 70Figure 3.18:Observed and modelled daily mean discharge at SW5. . . . . . . . . . . . . . . . . . 71Figure 3.19: Scatterplots between observed and modelled daily mean discharge. . . . . . . . . . 71Figure 3.20: Comparison of modelled and measured/MODIS evapotranspiration. . . . . . . . . . 73Figure 4.1: Monthly water extraction rates for dierent sectors between 2005 and 2016. . . . . 82Figure 4.2: ASADA Caimital seasonal household water use from 2012 to 2015. . . . . . . . . . 83Figure 4.3: Daily water use rates for artisanal use and for towns for dry and wet season. . . . . 84Figure 4.4: Mean total monthly modelled evapotranspiration for dominant land use types. . . . 87Figure 4.5: Monitored daily mean groundwater table depth for GW1, GW2 and GW3. . . . . . . 89Figure 4.6: Daily mean streamflow (observed and modelled). . . . . . . . . . . . . . . . . . . . . 91Figure 4.7: Examples of observed stormflow response to rainfall at the Potrero river. . . . . . . 92Figure 4.8: Seasonal flow duration curves based on modelled daily streamflow. . . . . . . . . . 92Figure 4.9: Socio-ecohydrological vulnerabilities to drought. . . . . . . . . . . . . . . . . . . . . . 95Figure 5.1: Overview of climate and water use impact modelling approach. . . . . . . . . . . . . 108Figure 5.2: ENSO monthly rainfall scenarios. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110Figure 5.3: Meteorological input data for ENSO scenarios. . . . . . . . . . . . . . . . . . . . . . . 111Figure 5.4: Monthly meteorology means for observed and GCM data. . . . . . . . . . . . . . . . 113Figure 5.5: Rainfall scenarios for 2074 - 2100. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116Figure 5.6: Wind speed scenarios for 2074 - 2100. . . . . . . . . . . . . . . . . . . . . . . . . . . 117Figure 5.7: Temperature scenario for 2074 - 2100. . . . . . . . . . . . . . . . . . . . . . . . . . . . 118xviFigure 5.8: Relative humidity and solar radiation scenarios for 2074 - 2100. . . . . . . . . . . . . 119Figure 5.9: Overview of modelled hydro-climate and social water demand scenarios. . . . . . . 123Figure 5.10: Sketch of uncertainties in modelled socio-hydrological system. . . . . . . . . . . . . 125Figure 5.11: Dierence of annual totals for ENSO scenarios relative to ENSO Neutral. . . . . . 127Figure 5.12:Monthly means for ENSO scenarios & flow duration curves. . . . . . . . . . . . . . . 128Figure 5.13: Daily rainfall and streamflow time series for climate change scenarios. . . . . . . . 130Figure 5.14: Climate change implications for water resources (annual totals). . . . . . . . . . . . 131Figure 5.15:Groundwater volume in storage for climate change scenarios. . . . . . . . . . . . . . 133Figure 5.16:Monthly means of water flows for climate change scenarios. . . . . . . . . . . . . . 134Figure 5.17:Monthly percentage change for streamflow and flow duration curves. . . . . . . . . . 135Figure 5.18:Monthly water demands for low and high population scenarios. . . . . . . . . . . . . 138Figure 5.19: Dierences between annual groundwater recharge and annual groundwater demand 140Figure 5.20:Monthly groundwater storage for low and high population scenarios. . . . . . . . . . 141Figure 5.21:Groundwater storage for climate change and socio-hydrological feedback. . . . . . 142Figure 6.1: Observed water table depths for two wells in the Potrero-Caimital aquifer. . . . . . . 149Figure 6.2: Observed annual rainfall and modelled groundwater recharge. . . . . . . . . . . . . . 150Figure 6.3: Conceptual non-resilient and resilient socio-hydrological systems. . . . . . . . . . . 152Figure 6.4: Groundwater recharge indicator. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155Figure 6.5: Step-by-step instructions for use of the groundwater recharge indicator. . . . . . . . 156Figure 6.6: Overall accuracy for prediction of groundwater recharge category. . . . . . . . . . . 158Figure 7.1: Key concepts, research questions, methods and results of this research. . . . . . . 164Figure A.1: Photo of Downstream Potrero River site (SW1) at baseflow. . . . . . . . . . . . . . . 187Figure A.2: Photo of Downstream Potrero River site (SW1) during low stormflows. . . . . . . . . 188Figure A.3: Photo of Downstream Potrero River site (SW1) during high stormflows. . . . . . . . 188Figure A.4: Photo of Tributary to Potrero River site (SW2). . . . . . . . . . . . . . . . . . . . . . . 189Figure A.5: Photo of data logger at Tributary to Potrero River site (SW2). . . . . . . . . . . . . . . 189xviiFigure A.6: Photo of Upstream Potrero River site (SW3). . . . . . . . . . . . . . . . . . . . . . . . 190Figure A.7: Photo of Upstream Potrero River site (SW3) with PVC tube for monitoring sensor. 190Figure A.8: Photo of Upstream Potrero River site (SW3) data logger. . . . . . . . . . . . . . . . . 191Figure A.9: Photo of Upstream Caimital River site (SW4). . . . . . . . . . . . . . . . . . . . . . . . 191Figure A.10:Photo of Upstream Caimital River site (SW4) surrounding rice fields. . . . . . . . . 192Figure A.11:Photo of Upstream Caimital River site (SW4) data logger. . . . . . . . . . . . . . . . 192Figure A.12:Photo of Downstream Caimital River site (SW5) during baseflow. . . . . . . . . . . . 193Figure A.13:Photo of Downstream Caimital River site (SW5) during stormflow. . . . . . . . . . . . 193Figure A.14:Photo of Downstream Caimital River site (SW5) data logger. . . . . . . . . . . . . . . 194Figure A.15:Photo of Varillal Groundwater Monitoring site (GW1). . . . . . . . . . . . . . . . . . . 194Figure A.16:Photo of Dulce Nombre Groundwater Monitoring site (GW2) . . . . . . . . . . . . . . 195Figure A.17:Photo of Dulce Nombre Groundwater Monitoring site (GW2) with CTD sensor. . . . 195Figure A.18:Photo of Gamalotal Groundwater Monitoring site (GW3). . . . . . . . . . . . . . . . . 196Figure A.19:Photo of irrigated melon fields during the dry season. . . . . . . . . . . . . . . . . . . 197Figure A.20:Photo of rainfed rice fields during the wet season. . . . . . . . . . . . . . . . . . . . . 197Figure A.21:Cross sections of channel geometry at monitoring sites. . . . . . . . . . . . . . . . . 199Figure A.22:Observed stage at the Downstream Potrero River site (SW1). . . . . . . . . . . . . . 200Figure A.23:Observed stage at the Upstream Potrero River site (SW3). . . . . . . . . . . . . . . . 201Figure A.24:Observed stage at the Upstream Caimital River site (SW4). . . . . . . . . . . . . . . 202Figure A.25:Observed stage at the Downstream Caimital River site (SW5). . . . . . . . . . . . . . 203Figure A.26:Example of discharge measurements at SW1. . . . . . . . . . . . . . . . . . . . . . . 204Figure A.27:Downstream Potrero River site (SW1) rating curve. . . . . . . . . . . . . . . . . . . . . 207Figure A.28:Upstream Potrero River site (SW3) rating curve. . . . . . . . . . . . . . . . . . . . . . 209Figure A.29:Upstream Caimital River site (SW4) rating curve. . . . . . . . . . . . . . . . . . . . . . 211Figure A.30:Downstream Caimital River site (SW5) rating curve. . . . . . . . . . . . . . . . . . . . 213Figure A.31:Groundwater monitoring sites (GW) and location of lithology logs . . . . . . . . . . . 215Figure B.1: MODIS mean and standard deviation for forest pixels. . . . . . . . . . . . . . . . . . . 217xviiiFigure B.2: MODIS mean and standard deviation for pasture pixels. . . . . . . . . . . . . . . . . . 217Figure B.3: Modelled evapotranspiration and observed/MODIS ET (Kc factor 1.5). . . . . . . . . 218Figure B.4: Modelled evapotranspiration and observed/MODIS ET (Kc factor 1.7). . . . . . . . . 219Figure D.1: GCM comparison for historical baseline period. . . . . . . . . . . . . . . . . . . . . . . 233Figure D.2: GCM comparison for 2074 - 2100. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234Figure D.3: Rainfall scenarios for 2074 - 2100 (Grossmann et al 2018). . . . . . . . . . . . . . . . 235Figure D.4: Comparison for observed and LARS-WG generated mean of monthly total rainfall. 239xixList of AcronymsAcronym NameACT Área Conservación Tempisque (part of MINAE)ASADA Rural Water Boards in Costa Rica (Asociacionesadministradoras de los Sistemas de Acueductos yAlcantarillados comunales)AyA National Water Utility Agency in Costa Rica (InstitutoCostarricense de Acueductos y Alcantarillados)BCSD Bias Correction and Spatial DownscalingCATIE Tropical Agricultural Research and Higher Education Centerin Costa Rica (Centro Agronómico Tropical de Investigación yEnseñanza)CMIP4 Coupled Model Intercomparison Project Phase 4CMIP5 Coupled Model Intercomparison Project Phase 5CTD Electrical conductivity, water temperature and water depthsensor (by METER Group Inc., formerly Decagon DevicesInc.)CTI Technical Interinstitutional Committee for Water Statistics inCosta Rica (Comité Técnico Interinstitucional Estadística delAgua)DEM Digital Elevation ModelDHI Danish Hydrologic InstituteEC Eddy CovarianceENSO El Niño Southern OscillationET Actual EvapotranspirationFREEWAT FREE and open source software tools for WATer resourcemanagementGCM General Circulation ModelGCM6 Climate change scenario used in research (mean of 6selected General Circulation Models)GIS Geographic Information SystemGSFLOW “Groundwater and Surface-Water Flow” modelGSM Global Systems for Mobile communicationsGUI Graphical User InterfaceGW GroundwaterHRU Hydrological Response UnitHydroSHEDS Hydrological data and maps based on SHuttle ElevationDerivatives at multiple ScalesHYLUC Hydrological Land Use Change modelKMNI Koninklijk Nederlands Meteorologisch InstituutIIASA International Institute for applied Systems AnalysisIMN National Meteorological Institute in Costa Rica (InstitutoMeteorólogico Nacional)xxAcronym NameINEC National Census and Statistics Institute in Costa Rica(Instituto Nacional de Estadistica y Censos)IPCC Intergovernmental Panel on Climate ChangeITCZ Inter-Tropical Convergence ZoneMAPSS Mapped Atmosphere Plant Soil SystemMIKE SHE Hydrological model based on the Système HydrologiqueEuropéen (SHE)MINAE Ministry of Environment and Energy in Costa Rica (Ministeriodel Ambiente y Energía)MODFLOW Modular Groundwater Flow ModelNAM Nedbor Afstromnings ModelNicoyAgua Non-governmental organization in Costa RicaNOAA National Oceanic and Atmospheric Administration (U.S.Department of Commerce)NSE Nash-Sutclie EciencyONI Oceanic Niño IndexPEST Parameter ESTimationPET Potential EvapotranspirationPRECIS Providing REgional Climates for Impacts StudiesPRMS USGS Precipitation-Runo Modeling SystemQGIS Open-source Geographical Information Systems softwareR Open-source coding languager Pearson correlation coecientR2 Coecient of determinationRCM Regional Circulation ModelRCP Relative Concentration PathwayRSME Root Mean Square ErrorRSR Ratio of RMSE over the standard deviationSAGA-GIS System for Automated Geoscientific AnalysisSC Sub-catchment in hydrological modelSD Secure DigitalSENARA National Groundwater Agency in Costa Rica (ServicioNacional de Aguas Subterráneas, Riego y Avenamiento)SINAC National System of Conservation Areas in Costa Rica(Sistemas Nacional de Áreas de Conservación)SRES Special Report on Emission ScenariosSW Surface waterSWAT Soil and Water Assessment ToolUNA National University of Costa Rica (Universidad Nacional deCosta Rica)USGS United States Geological SurveyVIC Variable Infiltration Capacity modelWEAP Water Evaluation and Planning ToolWhitebox GAT Geospatial Analysis ToolWL Weatherlink stationxxiList of SymbolsSymbol Name UnitA Wetted area of cross section at stage m2Ad Aquifer depth at equilibrium mas Ångström coecient -bs Ångström coecient -D Vapor pressure deficit kPD Temperature gradient of saturated vapor pressure kPCd Wetted depth of stream mET Actual evapotranspiration mm dy1ETreƒ Potential reference crop evapotranspiration mm dy1f Scaling factor for Kc -G Soil heat flux mm dy1GS Groundwater storage m3GSe Equilibrium groundwater storage m3h Stage (stream water depth) mhd Horizontal distance from farthest edge of aquifer to stream mKc Crop coecient -ks Saturated hydraulic conductivity of soil layer 1 mm dy1ks,q Saturated hydraulic conductivity of aquifer mdy1l Wetted length of aquifer in contact with stream mn Manning’s roughness coecient sm1/3n N1 Cloudiness fraction -n Bright sunshine hours per day horsN Total day length horsPe Eective precipitation mm dy1P Wetted perimeter of cross section at stage mPFD Preferred Flow Direction -Q Stream discharge m3 s1R Hydraulic radius mRd Total eective storage for soil layer 1 mmRn Net radiation exchange for crop cover mm dy1RRF Runo Resistance Factor -S Energy gradient (slope) of river reach -St Total incoming short-wave solar radiation MJm2 dy1Sy Specific yield of aquifer -S0 Extraterrestrial radiation MJm2 dy1U2 Wind speed at 2 m m s1 Psychometric constant kP°C1yd Height of current groundwater storage above equilibriumstoragemz1 Relative storage in soil layer 1 as fraction of total eectivestorage Rd-xxiiAcknowledgementsI am immensely grateful to my supervisor Mark Johnson for all his support throughout this project, whilealso giving me room to grow and explore at the same time, for inspiration and a great experience workingtogether. I also thank DouwSteyn for his always kind support asmy PhD committeemember and throughthe FuturAgua project, and for letting me benefit from his vast experience and wise outlook on research.I also kindly thank Diana Allen for her important role in my PhD committee, for sharing her expertise ongroundwater hydrology and for letting me participate in her lab meetings.I gratefully acknowledge financial support for this research through the Canadian Natural Science andEngineering Research Council (NSERC; G8PJ-437336-2012). Research was conducted as part of theFuturAgua project established through the Belmont Forum Freshwater Security and Coastal Vulnerabilityframework. A Social Sciences and Humanities Research Council of Canada (SSHRC) Connection Grantto the UBC FuturAgua team also supported communication of research results to stakeholders. I alsogratefully acknowledge a research stipend from CATIE (Centro Agronómico Tropical de Investigacióny Enseñanza), the Egil H. Lornthzen Scholarship and W.H. Matthews Scholarship from the Earth,Ocean and Atmospheric Sciences Department at UBC, and a Faculty of Science PhD Tuition Awardand International Tuition Award from UBC.I would like to thank the many stakeholders and kind people from Guanacaste who made this projectpossible, shared their local knowledge with me and welcomed me with open arms to their region. Ikindly thank Xinia Campos Palma and Alvar Campos de Lemos, as well as the other members of ourstakeholder group (in particular, Carlos Calvo, Nelson Gamboa Araya, Elizabeth Fernández, and AndreaSuarez Serrano) and organizations (Área Conservacion Tempisque (ACT), the Fundación NicoyAgua,the Dirección de Gestión Ambiental of the town of Nicoya, and the Universidad Nacional de Costa Rica)for their continuous and passionate support throughout this project, their participation in workshops andmeetings, and for facilitating access to data. I thank the many water agencies throughout Guanacasteand Costa Rica who shared data and information. In particular, I thank Max Gomez from the nationalwater agency AyA (Instituto Costarricense de Acueductos y Alcantarillados) in Nicoya for access to waterextraction data, the Direccion de Agua from the Ministerio del Ambiente y Energía (MINAE) for accessto water licensing data, the Comité Técnico Interinstitucional (CTI), composed of the Dirección de Agua,AyA and the national groundwater agency SENARA (Servicio Nacional de Aguas Subterráneas, Riego yxxiiiAvenamiento) for granting access to groundwater level data, and Pamela Garcia Serrano, Clara Agudeloand Carlos Romero (SENARA) for facilitating the data. I thank the Instituto Meteorológico Nacional (IMN)in San José for access to historical rainfall data. I thank the rural water board (ASADA) presidents fromCurime-Varillal, La Virginia, Samara and Caimital who met with me and shared their knowledge, and inparticular Danilo Rojas from Caimital for letting me digitize water use bills.This project would not have been possible without the friendly Guanacaste families who let me installmonitoring stations on their land (often right next to their own house), made sure the stations were safe,and provided an open door (and many a chat) throughout my numerous field visits. I thank MarielosBarrantes, Carmen Lopez and her family (Jennifer, Kevin, Oscar, and the late Alfonso Lopez), BettyVargas, Manuel Vargas Araya and the Costeña farm, Evelyn Aguilar, Alvar Campos, Blas Campos, andRafael Quiros). I thank my local field assistant and friend Grethel Rojas Hernandez, who always kept ingood humour in the field, shared insights from Guanacaste, and looked well after the monitoring stationswhen I was not in the field. I also thank Siria and Luis Obando for a good home to stay in Nicoya.I thank the FuturAgua team - it was a wonderful opportunity to work with such an interdisciplinary groupand I learned a lot from all of you. In particular, I thank Laura Morillas for being a great team partnerand her support with my project, Jennifer Romero Valpreda and Pavel Bautista Solis for all their greatwork as project coordinators, Tim McDaniels for helpful discussions and for leadership of the FuturAguateam, Iris Grossmann for climate change discussions, and Alejandra Echeverri, Paige Olmsted, ClaudiaCastro Miravalles and Matthew Babcock for the good times spent together in the field. I also thank TomKeddie for his kind support in developing the Arduino data loggers.I thank all current and former Ecohydro Lab members, especially Ashlee Jollymore, Brenda D’Acunha,Cameron Webster, Iain Hawthorne, Michael Lathuilliere, Mollie McDowell, Morgan Haines, and YeonukKim, as well as many other IRES/EOAS/UBC students, especially Mekdes Ayalew Tessema, TugceConger, Jackie Yip, and Lucy Rodina. I thank the Institute for Resources, Environment and Sustain-ability (IRES) and the Department of Earth, Ocean and Atmospheric Sciences (EOAS) communitiesfor inspiring me from many dierent interdisciplinary directions, and the EOAS and IRES sta for theiradministrative support. I also kindly thank Tamsin Lyle for inspiration and mentorship.I thank my Vancouver family Amanda, Ana, Colin, Christina, Rafael and Sarah for making me feel athome in Vancouver, and the girls, Franka, Ramona and Brynn for always bringing a smile to my face. Ialso thank André, Hayley, Kathryn, Magnolia and Vernon for embracing me into their family. I thank myextended family (my Oma, my aunts/uncles and cousins) and my friends (Angela, Anja, Astrid, Damaris,Juliane, Helena, Kathrin, Larissa and Maria) who supported me from across the world from Germany,England and Brazil. I am especially grateful to my parents Renate and Wolfgang and my sisters Corinnaand Alexandra who were always there for me along this journey, and to Colin for all his continuous loveand encouragements.xxivDedicationTo my grandfather Dipl. Ing. Josef Welter (1911 - 2017),who inspired me towards this path.To Colin, my parents and my sisters.xxvChapter 1Introduction1.1 Water, humans, and droughtsWater and humans are intimately connected, and water with its many facets between too little and toomuch has always shaped human lives. For thousands of years, humans have also influenced the flowof water through the environment (Biswas 1970; Dermody et al. 2014; Kuil et al. 2016; Montanari et al.2013; Said 1993). With the rise of the Anthropocene (Crutzen 2006), however, the inter-relation betweenwater and humans is becoming more important than ever, as many watershed processes are now drivenby feedback between social and natural system components (Montanari et al. 2013). Variability ofthe climate, and thus, of water flows in streams and groundwater, has always been part of naturalsystems, but human-induced climate change has been accelerating in recent decades and changingthe hydrological cycle towards more frequent extremes of droughts and floods (IPCC 2014). Combinedwith rising water demands of a growing population and water management practices that have not yetbeen adapted to a changing climate and changing demands, this has led to water security challengesfor many communities across the world (Hoekstra et al. 2018; Wada et al. 2011a; Wada et al. 2014).An example is the Pacific coast of Central America, where seasonal changes between intense wetseasons and long dry seasons have historically required societies to live between too much and toolittle water. Fast population growth along with agricultural intensification, water governance challenges,and recurrent meteorological droughts have caused wide-spread water insecurities in recent years(Ballestero et al. 2007; Esquivel-Hernández et al. 2017a; Hidalgo et al. 2015b; Kuzdas et al. 2015a).The El Niño Southern Oscillation (ENSO) is one of the major drivers of rainfall variability in the region,and El Niño conditions tend to lead to intensified droughts when wet season rainfall is substantiallyreduced (Steyn et al. 2016). This happened during the El Niño from 2014 to 2015, when nationaldrought emergencies were declared, 3.5 million people were food insecure in the seasonally-dry tropicsof Central America, and many community pumping wells ran dry (FAO 2016; Vignola et al. 2018).Groundwater is typically the primary water resource available in the region during the long dry seasonwhen streams have low baseflows or are dry if they are not interacting with groundwater (Ballestero et al.2007). The intense rainfall of the wet season replenishes aquifers in the tropics through groundwater1recharge (Jasechko and Taylor 2015; Sánchez-Murillo and Birkel 2016), which is essential for providingwater supplies during the following dry season. But reduced groundwater recharge in years of lowerwet season rainfall, in combination with continued high extraction rates, can lead to groundwater levelsthat are below the depth of many community pumping wells, thus leading to water shortages. The highstreamflows of the wet season renew surface water supplies and also contribute to aquifer rechargevia seepage through streambeds and floodplains. While it has been recognized for many years thatgroundwater and surface water should be managed as a single water resource (Winter et al. 1998), bothin practice and in scientific research often only one of the two components is addressed and activelymanaged.Eective water management in the developing countries of Central America is further hampered bysparse hydrological and water use data (Westerberg 2011). Data on groundwater supplies are rare and,similarly to many other regions around the world (Famiglietti 2014), the hidden nature of groundwaterhas led to over-extraction and mismanagement (Ballestero et al. 2007). Some monthly measurementsof streamflow exist, but high-frequency automated monitoring of streamflow is sparse, yet needed tocapture fast stormflow runo typical of the region and to allow full assessment of water supplies andchanges. Information on extraction rates from surface water and groundwater resources for use bydierent sectors also has many unknowns. Understanding of water needs in relation to water availabilityis, however, an essential first step for improving water security in the current system (Srinivasan et al.2017), and is also needed to prepare for a future in which meteorological droughts in the region are likelyto become more frequent under climate change (Imbach et al. 2018).A number of climate studies have explored the complex climate of the region and how it may changein the future. Overall, these indicate a likely a transition towards a drier climate (Hidalgo et al. 2013;Imbach et al. 2018; Rauscher et al. 2008). However, the translation of these climate change scenariosinto impacts on surface water and groundwater is limited, and if existent, only addresses surface waterimpacts at regional scales across all of Central America (Hidalgo et al. 2013; Imbach et al. 2012).Thus, the propagation of meteorological droughts (rainfall deficit) under climate change to hydrologicaldroughts (groundwater and streamflow deficit), and further, the societal role in drought propagation(human-induced drought) (Van Loon et al. 2016b) has so far not received much attention in the scientificliterature of this region.New approaches are needed to address the interacting drivers of changes in societal water demands andwater supplies and to improve water security under drought. The emerging concept of socio-hydrologyoers a holistic framework under which these two system components of hydrology and society canbe analyzed (Sivapalan et al. 2012). Socio-hydrology argues for explicitly including societal actionsinto hydrological analysis, and recognizing society and water as one coupled system in which thetwo sub-components interact and co-evolve (Montanari et al. 2013; Sivapalan et al. 2012; Thompson2et al. 2013; Wagener et al. 2010). As such, it draws on the concept of socio-ecological systems(Folke 2006; Folke et al. 2005; Troy et al. 2015), which describes the interactions between societyand ecological (natural) systems. Most research in socio-hydrology to date has focused specifically onhuman-water systems. Yet, humans not only manipulate water in streams and groundwater, but alsoaect evapotranspiration processes through land use management decisions. While evapotranspirationand natural water needs constitute important water demands in most watersheds, they are often notexplicitly included in socio-hydrology. Falkenmark and Folke (2002) first coined the term of socio-ecohydrology, and referred to water in streams and groundwater as blue water, while water availablein soil moisture for evapotranspiration is termed green water (Falkenmark and Folke 2002; Falkenmarkand Rockström 2006; Lathuillière et al. 2016). Socio-ecohydrology blends the terms ecohydrology (i.e.,interaction between ecological systems and hydrology; Dolman et al. 2014) and socio-hydrology, buthas only been rarely used, such as in a study on evapotranspiration from rainwater harvesting pondsin India (Van Meter et al. 2016), and in a discussion on urban water challenges (Pataki et al. 2011).However, the term highlights the diverse water demands and supplies that exist in human-dominatedwatersheds, and which play important roles in particular under drought conditions.Another similar concept to socio-hydrology is the hydrosocial cycle that highlights the many ways thatwater shapes human lives, and similarly to socio-hydrology, its proponents argue for joint analysis ofhuman and water systems (Linton and Budds 2014; Swyngedouw 2009).To date, much focus of socio-hydrology has been on developing stylized and conceptualized modelsof water and human interactions for predicting possible trajectories of system evolution (Mostert 2018;Roobavannan et al. 2018). For instance, a stylized model characterized a ’pendulum swing’ between asocietal focus on agriculture and economic gains (with resulting environmental degradation) to a focuson ecosystem health, which was attributed to shifts in ’environmental awareness’ (Roobavannan et al.2018; Van Emmerik et al. 2014). Other stylizedmodels have described feedback between water scarcity,city water supply systems, and human water use (Srinivasan 2015) or between societies and floods insettled floodplains (Di Baldassarre et al. 2013). In these stylized conceptual models, societal actionsare often portrayed indirectly through proxies such as ’environmental awareness’ (Van Emmerik et al.2014) or ’community sensitivity’ (Elshafei et al. 2014) that aim to describe the shifts in human normsand values leading to changes in observed human-water interactions (Roobavannan et al. 2018).Recent research has called for a more field-based approach to socio-hydrology (Massuel et al. 2018).By moving from theoretical models of socio-hydrological systems to field-based and on-the-groundresearch, researchers can learn from local knowledge of societal water dynamics, which are essentialfor understanding complex human – water interactions. At the same time, researchers can ensure thattheir research is situated within the context of the local community, and that research results can supportcommunities in, for instance, developing adaptation strategies to hydrological hazards such as droughts.3In this way, socio-hydrology expands to community-based research. Community-based research hasbeen defined as research in close collaboration with community members that is change-orientated andaddresses a community-identified need (Hall et al. 2015; Strand et al. 2003). An important componentof community-based research is also the dissemination of knowledge, including the communication ofscientific results to decision makers (Hall et al. 2015; Strand et al. 2003).In a community context, hydrological field monitoring also gains new dimensions. Monitoring has tobe integrated within the community, and both hydrological and social considerations have to be con-sidered when implementing monitoring networks in watersheds that are populated and influenced byhuman activities, in contrast to the natural and undisturbed watersheds that have been studied in tradi-tional hydrology. Recent developments of open-source data recorders (data loggers) can support suchcommunity-based monitoring due to their low cost and ease of use (Buytaert et al. 2014; Paul et al.2017).1.2 Objective and research questionsDrought and water security problems in many watersheds of the seasonally-dry tropics of Central Amer-ica require new approaches to move towards more drought-resilient communities in the region. Hydro-logic monitoring is an essential first step towards improving socio-hydrological knowledge for droughtadaptation in the data-scarce region, providing information on current dynamics of streamflow andgroundwater. This can then be combined with information on societal water needs to assess the currentsystem for vulnerabilities to drought. Yet, the climate is changing and population is growing rapidly in thisregion, necessitating investigations of potential future impacts of climate change and water use changeon water resources to inform adaptation strategies to drought. In a community-based context, it is furtherimportant to communicate scientific results to water managers in the form of hands-on tools to supportthem in increasing the societal resilience to drought.Therefore, in this thesis, my overall research objective is to assess current and future socio-hydrologicaldynamics and impacts on streamflow and groundwater supplies, with the overarching goal to provideinformation that can guide the future development of drought adaptation strategies of rural communitiesin seasonally-dry tropical watersheds of Central America, with specific focus on two seasonally-drywatersheds in Costa Rica. While there are many socio-hydrological inter-relations, this research focuseson societal water use as a first step towards water security. I approach my research objective throughthe following three research questions, aimed at investigation of the current system (research question1), the future system (research question 2) and drought adaptation (research question 3). Each of theseresearch questions is addressed through sub-questions as steps to guide the analysis:4Current System1. What are the current socio-ecohydrological dynamics with respect to water use, and what vulner-abilities to drought emerge? (Chapter 4)• What is the current annual and seasonal domestic, agricultural and evapotransporative wateruse?• And, on the other hand, what are the key hydrological annual and seasonal dynamics inrelation to water needs?• Further, what socio-ecohydrological vulnerabilities to drought emerge from this analysis, andwhat key management issues for drought adaptation and improvement to water security canbe identified?Future System2. What are the impacts of current inter-annual climate variability (ENSO), future climate change andgrowing water demands on streamflow and groundwater, and how do those impacts change ifsocio-hydrological feedback between climate and water use is considered? (Chapter 5)• How does inter-annual climate variability driven by ENSO impact water resources of thepresent and near future (~next decade)?• What are the implications of end of the 21st century climate change scenarios on streamflowand groundwater recharge?• What eects do the multiple drivers of climate change and a growing population have onwater resources?Drought Adaptation3. How can water managers increase the resilience of their communities to seasonal drought andprepare in years of reduced rainfall for the oncoming dry season? (Chapter 6)• What is the relation between seasonal rainfall and groundwater recharge for the Potrero-Caimital aquifer?• How can the socio-hydrological resilience of the system to seasonal drought be character-ized?• And what decision support tool could support local water managers for increasing the socio-hydrological resilience of communities and prepare in years of reduced rainfall for the oncom-ing dry season?51.3 Drought-prone Guanacaste in Costa RicaTo answer these research questions, I focused on the seasonally-dry tropics in Costa Rica, and specif-ically, on two watersheds within the Province of Guanacaste in Northwestern Costa Rica (Figure 1.1).As is the case throughout much of the region, Guanacaste has experienced significant drought impactsin recent years, which have led to increasing water conflicts between agriculture, tourism, and domesticwater users (Kuzdas et al. 2015a; Esquivel-Hernández et al. 2017a; Ramírez-Cover 2008).Figure 1.1: The Province of Guanacaste and the study watersheds (Potrero and Caimital watersheds),Costa Rica, Central America. Background: Open Street Map Stamen; Natural Earth Data.I conducted this research as part of a large and inter-disciplinary research project called “FuturAgua”based in Guanacaste that aimed to contribute scientific understanding to help improve the freshwatersecurity in the region (McDaniels et al. 2013; Romero Valpreda et al. 2017). FuturAgua was deeplyrooted in the local community through a wide network of stakeholders and local partners that supportedand participated in the project, and thus provided a good background for community-based research.During stakeholder workshops in the initial stage of my research project, the Potrero and Caimitalwatersheds in Guanacaste (Figure 1.1) emerged as important water supplies of concern for manysurrounding communities. A strong desire was expressed by local stakeholders to increase scientific6understanding of water dynamics in these watersheds. In the Potrero and Caimital watersheds (whichshare the Potrero-Caimital aquifer), water conflicts have emerged between agriculture and domesticwater users, and recent droughts have had significant impacts on the water security of communities.The rural agricultural communities depend on their local surface water and groundwater resources, andany changes in water supplies have direct consequences on their livelihoods. Given the intense humanuse of these watersheds for commercial and smallholder agriculture combined with water extractionfor rural towns, the Potrero-Caimital aquifer and overlaying watersheds provide an example of a socio-hydrological system. Their mixed land use of agricultural crops, pasture, forest, and rural villages is alsotypical for the region. Thus, the Potrero and Caimital watershed oered a good field site location for myresearch.1.4 Research approach and significanceI approached my research questions through a combination of field data collection and hydrologicalcomputer-based modelling under the overall framing of socio-hydrology. First, as available hydrologicaldata were limited in my study watersheds, I installed a hydrological monitoring network of streams andgroundwater levels in the Potrero and Caimital watersheds. For this, I cooperated with local stakeholdersand community members, and developed open-source data loggers for hydrological monitoring. I alsoworked with stakeholders and community members to assemble water use data in order to gain a betterunderstanding of water demand in the region. Next, I set up a hydrological computer model of the twowatersheds and aquifer to assess impacts of current annual climate variability (driven by the El NiñoSouthern Oscillation), future climate change and changes in water use on water resources, and alsoexplored potential socio-hydrological feedback between water use and a drier climate. Lastly, I used myresearch to develop a tool for water managers in the region that can support preparing for droughts.This thesis thus seeks to contribute to the emerging field of socio-hydrology, using a community-basedresearch approach. Further, this thesis aims to generate new knowledge on socio-hydrological dynamicswith respect to water use in rural watersheds of the seasonally-dry tropics, and provide a basis fordeveloping future drought adaptation strategies. My thesis work also aims to advance the managementof surface water and groundwater as a unified water resource by addressing both in a combined anal-ysis, as scientific research is still often fragmented between surface water and groundwater resources.Another goal of the thesis is to contribute new knowledge on climate change impacts on water resourcesin the seasonally-dry tropics of Central America, where currently, limited translation of climate changescenarios to impacts on streamflow and groundwater exists. Lastly, the thesis aims to contribute to thecommunication of scientific information to decision makers through the development of a novel droughtadaptation tool.71.5 Overview of thesisThis thesis consists of three main parts (Figure 1.2):1. Developing an understanding of the current system, with focus on hydrological dynamics andsocietal water use patterns.2. Modelling potential future impacts of climate change and population growth on water resources.3. Developing a drought adaptation tool.Specifically, in Chapter 2, I describe my study watersheds in more detail, and provide background onthe implementation of the hydrological monitoring network and assembly of water use data, includinga description of the community-based approach and development of open-source data loggers. InChapter 3, I describe the setup and evaluation of the hydrological computer model of the two watershedsand their underlying aquifer. In Chapter 4, I explore current system dynamics, using data obtained inthe field (hydrological monitoring and water use data), as well as model results of the current system.In Chapter 5, I use the hydrological model to explore the impact of a range of climate and water usescenarios on water resources and discuss potential consequences for communities. In Chapter 6, Idevelop a “groundwater recharge indicator” tool as support for local water managers in preparing forseasonal droughts. Figure 1.2 provides an overview of my thesis.8Figure 1.2: Overview of thesis.9Chapter 2Community-based monitoring ofstreams and groundwater usingopen-source data loggersThis chapter provides first a brief field site overview of the two study watersheds (Potrero and Caimitalwatersheds) and their shared aquifer. Further, it describes how this research project was basedwithin thecommunities of the region and how interaction with local stakeholders was important for the hydrologicalmonitoring program. As I conducted the monitoring with open-source data loggers, I also describe thedevelopment of these data loggers in this chapter. This is followed by details on the implementationof the stream and groundwater level monitoring network and the installation of a meteorological stationoperated by research partners. Lastly, I describe the assembly of water use data from the Potrero andCaimital watersheds.2.1 Field siteThe Potrero and Caimital watersheds are located in the Province of Guanacaste in northwestern CostaRica and share the same aquifer (Potrero-Caimital aquifer) (Figure 2.1). The surface water dividebetween the two watersheds crosses through the relatively flat alluvial valley that is underlain by theaquifer (Figure 2.2). Considering that the Potrero-Caimital aquifer extends through both watersheds, andthat groundwater extraction, surface water – groundwater interactions and other hydrological processeswill aect and be aected by processes in both watersheds, I include both watersheds in the analysis.10Figure 2.1: Location of the Potrero and Caimital watersheds in Guanacaste, Costa Rica. Mean annualrainfall was modelled by Karp et al. (2017).Surface waterThe Potrero watershed (36 km2) drains to the northeast into the Grande River, and the Caimital water-shed (41 km2) drains to the southwest into the Quirimán River, a tributary to the Nosara River (Figure2.2). The two river mainstems (the Potrero and Caimital rivers) are groundwater-fed and perennial, buthave low baseflows during the dry season. Ephemeral tributaries contribute to streamflow from thehillsides during the wet season and are dry during the dry season (Figure 2.2).11Figure 2.2: Topography, rivers, and aquifer in the Potrero and Caimital watersheds. Digital ElevationModel (DEM) from HydroSHEDS (Hydrological data and maps based on Shuttle Elevation Derivatives atmultiple Scales; Lehner et al. (2008); ~90 m resolution). River files from Area Conservación Tempisque(ACT). Aquifer extent from Servicio Nacional de Aguas Subterráneas, Riego y Avenamiento (SENARA).Elevation contour lines from Municipalidad de Nicoya.ClimateThe Province of Guanacaste has a wet-dry (seasonally-dry) tropical climate. Mean total annual rainfall inNicoya (nearby the Potrero and Caimital watersheds; Figure 2.1) is 2130 mm (mean from 1980 to 2016;ranging from 1310 mm to 3230 mm) (Figure 2.3), and the Province of Guanacaste is influenced by theseasonal variation of the Inter-tropical Convergence Zone (ITCZ). During the southerly position of theITCZ from November to March, the dominant climate forcings are the downwelling northeast trade windsthat bring warm and dry conditions over the Central American Cordillera east of Guanacaste (Steyn et al.2016; Waylen et al. 1998). In April, when the ITCZ migrates northward and Guanacaste is within therange of the convergence zone of the tropics, the wet season is initiated and weather is dominated byheavy convective rainstorms.12Figure 2.3: Monthly total rainfall for Nicoya from 1980 to 2016. Boxplots depict the median (horizontalline within box), the first and third quartile (extents of central box), range of data (“whiskers” above andbelow the box), and outliers that are located outside of 1.5 times the interquartile range. Compositedataset from the Nicoya Instituto Meteorólogico Nacional (IMN) station, the Weatherlink station and theEC station.The wet season is interrupted by a period of lower rainfall in July and August, a mid-summer droughtknown locally as “El Veranillo de San Juan” (Figure 2.3). The mid-summer drought and the double-peaked annual rainfall cycle are features of the southwest coast of Central America (Steyn et al. 2016;Magaña et al. 1999; Waylen et al. 1998; Waylen and Quesada 2002). Guanacaste also experienceshigh inter-annual variability in rainfall due to a complex interplay of local and remote climate processes.One of the main drivers is ENSO with significantly lower rainfall during El Niño years (Steyn et al. 2016).Geology and soilThe geology of the Nicoya peninsula is dominated by the Nicoya Complex that extends along thePacific coast (Denyer and Baumgartner 2006). The Nicoya Complex is obducted oceanic crust withexposures mostly limited to the Nicoya Peninsula (Sinton et al. 1997). It is a basaltic sequence olderthan Lower Campanian-Santonian (> 74 Ma) and mainly composed of olivine-tholeiite basalts (Denyerand Baumgartner 2006). The Nicoya Complex is the dominant bedrock within Potrero and Caimitalwatersheds, and dominates the geology of the hills surrounding the river valley (Figure 2.4). In the hills,the topography is generally characterized by slopes > 10 % (Figure 2.2). Loose and unconsolidatedsandy colluvial deposits dominate along the transition from hills to the relatively flat valley (Losilla andAgudelo 2003). Within the valley plain (slope < 10 %), Quaternary alluvial-colluvial sediments overlay alayer of fractured basaltic rocks and the low-permeability bedrock of the Nicoya Complex (Figure 2.4).13The alluvial-colluvial deposits consist of a top layer of 3 to 8 meters of silty clay and loam above a layerof sand to gravel with clay lenses (Losilla and Agudelo 2003; Denyer et al. 2013a, 2013b).Figure 2.4: Geology of the Potrero and Caimital watersheds. (a) Map of geological zones (Data sources:SENARA); and (b) Geological cross section developed based on SENARA (Agudelo (2006), Garcia-Serrano (2015), and Losilla and Agudelo (2003)). Topographical cross section based on HydroSHEDSDEM. MT-344, MT-343, MT-238 and MT-351 are wells with lithology logs (obtained from SENARA).14GroundwaterThe Potrero-Caimital aquifer extends through the alluvial-colluvial valley zone, and its lower boundary isthe contact with the underlying low-permeability rocks of the Nicoya Complex (Figure 2.4) (Agudelo2006). The aquifer is unconfined to semi-confined (Agudelo 2006). Groundwater recharge to thePotrero-Caimital aquifer consists of both diuse (direct) recharge from infiltration and percolation throughthe soil layer, and focused (indirect) recharge from downwardmovement from surface water to the aquifer(Agudelo 2006).Land useDominant land use in the two watersheds includes forest (52%), pasture (38%), agriculture (8%), as wellas residential (2%) based on a 2010 land use classification by Garcia-Serrano (2015) (Table 2.1; Figure2.5). Most agricultural fields are double-cropped with rainfed upland rice (non-paddy rice) during thewet season, and groundwater-irrigated melons during the dry season, although some fields are plantedwith rice during the wet season and lie fallow in the dry season.Figure 2.5: Land use/land cover of the Potrero and Caimital watersheds. Land use/land coverclassification based on Digital Globe ESRI from 2010 (Garcia-Serrano 2015).15Table 2.1: Land use/land cover (in hectares) of the Potrero and Caimital watersheds.Forest Agriculture (rice) Agriculture (melon/rice) Pasture Residential TotalArea [ha] 3,950 185 405 2,887 202 7,629Water demand and governanceThe Potrero and Caimital watersheds provide domestic water supply for many communities in the sur-rounding region. Domestic water demand includes small rural villages located within the watersheds(total population of 3,823 in 2014, data obtained from Ministerio de Salud in Nicoya). Furthermore,water is transferred via pipelines to the nearby towns of Nicoya (24,833 inhabitants, District of Nicoya;National Census and Statistics Institute in Costa Rica (Instituto Nacional de Estadísticas, INEC 2011),and Hojancha (4,245 inhabitants, District of Hojancha; (INEC 2011)) (Figure 2.6). Groundwater usedominates, although the town of Nicoya has a water treatment plant for a surface water intake from thePotrero River, supplying 27% of the total demand of Nicoya in 2016.Figure 2.6: Water demand in the Potrero and Caimital watersheds. Shown are villages and townsdepending on water supply of the Potrero and Caimital watersheds and aquifer, as well as groundwaterextraction wells and a surface water extraction site.16Local oces of the national water utility agency (AyA, Acueductos y Alcantarillados) deliver water supplyto households in Nicoya and Hojancha via transfer pipelines. In the villages, water supply is admin-istrated by community-elected and often volunteer-based boards called ASADAs (Asociación Admin-istradora de los Sistemas de Acueductos y Alcantarillados Comunal) that operate in rural areas ashighly decentralized units of the AyA (Kuzdas et al. 2015b). Most ASADAs operate a groundwater well,or in some cases have access to a surface water spring and storage tank. The operating ecienciesof these water boards depend on personal and financial capacities of the specific ASADAs (Kuzdaset al. 2015b). Records on water extraction rates by ASADAs are limited and total (monthly or yearly)water use is not reported to higher-level agencies. Many households also operate their own shallowpumping well for outdoor water use (which is typically called “artisanal well” in the region). Water licensesfor agricultural, domestic and industrial water extraction rates are obtained through the Dirección deAgua (Water Department) of the Ministry of Environment and Energy Ministerio de Ambiente y Energia,MINAE).2.2 Community-based monitoringThis research has been situated within the local communities from the beginning. It was part of theFuturAgua research project, an international and inter-disciplinary project focused on drought issues inGuanacaste frommultiple perspectives and approaches (McDaniels et al. 2013, 2014; Romero Valpredaet al. 2017). The FuturAgua project was supported by an involved group of local stakeholders andrepresentatives from dierent water organizations in Nicoya, Guanacaste, who recognized the impact ofdrought on their communities, and who were motivated to support drought-relevant research. The groupof stakeholder organizations included the National System of Conservation Areas (Sistemas Nacionalde Áreas de Conservación, SINAC) - Tempisque Conservation Area (Área Conservación Tempisque,ACT) from the Ministry of Environment and Energy (Ministerio del Ambiente y Energía, MINAE), the non-governmental organization NicoyAgua, the Environmental Management section (Dirección de GestiónAmbiental) of the town of Nicoya (Municipalidad de Nicoya), the National University of Costa Rica(Universidad Nacional Costa Rica, UNA), and the Tropical Agricultural Research and Higher EducationCenter (Centro Agronómico Tropical de Investigación y Enseñanza, CATIE) of Costa Rica. Stakeholderswere engaged from the proposal development stage. Thus, when the FuturAgua project started andI joined the research team, some initial stakeholder connections were already established. Theseconnections then proved essential for the development of my research project.During the initial research phase, it was of primary importance to build trust and a personal connectionto stakeholders and to learn from their local knowledge. The Potrero and Caimital watersheds emergedfrom discussions with local stakeholders as the focal watersheds for this research given their importance17for water supply, the possibility for future water conflicts, and the availability of prior studies and datasets(Agudelo 2006; Losilla and Agudelo 2003; Fernandez-Sing 2009; Garcia-Serrano 2015; Kuzdas et al.2015b; Morataya Montenegro 2004).Local stakeholders also provided essential support for finding locations for the hydrological monitoringstations. As many rural villages are spread throughout the watersheds, social considerations had tobe combined with hydrological ones (Table 2.2). I defined the hydrological goals of locating streammonitoring sites at watershed outflows to capture total watershed streamflow, and within the upstreamreaches of the streams to capture headwater flows. Further, as I was working with low budget stationsand would not be building a weir or other infrastructure, locations had to be found where the sensor tubecould be solidly installed along the river bank. I was also interested in monitoring groundwater levels.As there was no budget for drilling an observation well, I aimed to utilize artisanal wells, which manyhouseholds have on their property, for groundwater level monitoring. For this, I had to find wells in goodcondition and in locations spread throughout the aquifer extent. I was interested in monitoring wells thatwere not currently being pumped for observing changes in groundwater levels, as well as a groundwaterwell that is currently being pumped to estimate household water use.Table 2.2: Hydrological and social considerations for selecting monitoring site locations.Hydrological considerations Social considerationsStation location at watershed outflow Access and safetyStation location within upstream river reaches Minimize damage or theftSecure install along river bank Landowner interested to cooperate with researchersGroundwater wells in dierent locationsNext, I conducted exploratory surveys around the watersheds with local stakeholders and representa-tives from ACT and NicoyAgua in order to identify locations that were hydrologically suitable, and further,that also met social considerations such as access and safety, and minimizing potential damage to ortheft of the station. Another important aspect was finding landowners who were interested in cooperatingby permitting a station to be installed on their land. For this, initial contact through local stakeholderswas essential. It also helped that local landowners were aware of droughts and water shortages, andhappy to support a project seeking to conduct research on these issues. In most cases, the monitoringstation locations were in close proximity to houses of the landowners, which ensured the safety of thestations.During the following monitoring phase, the FuturAgua team and I continued to build relationships with lo-cal stakeholders and landowners through workshops organized under the FuturAgua project, meetings,and time spent in the field. Many of the landowners also developed stewardship strategies for monitoringstations, notifying researchers if there were issues at the site. I trained a local field assistant whoconducted station maintenance and data downloads when I was not in the field. It was also important18to keep stakeholders informed about the research progress during this time and share research resultsthrough workshops.Throughout the course of the project, I also conducted many meetings with water ocials in Guanacasteand in Costa Rica’s capital San José. Specifically, I met with water ocials from the AyA in Nicoya,Hojancha and San José, ASADA presidents, the Municipalidad de Nicoya, the groundwater agencySENARA in San José, the local governmental conservation authority ACT in Nicoya, and the non-governmental organization NicoyAgua. These meetings allowed me to better understand the local watergovernance structure as well as current and emerging water issues.2.3 Open-source data logger developmentIn Guanacaste, and in many other regions around the world, hydrologic monitoring remains sparse. Thelack of data makes it challenging to inform water resource management decisions and develop climatechange adaptation strategies (Montanari et al. 2013; Wohl et al. 2012). The goals of the current Inter-national Hydrologic Scientific Decade (2013-2022; Montanari et al. 2013) highlight this challenge withcalls for the development of new and innovative monitoring techniques to improve hydrologic monitoringplatforms, networks and databases which are necessary for well-informed water resource managementdecisions.Continuous data logging systems are essential for hydrologic research andmanagement of environmen-tal and agricultural systems. These systems are needed to obtain data at the high temporal resolutionnecessary to capture the episodic nature of dynamical hydrologic systems. One of the challenges forextending continuous hydrologic monitoring networks is the high cost of conventional data loggers anddata acquisition systems. However, recent advances in open-source software and hardware technolo-gies show potential for the development of low-cost logging systems that can be deployed with highspatio-temporal coverage. The Arduino project (https://www.arduino.cc) is based on an open-sourceelectronic microcontroller board with a processor, clock, USB connection and digital and analog inputports, and it is programmed using open-source software. The project embraces the philosophy of open-source projects (with open access to source code and hardware schematics), and is driven by a largeinternational user community contributing to ongoing improvement, development, and application ofthe technology (Hut 2013). Ongoing development of the Arduino as a low-cost environmental loggerincludes the “ALog” (Wickert 2014) and the “Mayfly Logger” (Stroud Water Research Center; Hicks et al.2015). Arduino-based data loggers show great potential for extending hydrologic monitoring networks,but are still a nascent technology. While some studies have explored the potential for using Arduinologgers for hydrologic (e.g., Islam et al. 2014; Hicks et al. 2015) and other environmental monitoring(e.g., Bitella et al. 2014; Lopez and Villaruz 2015), field applications spanning extreme environmental19conditions such as high humidity and temperature or longer time periods are rare. Below, I presentadvances in the development of a low-cost Arduino-based data logger towards establishing a hydrologicmonitoring network in a data-scarce region of the tropics, and share the challenges faced and successesobtained with this new technology over a monitoring period of over two years.Ecohydro LoggerArduino systems typically consist of the main Arduino board with processor and additional hardware(‘shields’) that can be attached to the main board to extend capabilities and add extra features. Thefirst version of the Arduino-based data logger used in this research was developed using an ArduinoMega 2560 board combined with a standard Arduino Secure Digital (SD) card shield. The Arduinowas programmed in the open-source Arduino Software IDE (C/C++) via USB cable to query read-ings from the sensor every ten minutes and record the data on an SD card. A key component ofthe program, the Serial Digital Interface (SDI) library, was developed by the Stroud Water ResourcesCenter (https://Github.com/StroudCenter/Arduino-SDI-12/). I deployed Arduino boards with water depth,temperature and electrical conductivity sensors (model CTD, by METER Group Inc. (formerly DecagonDevices Inc.), Pullman Washington, USA) during the first year of monitoring (March 2014 – April 2015).Subsequently, I designed, in collaboration with a micro-electronic engineer, a custom Arduino-baseddata acquisition system which I installed in the field in April 2015 (Figure 2.7). This “Ecohydro Logger”includes a customized shield combined with an Arduino Mega 2560 board. To reduce power consump-tion of the Arduino Mega board, the USB serial power was disabled with a shunt and all LEDs wereremoved. The customized shield includes a hard-mounted SD card reader, fuses for the solar chargecontroller and battery, as well as screw terminals for more stable cable connections. We further addedexternal clock hardware to the shield with a wake function to permit a nearly complete power downof the system between measurements. The SDI-12 (https:// Github.com/ EnviroDIY/ Arduino-SDI-12),Real Time Clock (https:// Github.com/ cjbearman/ ds1306-arduino) and LowPower (https:// github.com/rocketscream/ Low-Power) Arduino libraries were used. Code for programming the Arduino EcohydroLogger and for data processing (using R programming language), as well as hardware details includ-ing the electronic circuit design for the Ecohydro Logger are available on GitHub (https:// github.com/UBCecohydro/ Ecohydro.Arduino).20Figure 2.7: Monitoring setup at the Potrero river. The open-source Arduino-based “Ecohydro Logger” isconnected to a CTD sensor (electrical conductivity, temperature, water depth). The system is remotelylocated and powered by solar panels.The cost of the Ecohydro Logger, including the Arduino Mega board and the Ecohydro shield, is ap-proximately US$100, in contrast to Campbell Scientific’s lowest cost data logger (the CR200X) currentlypriced from US$480. The Ecohydro Logger can be connected to any SDI or analog sensor. Consideringthe open-source hardware approach, the data logger is highly customizable to researchers’ needs. Forexample, the platform can be extended to include automated data uploading to an Internet-connecteddatabase by adding a Global Systems for Mobile communications (GSM) shield to the platform.A river stage time series obtained by automated data logging at ten-minute intervals is presented inFigure 2.8, shown here as raw data with gaps to draw attention to the capabilities and challengesexperienced with the Arduino setup during development of the data acquisition system. There wereseveral reasons for discontinuities in this time series, including those typical for installations in remotelocations (an interrupted solar panel cable connection due to a tree branch falling onto the wiring, andan interrupted sensor cable connection after a storm), and those related to the ongoing developmentof the Arduino-based data acquisition system (power problems in particular during the wet season withincreased cloud coverage and growing vegetation, loose cable connections, and code errors related tochange in date).21Figure 2.8: Time series of stream stage (raw data, 10-minute intervals) collected using open-sourceArduino-based data loggers. References to causes of data interruptions during development of thesystem (43% of data missing): Interrupt 1 - damage to solar panel cable connection; 2 Arduino DataLogger Version 1 power issues; 3 - interrupted sensor cable connection after storm; 4 - code error relatedto change in date (corrected).The new Ecohydro Logger overcame the shortcomings experienced with the first version, and dataloss was minimized following installation of the new system. The reduced power consumption, with adecreased electrical current demand from 80 mA (average) to 16 mA (average, considering both “sleep”and “wake” cycles), and the addition of screw terminals for cable connections significantly increasedrobustness of the platform when deployed in the field. Despite the extreme conditions of the wet-dry tropics (with high air temperature and moisture content in the wet season), the Ecohydro Loggerperformed well, even when located within the tropical rainforest, and the installation with a double-boxsystem (Figure 2.7) proved to be reliable. The Ecohydro Logger was able to capture long periods ofhigh-frequency stream data, recording the flashy nature of the streamflow response to rainfall previouslyunavailable in this region. The modified version of the system also allowed for development of stage-discharge rating curves using the salt gauging method (Moore 2005), providing a proof-of-concept forthe extensibility of the base platform.The use of an open-source and inexpensive Arduino-based data acquisition system can provide a pow-erful means for extending hydrologic monitoring networks into regions where high-frequency monitoringis limited. A growing hydrologic database and understanding may be able to contribute to improvedwater management and climate change adaptation strategies. The Ecohydro Logger developed in this22research is significantly less expensive than conventional commercial data loggers. Furthermore, the ac-cessible, extensible and open-source nature of the Arduino-based data acquisition system shows greatpotential for integration into socio-hydrology that can empower local citizens to contribute to increasingknowledge on water resources in their communities.2.4 Hydrological monitoring network2.4.1 Stream monitoringI installed a stream monitoring network in the Potrero and Caimital watersheds in December 2013 (1station) and March/April 2014 (4 stations) and equipped these with Arduino data loggers. To increasespatial resolution in stream stage monitoring, I chose a nested approach, with monitoring locations nearthe outflows of the Potrero and Caimital watersheds, upstream at both rivers, and at a small ephemeraltributary stream (Table 2.3; Figure 2.9; Figure 2.10).Figure 2.9: Monitoring site locations. SW = surface water (stream), GW = groundwater, EC = EddyCovariance station operated by FuturAgua; WL = Weatherlink meteorological station operated by farmowners.23Table 2.3: Location, elevation in meters above sea level, and upslope area in hectares of streammonitoring sites. Upslope area delineation based on 90 m DEM (HydroSHEDS).ID Location Station name (river) Elevation Upslope area[m..s.l.] [h]SW1 Casitas Potrero River Downstream 104 33,850SW2 Varillal Tributary to Potrero River 155 7,934SW3 Costeña Potrero River Upstream 160 6,364SW4 Costeña Caimital River Upstream 153 3,114SW5 La Virginia Caimital River Downstream 138 40,892Figure 2.10: Upslope area for stream monitoring stations (based on 90 m HydroSHEDS DEM).At each monitoring station, a water depth, temperature and electrical conductivity sensor (model CTD,by METER Group Inc. (formerly Decagon Devices Inc.), Pullman Washington, USA) measured theseparameters at ten-minute intervals via the Arduino data logger. I tested the factory calibrations of theCTD sensors in the laboratory prior to field installation, and again in fall 2015 using standard proceduresrecommended by Decagon Devices (Decagon Devices, 2016). As the CTD sensors output a digitalsignal, there is no user calibration, as is typical with analog sensors. In addition, my field assistantand I made monthly physical measurements of water height and discrete measurements of electricalconductivity and water temperature using independent handheld conductivity and temperature sensors24(model GS3, by METER Group Inc.). These monthly measurements were used to assess data qualityobtained from the data logger-connected sensors.At all stations, the sensor was installed at the base of the main thalweg of streams and protected withina perforated PVC tube. The tube and sensor were secured against roots and trees along the streambank for stability during stormflows (see Appendix A for photos of installations at all stations and crosssections of the stream at the monitoring stations). Each system was powered by two solar panels (with amaximum voltage of 21.9) that were connected to a solar charge controller, providing power to recharge12V/7Ah gel-cell batteries. Data were downloaded once per month. At each of the five streammonitoringstations, water depth was recorded every ten minutes via the CTD sensor. However, due to continuedriver bank erosion problems at the Tributary to Potrero River site (SW2), data for this station had to bediscarded and was not further processed (Appendix A).Stage data were gap-filled, cleaned and aggregated to 30-minute averages. To develop location-specificrelationships between measurements of stream stage and stream discharge, I conducted a series ofdiscrete discharge measurements at each monitoring site with the salt solution slug injection method(Moore 2005). For details on discharge measurements and rating curve development, see Appendix A.2.4.2 Groundwater level monitoringI also installed three groundwater level monitoring stations in December 2013 (1 site) and March/April2014 (2 sites) (Figure 2.9, Table 2.4). Installation design at the groundwater monitoring sites was similarto the stream monitoring sites. That is, I used the Ecohydro Logger, solar panels as power supply, and aCTD sensor to measure groundwater level. It was not possible to drill new observation wells due to costsand logistics; therefore, monitoring was carried out at existing artisanal (household) groundwater wells.Two of these groundwater wells were used infrequently for pumping, while the third one was pumpeddaily. The two unused groundwater wells provided information on natural groundwater levels, while theother well was used to obtain information on daily water use from an otherwise unmetered source.The CTD sensor was lowered into the groundwater well to just above the bottom of the well. The CTDsensor measures the depth of the water column above the sensor via a vented dierential pressuretransducer. This depth measurement was combined with sensor cable length in the well and elevationabove sea level at the ground surface of the station to calculate the groundwater elevation. Duringmonthly site visits for data downloads, groundwater levels were also measured using a water level dipmeter to ensure consistency with the depth measurement of the CTD sensor. More details on installationand lithology at well monitoring sites are provided in Appendix A.25Monthly measurements of groundwater level for wells throughout the Potrero-Caimital aquifer are alsoconducted by the Comité Técnico Interinstitucional Estadística del Agua (CTI), which is a collaborationbetween the SENARA, AyA and MINAE. I obtained these groundwater level data from 2005 to 2015.Table 2.4: Elevation at ground surface of groundwater monitoring sites and at well bottom in metersabove sea level. Details on well screen elevations were not available, but according to well owners, thewell screens are just above the well bottom.ID Location Elevation at ground surface Elevation at well bottom[m..s.l.] [m..s.l.]GW1 Varillal 155 150GW2 Dulce Nombre 169 158.5GW3 Gamalotal 149 1342.4.3 Meteorological monitoringIn July 2014, I helped to install a meteorological and eddy covariance (EC) equipped monitoring stationwithin the fields of a melon-rice farm (EC Station, Figure 2.9). The station is located at the watershed di-vide between the Potrero and Caimital watersheds within the alluvial valley. The EC system is composedof an ultrasonic anemometer (Young 81000 model) and an open path infrared gas analyzer (LI-COR7500A) for measurement of actual evapotranspiration. The monitoring station also included a multi-parameter meteorological instrument (Vaisala WXT520) for rainfall, air temperature, relative humidity,wind speed and direction, and air pressure measurements, as well as additional soil sensors (includingsoil moisture and temperature) and radiation sensors (including shortwave solar radiation). Data wererecorded at this station in 30-minute intervals since July 2014 within the purview of the FuturAguaproject. An additional weather station (Davis Weatherlink Vantage Pro2 Plus) located 600m from theEC station is operated by the farm owners. The Weatherlink station has recorded hourly values ofrainfall, air temperature, relative humidity, wind speed and direction, air pressure since 2007, and short-wave solar radiation since 2014. Data obtained from the Weatherlink station were used to gap-fill themicrometeorological records from the EC station during occasional data outages.2.5 Water use data assemblyIn my research, I focused on population and water use data to explore socio-hydrological patterns in thewatersheds. Population data were available through census data (INEC 2001, 2011); however, the datawere not disaggregated to the level of towns. With support from local stakeholders, I obtained populationdata at the level of the rural towns of the Potrero and Caimital watersheds from the Ministerio de Saludin 2014.26Through several meetings with national water agency (AyA) representatives in Nicoya and in Hojancha, Iobtained monthly water extraction volumes for these two towns. The rural villages within the Potrero andCaimital watersheds are governed by rural water boards (ASADAs). I met with several ASADA presidents(from Curime, La Virginia, Dulce Nombre, Samara and Caimital) to learn about their ASADAs and wateruse. However, while household water use is metered at each house, only the ASADA of Caimital hadkept records of water bills, and my field assistant and I digitized and anonymized these hand-writtenmonthly water bills for each household from 2012 to 2015.This assembly of the water use data - and learning from local knowledge on the water challengesof the region – was only possible through the community-based approach of this research, wherestakeholder partners facilitated initial contacts, allowing me to build relations with local water managers.Similarly, the community-based approach was essential for the hydrological monitoring, where manyof the landowners of the monitoring stations developed station stewardship, and where stakeholdercooperation had been essential for identifying monitoring sites.27Chapter 3Hydrological model3.1 Hydrological modellingField measurements form the basis of hydrological analysis. However, field techniques have limitationsin both space and time, and we cannot measure every process in a hydrological system (Beven 2012).Thus, if we want to synthesize current watershed dynamics, and further, explore a future for which wecannot have any measurements, we need the help of a tool (Beven 2012). A hydrological model canprovide such a tool for the simulation of watershed processes (Beven 2012; Bronstert et al. 2005).In any context, a model is a simplification of a complex real world where one focuses on capturingkey elements and processes that are important for the task at hand (Beven 2012; Bronstert et al.2005). A model is therefore helpful for formalizing knowledge and synthesizing the components of acomplex system. More specifically, a hydrological model describes a range of hydrological processes ina watershed (or any other hydrological unit), for which the research questions determine the focus andthe spatial and temporal scale of the modelled processes (Beven 2012; Bronstert et al. 2005). In mostcases, the hydrological model is developed within a computer environment, although physical scalemodels also exist.For my research, I used a hydrological model to support three goals:• First, it will provide a means for synthesizing current water flows in the Potrero and Caimitalwatersheds, and support me in assessing the complex socio-hydrological dynamics in these wa-tersheds.• Second, it will allow me to explore the potential impacts of future climate change and water usechange scenarios on water supplies in these watersheds.• And third, I can use the model results to develop an adaptation tool.Hydrological modelling consists of several distinctive steps (Beven 2012): First, the modeller decideswhich processes to include in the hydrological model, which is based on the research questions andthe perception of the modeller of the watershed processes. These processes and relations between28fluxes and storage terms are then described in the conceptual model. In the following procedural model,the modeller implements the processes and watershed characteristics in a computer model. This isfollowed by calibration during which model parameters are optimized so that model results are similarto field measurements, and finally, evaluation, when model performance is assessed. The followingsub-sections describe these modelling steps.3.2 Conceptual model overviewThe conceptual model describes the dominant processes relevant to the research question. For myresearch in the Potrero and Caimital watersheds, I am specifically interested in generation and variabilityof water supplies throughout dierent seasons and years (water budget), i.e., specifically the processesof streamflow (rainfall-runo generation) and groundwater recharge. Equally though, I am interested insocial water dynamics, such as surface water and groundwater extraction and agricultural irrigation.The conceptual model in Figure 3.1 provides an overview of these dominant processes, while noting thatsection 3.4 will dive into the detail of conceptualizations that are applied within the hydrological model.Figure 3.1 shows both water fluxes and storage terms within the system. Rainfall provides inflow to thehydrological system, which can either infiltrate into the soil, or contribute via surface runo to streams.Infiltrated water is stored in the unsaturated zone as soil moisture, fromwhich evapotranspirative demandis satisfied (and exits the system), and from where it can contribute to streamflow via interflow, or todiuse groundwater recharge via percolation to the groundwater table. Saturated zone (groundwater)storage can contribute to stream storage via baseflow, and streams can contribute to groundwater stor-age via focused groundwater recharge through the streambed (surface water - groundwater interaction).Total watershed streamflow is also an outflow from the (watershed) system. Society extracts water fromboth surface water (streams) and groundwater storage, some of which returns to the system via irrigationreturn flow.29Figure 3.1: (Overview) conceptual model of dominant processes showing storages and exchanges(systems model). Arrows indicate water fluxes, boxes indicate water storage. Adapted and extendedfrom Freeze and Cherry (1979).I also developed a landscape-based conceptual model specific to the characteristics of the Potreroand Caimital watersheds (Figure 3.2), based on previous studies and geological information from thewatersheds (Agudelo 2006; Garcia and Agudelo 2014; Losilla and Agudelo 2003) and study site char-acteristics described in Chapter 2. Evapotranspiration rates are determined by meteorology, as wellas location-specific land cover and available soil moisture. The three geological zones include thebedrock of the Nicoya Complex (basalt zone) within the hills, unconsolidated colluvial sediments in thetransition from hills to valley (colluvial zone), and unconsolidated alluvial sediments within the alluvialvalley (alluvial zone). The bedrock consists of a layer of fractured (more permeable) basalt aboveless fractured (lower permeability) basalt. The basalt extends under the alluvial-colluvial deposits inthe alluvial valley, and is assumed to be the lower boundary of the Potrero-Caimital aquifer. Streamtributaries from the hills feed into the main stem of the Potrero and Caimital rivers in the central alluvialvalley. Runo and interflow from the basalt zone contribute to tributary streamflow. Stream reaches inthe colluvial and alluvial zone are in contact with groundwater (surface water – groundwater interaction).Human water extraction from both surface water and groundwater are also included. More detailedconceptualizations of modelling processes, such as rainfall-runo generation, are discussed in section3.4.30Figure 3.2: Landscape-based conceptual model of the Potrero and Caimital watersheds. Here, aconceptual cross section from one hill side to the main stem of the Potrero / Caimital River is shown.SW - gw interaction = Surface water - groundwater interaction.3.3 Software selectionMany dierent software applications for hydrological modelling are available, which dier widely in focus,spatial extent, and modelled processes. Based on my research questions and conceptual model of thePotrero and Caimital watersheds, selection criteria emerged to guide the decision of which modellingsoftware to use (Table 3.1). First of all, I needed to choose a watershed model (in contrast to a hillslope or global model) to capture the water budget of the Potrero and Caimital watersheds. The modelshould allow accounting for spatial variability in geology, land use, slope and other characteristics.Further, the model should allow simulation of evapotranspiration and rainfall-runo generation (i.e.,soil infiltration, surface runo and interflow to streams), but also groundwater recharge and interactionbetween surface water and groundwater. Furthermore, societal aspects such as agricultural irrigationand water management aspects should be included. In the Potrero and Caimital watersheds, data onhydrology, soils and other characteristics are sparse, further necessitating a hydrological model that31can be used to simulate watersheds with limited input data. Another consideration was the cost of thesoftware. The costs should be within the research budget, but also, considering the community-basedaspect of this research, allow potential future implementation in Costa Rica for water management oruniversity training where high commercial software costs could be prohibitive.Table 3.1: Criteria for hydrological model selection.Type CriteriaSpatial extent Watershed modelSpatial scale Spatial variability in geology, land use, slope, ...Temporal scale Daily time steps to capture streamflowScenarios Climate and water use scenarios can be appliedModelled processes - Evapotranspiration- Rainfall-runo generation- Streamflow- Groundwater recharge- Surface water - groundwater interaction- Agricultural irrigation- Societal aspects (surface water & groundwater extraction, demandmanagement)Input data requirements Usable for watersheds with limited data availabilitySoftware costs Ideally non-commercial (for use in developing countries such as Costa Rica)Below, I discuss briefly a number of relevant hydrological modelling software, and their advantagesor disadvantages in light of the selection criteria (summarized in Table 3.2). Further details on specificmodels can be obtained through the indicated references. As all models are simplified representations ofreality, watershed processes are portrayed in dierent manners. The main dierences between modelsare determined through their spatial distribution and descriptions of processes. Fully-distributed andprocess-based models describe watershed characteristics at the level of grid cells, based typically onnonlinear dierential equations that are solved numerically. These models require many parameters,which have to be determined for each grid cell, and their data input needs are therefore high (Beven2012). Semi-distributed models are simpler and are based on the distribution of responses within awatershed, for instance using the concept that many points within a watershed behave in a hydrolog-ically similar way in relation to rainfall-runo processes, and can therefore be classified according totheir functional hydrological response (Hydrological Response Units; HRU) (Beven 2012). Each HRUrepresents areas of similar characteristics (such as land use, soil type, or slope) for which it can beassumed that the hydrologic response to rainfall in regards to runo generation, groundwater rechargeor evapotranspiration is similar.32The MODFLOW (Modular Groundwater Flow) model is a hydrogeological grid-based numerical modeldeveloped by the United States Geological Survey (USGS) (Harbaugh 2005). The code is open-source,but most of the graphical user interfaces (GUI) of MODFLOW are commercially-distributed software andcostly, however some free GUIs also exist. MODFLOW is powerful for groundwater flow modelling, but itdoes not allow for modelling of rainfall-runo and groundwater recharge processes. It can, however, becoupled with the USGS Precipitation-RunoModelling System (PRMS), which is a semi-distributed HRUbased rainfall-runo model, and the coupled model is called the GSFLOW (Groundwater and SurfaceWater Flow) model (Markstrom et al. 2008). As it is based on MODFLOW, GSFLOW also has high datainput requirements especially for the hydrogeology component, and further, does not explicitly allowintegration of societal aspects.The FREEWAT (FREE and open-source software tools for WATer resource management) was devel-oped by a consortium of universities and institutions in Europe and is funded by the European Union(Foglia et al. 2018; Rossetto et al. 2018). It shows great potential for research such as this study, asit is targeted towards working with stakeholders, facilitating data and model sharing across platforms,and is free and open-source (Foglia et al. 2018). It is operated through the open-source GIS softwareQGIS which facilitates pre- and post-processing of model input and output. Its hydrological modellingcapacities are mainly based onMODFLOW, and thus, at least currently, focus on groundwater modelling,even though further extensions are planned for the future. Being a newly developed model, only a testversion of FREEWAT was available at the beginning of the hydrological modelling for this research (in2018 a full version was released).MIKE SHE is a distributed grid-based model that is an extension of the Système Hydrologique Européen(SHE) model, developed by the Danish Hydrologic Institute (DHI 2012). It allows integrated watershedmodelling, including modelling of rainfall-runo processes, flow routing in streams, groundwater flow,and coupled modelling of surface water and groundwater interactions. However, it has high data in-put requirements, and as it is a commercially-distributed software, it is costly. It also has no explicitintegration of societal water management aspects.The Soil andWater Assessment Tool (SWAT) was developed by the Agricultural Research Service at theUnited States Department of Agriculture (USDA) and the AgriLife Research at The Texas A&MUniversity(USDA-ARS and Texas A&M 2017). Historically this model focused on modelling large river basins, butsmaller watersheds can also be modelled (Arnold et al. 1998; USDA-ARS and Texas A&M 2017). It isspatially semi-distributed using the HRU approach, and allowsmodelling of the hydrological processes ofinterest in this research. It has detailed soil and vegetation routines, as its primary use is for agriculturalwater management applications, but it has limited integrations of other societal system components.It is free of charge and open-source, but its most-used graphical interface depends on the commercial33and expensive ArcGIS software (however, recently SWAT also interfaces with open-source GIS softwaresuch as QGIS).The Water Evaluation and Planning Tool (WEAP) was developed by the non-profit Stockholm Environ-ment Institute (Yates et al. 2005). Small license fees are charged in developed countries, but users indeveloping countries such as Costa Rica have free access to the software. WEAP is a semi-distributedmodel using the HRU approach. The model includes evapotranspiration, rainfall-runo generation,groundwater recharge and storage, as well as surface water – groundwater interaction. Furthermore,the model allows one to combine the hydrologic system components with social system components.For instance, it includes water demand modelling routines, where water users (such as dierent towns)can be described by specific characteristics, such as population, per capita water use, water supplysources that a town draws on (which is then connected to the hydrological model components), waterallocation schemes, and also agricultural irrigation practices.Table 3.2: Advantages and disadvantages of potential modelling software for use in this research.MODFLOW (Harbaugh 2005); GSFLOW (Markstrom et al. 2008); FREEWAT (Foglia et al. 2018;Rossetto et al. 2018); MIKE SHE (DHI 2012); SWAT (Arnold et al. 1998); WEAP (Yates et al. 2005).Software Advantages for use in this research Disadvantages for use in this researchMODFLOW Spatially-distributed grid-based numericalgroundwater model.High input data requirements; no rainfall-runo andsurface water modelling component.GSFLOW Spatially-distributed grid-based numericalgroundwater model; Coupled surface water –groundwater model.High input data requirements; no explicit integration ofsocietal dynamics.FREEWAT Free and open-source software directed forstakeholder interaction; Spatially-distributedgrid-based numerical groundwater model.Currently no rainfall-runo and surface watermodelling component; software only as test versionavailable.MIKE SHE Spatially-distributed grid-based numerical watershedmodel with many modules, including rainfall-runo,groundwater flow and surface water –groundwaterinteraction.High input data requirements; no explicit integration ofsocietal dynamics; high costs for software.SWAT Semi-distributed hydrological model with less inputdata requirements than spatially-distributed models.Rainfall-runo modelling focus. Free of charge.Simplified representation of hydrological processesthrough HRU approach; focus on soil and vegetationdynamics with detailed needs for soil and vegetationinput data; No groundwater flow modelling; no explicitinclusion of other societal water management aspects.WEAP Semi-distributed hydrological model with less inputdata requirements, also usable in data-scarce regions.Rainfall-runo, surface water routing, groundwaterrecharge/ storage, and interactions between surfacewater and groundwater. Explicit integration of societalaspects (water demand modelling, water allocation,water transfers).Simplified representation of hydrological processesthrough HRU approach. Small costs for use indeveloped country. No groundwater flow modelling.34All hydrological models have advantages and disadvantages for use in this research (Table 3.2), andtrade-os between the dierent hydrological modelling approaches were necessary. One of the majorconstraints for research in the Potrero and Caimital watersheds was the limited input data availability,which would make the high input data requirements of the fully-distributed models (MODFLOW, GS-FLOW, MIKE SHE and FREEWAT) challenging. The stand-alone version of MODFLOW (and thus,FREEWAT) does not include rainfall-runomodelling, and surface water – groundwater interaction. TheSWAT and WEAP models are both semi-distributed based on the HRU approach, but while the focus ofSWAT is on agricultural water management and soil-vegetation processes specifically, the WEAP modelallows explicit integration of societal aspects through its demand site modelling and water managementroutines. Considering these aspects, I chose the WEAP model for my research.3.4 Process conceptualizationOne of the main advantages of the WEAP model is that it allows integration of hydrological processmodelling (generation of water supplies and water flows through a watershed) with a water managementmodule, in which water demand sites can be connected to various water supply sources, and watermanagement strategies can be included in the model (Yates et al. 2005). The hydrological or bio-physical component includes modules of rainfall-runo generation (rainfall, evapotranspiration, interflow,groundwater recharge generation), an alluvial groundwater model (including the possibility of surfacewater – groundwater interaction) and streamflow routing. Water demand can be specified at the level oftowns or disaggregated to per capita needs, and can be included for dierent sectors such as domestic,business or agricultural irrigation.Spatial disaggregation in WEAP is first based on division of the study watershed into sub-catchments(SCs), which are often similar to tributary catchments. For every time step in the model, a water balancewith outflows from and storage changes within each SC is calculated. The time step can be chosen asdaily, weekly or monthly. Outflows from a SC can be routed to rivers or aquifers. Each SC is fractionallydivided into HRUs which represent areas of similar characteristics (such as land use, soil type, or slope).Climate is assumed uniform over each SC. Rainfall provides the water supply to a catchment that thenbecomes gradually depleted through natural processes and human demands. First, evapotranspirativedemand is satisfied within the catchment, and then residual water contributes to streamflow, groundwaterrecharge and soil water storage, and is eventually available as an input to the human water demandsystem. A simplified flow structure of the dierent model routines for each time step is presented inFigure 3.3.35Figure 3.3: Simplified flow structure of model routines in WEAP for each time step. GW - SW =groundwater - surface water interaction.Evapotranspiration (ET) is estimated in WEAP by the crop coecient method (Doorenbos and Pruitt1977). Evapotranspiration from each HRU within a SC is estimated as the product of a “crop” coecient(Kc), that incorporates land cover characteristics and water requirements, and a reference crop potentialevapotranspiration (ETreƒ ) that represents atmospheric water demand based on meteorological condi-tions (see also Equation 3.2 below). The daily ETreƒ is calculated by WEAP using the reference croppotential evapotranspiration equation (Shuttleworth 1993 equation 4.2.31; Equation 3.1), which is basedon the Penman-Monteith equation but modified for a standardized crop of grass (i.e., several resistancesare here assigned values for a specific, well-defined reference surface):ETreƒ =+ (Rn  G) + + 900T+ 275U2 D (3.1)where ETreƒ⇥mmdy1⇤ is the reference crop potential evapotranspiration,  ⇥kP °C1⇤ is thetemperature gradient of saturated vapor pressure, ⇥kP °C1⇤ is the psychometric constant with = (1+ 0.33U2), T [°C] is the air temperature, 900 and 275 are specific values assigned for the modi-fied Penman-Monteith equation for a standardized crop according to Shuttleworth (1993), U2⇥ms1⇤ isthe wind speed at 2m, Rn⇥mmdy1⇤ is the net radiation exchange for the crop cover,G⇥mmdy1⇤is the soil heat flux, and D [kP] is the vapor pressure deficit. Rn and G are given here inmmdy1for the modified Penman-Monteith equation, according to Shuttleworth (1993).The reference crop evapotranspiration equation is calculated for each time step (here, daily). As input,WEAP uses the daily averages of air temperature, relative humidity, wind speed, as well as cloudinessfraction to estimate the net solar radiation. In theWEAPmodel, cloudiness fraction describes the numberof bright sunshine hours per day over the total hours of daylight.The rainfall-runo response in a SC can be modelled by dierent modules within the WEAP model. Iused the soil-moisture module that allows modelling surface runo processes by representing each SCwith two soil layers (Figure 3.4) (Sieber and Purkey 2015; Yates et al. 2005). Hydrological processes inthe top soil layer 1 are modelled separately for each HRU in a SC, while the lower soil layer 2 extendsthrough the entire SC and percolation from HRUs in soil layer 1 contributes to the combined soil layer 2(Figure 3.4a). If the SC is underlain by an alluvial aquifer, the lower soil layer 2 is replaced by the aquifer36layer (Figure 3.4b). If several SCs are connected to the same aquifer (such as SC C and SC D in Figure3.4b), percolation from HRUs in these SCs enters the same aquifer storage unit.Figure 3.4: Conceptualization of rainfall-runo processes in WEAP (soil-moisture module); Figureadapted and extended based on Sieber and Purkey (2015). (a) Sub-catchments, no underlying aquifer:Percolation from soil layer 1 in each HRU enters soil layer 2 (that is shared within one sub-catchment).(b) Sub-catchments are underlain by an aquifer, which is shared across all SCs that are connected tothis aquifer. Percolation from soil layer 1 enters aquifer.For each fractional area (HRU) in a SC, a water balance is computed at each time step (Equation 3.2).Equation 3.2 is solved for each HRU for each time step according to changes in input precipitation andpotential evapotranspiration. The storage change in soil layer 1 (top layer) for each time step in eachfractional area (HRU) in a sub-catchment is given by Equation 3.2 (Sieber and Purkey 2015);Rdjdz1,jdt= Pe (t) ETreƒ (t)Kc,j (t) 5z1,j 2z21,j3! Pe (t)zRRFj1,j  PFDj ks,jz21,j1 PFDjks,jz21,j(3.2)where the first term on the right hand side indicates the eective precipitation, followed by evapotranspi-ration, surface runo, interflow and percolation to the lower soil layer (or aquifer), and where Rdj [mm]is the total eective storage of soil layer 1 for each HRU fraction j, z1,j [] is the relative storagein soil layer 1 as fraction of total eective storage Rdj for each HRUj (soil moisture), t is the timestep (here, daily), Pe [mm] is the eective precipitation, ETreƒ [mm] is the reference crop potentialevapotranspiration, Kc,j [] is the crop coecient for each land cover type, RRFj [] is the runo37resistance factor of land cover, ks,j⇥mmdy1⇤ is the soil layer 1 saturated hydraulic conductivity, andPFDj [] is the preferred flow direction, a partitioning coecient that fractionally divides water flowshorizontally and vertically.Change in water storage in the upper soil layer is thus determined by antecedent soil moisture, inputthrough rainfall, and loss through evapotranspiration, surface runo, interflow and percolation to thelower soil layer 2 (or aquifer) (Equation 3.2). Surface runo and interflow are routed to streamflow, whilepercolation through the upper soil layer enters the lower soil layer, or is routed to an aquifer storage(Figure 3.4). Aquifer characteristics are not disaggregated and the aquifer is modelled as one storageunit to which recharge from many SCs can be routed, which can interact with surface water, and fromwhich demands sites can extract water.In WEAP, it is possible to dynamically model interactions between surface water and an alluvial aquifer.Stream reaches can be gaining water from the aquifer (gaining stream) or contributing focused rechargeto the aquifer (losing stream), depending on the level of groundwater in the aquifer (Sieber and Purkey2015). Surface water – groundwater interactions aremodelled through an idealized ’groundwater wedge’along the length of the river reach (Sieber and Purkey 2015). If aquifer levels rise higher than level ofequilibrium with the stream, seepage occurs from the aquifer to the stream, and vice versa, if streamwater levels rise, seepage occurs from stream to aquifer. Specifically, first, the ’equilibrium ground-water storage’ (GSe⇥m3⇤) for one side of the ’groundwater wedge’ is estimated assuming equilibriumconditions between stream and groundwater table,GSe = hd  Ad Sy (3.3)where hd[m] is the horizontal distance from farthest edge of aquifer to stream, l [m] is the wettedlength of the aquifer in contact with the stream, and Ad [m] is the aquifer depth at equilibrium and Sy[]is the specific yield of the aquifer (Sieber and Purkey 2015). The height yd [m] of current groundwaterstorage GS⇥m3⇤above or below equilibrium storage is calculated asyd =GS  GSehd  Sy(3.4)If the groundwater table rises relative to the stream level, seepage from aquifer to stream increases, andin contrast, the lower the groundwater table becomes relative to stream levels, the higher the flow fromstream to aquifer. The seepage S⇥m3 dy1⇤ from both sides of the stream is given asS = 2✓ks,qydhd◆ d (3.5)38where ks,q⇥mdy1⇤ is the saturated hydraulic conductivity of the aquifer and d [m] is the wetteddepth of the stream (Sieber and Purkey 2015). After estimation of seepage, WEAP calculates totalgroundwater storage for each time step based on groundwater storage of the previous time step, andthe dierence between diuse recharge entering the aquifer, anthropogenic extraction volumes, andseepage.Water demand sites such as towns can be added in WEAP, and are described by an annual activity level(this could be population for domestic demands, or production sites for business demands), an annualwater use rate, and a daily variation of demand to reflect dierent water use throughout the year. Foreach demand site, the water supply sources can be defined. A demand site can draw from several watersources (i.e., a stream and/or an aquifer), and water is transferred from a water source to a demand sitevia ‘transmission’ links (which would be equal to water pipelines or aqueducts). During transmission,water loss through leakage can occur. Water consumption at a demand site constitutes a watershedoutflow (i.e. water does not return), whereas for instance water extraction can also contribute to irrigationreturn flow. Irrigation is applied directly to the HRU that was defined as irrigated agriculture (seasonalextents of irrigation can be defined). The irrigation module within WEAP applies irrigation based on soilmoisture content. Modelling irrigation as part of the catchment routines allows the irrigation water tore-enter the catchment system, i.e., to model evapotranspiration and infiltration into the soil.3.5 Model setup and input parameters3.5.1 Model time steps and initial modelling periodFor the WEAP model of the Potrero and Caimital watershed, I chose a daily model time step. This wasmostly based on the field observation that streamflow response to rainfall events was flashy in thesetropical watersheds, which a monthly model would not be able to capture. A higher time resolution thandaily would, however, not be possible given limited available input data.I chose the period from 2005 to 2016 as my initial modelling period for model calibration and evaluation,as well as assessment of current system dynamics. This period included both high and low rainfall years,thus reflecting a range of hydrological conditions. Furthermore, I was able to acquire meteorologicaldata, groundwater levels and water extraction data for this period. Streamflow data were only availablefrom 2014 to 2016 from my monitoring, and this period constituted the principal model calibration period.393.5.2 Spatial distributionAs described in section 3.4, spatial distribution inWEAP consists of sub-catchments (SCs) and fractionalareas within the SCs of dierent land cover, soil and topography characteristics (HRUs). Below I describethe delimitation of the SCs and HRUs for the Potrero and Caimital watershed model.I used the following criteria to guide the delimitation of SCs:1. Upslope area of the monitoring sites.2. SCs of the tributaries to allow routing of tributary runo into main stems of the Potrero and Caimital.3. Aquifer extent to allow modelling diuse groundwater recharge.Criteria one and two were based on a digital elevation model (DEM). I used the HydroSHEDS (3-seconds) DEM that has an approximately 90 m spatial resolution in the region (Lehner et al. 2008).I hydrologically-conditioned the DEM with a detailed local stream layer (obtained from the ACT), anddelimited the SCs using the open-source programsWhitebox GAT (Whitebox Geospatial Analysis Tools)(Lindsay 2016), SAGA-GIS (System for Automated Geoscientific Analyses) and the SAGA plug-in inQGIS (Conrad et al. 2015; Olaya 2004; QGIS Development Team 2017). Criteria three was basedon extents of the Potrero-Caimital aquifer, as reported by SENARA (Losilla and Agudelo 2003), anddescribes SCs in the alluvial-colluvial valley (alluvial and colluvial zones) of the Potrero and Caimitalwatersheds. In contrast, SCs in the hills (basalt zone) are not underlain by the Potrero-Caimital aquifer.Using the three criteria, a total of 79 SCs were delimited for the Potrero and Caimital watersheds (Figure3.5).40Figure 3.5: Sub-catchments developed for the WEAP model of the Potrero and Caimital watersheds,and modelled stream reaches.Next, each SC was divided into HRUs or fractional areas that were assumed to have a similar hydrologi-cal response to rainfall events. The first criterion for the HRU definition was land cover, which was basedon a land use map delineated from satellite imagery from Digital Globe ESRI (Date: December 28, 2010)with a 5m resolution, and was developed in support with field visits byGarcia-Serrano (2015). I simplifiedthe land use classifications into the main categories of forest, pasture, residential, agriculture (rice only),and agriculture (rice and melons, which is a double cropping system of the two crops grown sequentiallyduring a single 12-month period). The separation into the rice and melon categories in the agricultureclassification accounts for the fact that rainfed rice is grown on a larger area than groundwater-irrigatedmelons, and that part of the rice fields (wet season) lie fallow during the melon season (dry season),with consequences for evapotranspiration and irrigation.To assess potential changes in land use between the start of the modelling period (2005) to the endof the modelling period (2016), I conducted a land use change analysis using QGIS and Google Earthto compare the 2010 land use map to the 2005 and 2016 imagery for the extent of the watersheds.Specifically, I digitized land use from Google Earth imagery from March 2005 as well as from February2016 and compared them to the 2010 land use map (Table 3.3). Overall, all land use changes between412005 and 2016 aected less than 4% of the total watershed area. Considering the low percentages ofland use change found, and the limitations of this analysis due to the lack of high resolution imagery, I as-sumed for hydrological modelling purposes that the land use of the studied watersheds was adequatelydescribed by the mid-point 2010 land use map and was treated as constant throughout the modellingperiod from 2005 to 2016.Table 3.3: Change in land use/land cover in the Potrero and Caimital watersheds from 2005 to 2010,and from 2010 to 2016.Forest Agriculture Pasture Residential2010 Land use/land cover [ha] 3,950 591 2,887 202Change from 2005 to 2010 [relative to 2005 coverage] + 1% 0.0 - 3% +15%Change from 2010 to 2016 [relative to 2010 coverage] +6% +5% -6% + 1%The second criterion for HRU definition was soil and geology, which was based on the three geologicalzones: basalt within the hills, colluvial material (loose and unconsolidated sandy colluvial deposits) inthe transition from hills to valley, and alluvial material (with a top layer of silty-clay material over a layerof sand to gravel with clay lenses) (see also section 2.1 and Figure 2.4 for geological zones). Thissimplification into three zones was based on available soil and geology for which no more detailedspatial data existed.Land cover within the basalt zone is dominated by pasture and forest, and no residential or agriculturalclasses are present. The basalt zone is not underlain by the Potrero-Caimital aquifer (Figure 2.4),and rainfall-runo processes in this zone were modelled with two soil layers (soil layer 1 and 2, seeFigure 3.4a). In contrast, rainfall-runo processes in the alluvial and colluvial zones were modelledas conceptualized in Figure 3.4b, where the top soil layer 1 of each HRU and SC is underlain by thePotrero-Caimital aquifer.Slope was the third HRU criterion. It was determined based on the HydroSHEDS DEM, aggregated intofour slope classes (Figure 3.6). The alluvial-colluvial valley zone was mostly flat (0 to 10% slope). Thebasalt zone is generally steeper than the alluvial-colluvial valley zone and all four slope classes (0-10%;10-20%; 20-30%; 30-80%) were represented within the HRUs of this zone.42Figure 3.6: Slope categories for HRUs based on HydroSHEDS DEM.Within an SC of the basalt zone there could be eight dierent fractional areas (HRU), and within an SCof the alluvial and colluvial zones (connected to aquifer) there could be ten dierent fractional areas,resulting in a total of 18 dierent HRU classifications (Table 3.4). Within the 79 SCs, a total of 3,327HRUs were delimited (Figure 3.7). Considering limited data availability, I assumed that an HRU with thesame classification (i.e., same land cover, slope, and soil characteristics) in one SC is characterized bythe same soil parameters as the same HRU classification in another SC.43Table 3.4: HRU classifications for the Potrero and Caimital watersheds.HRU # Geological zone Land use/land cover Slope1 Alluvial Agriculture - rice/melon < 10%2 Alluvial Agriculture - rice/fallow < 10%3 Alluvial Forest < 10%4 Alluvial Pasture < 10%5 Alluvial Residential < 10%6 Colluvial Agriculture - rice/melon < 10%7 Colluvial Agriculture - rice/fallow < 10%8 Colluvial Forest < 10%9 Colluvial Pasture < 10%10 Colluvial Residential < 10%11 Basalt Forest 0-10%12 Basalt Forest 10-20%13 Basalt Forest 20-30%14 Basalt Forest 30-80%15 Basalt Pasture 0-10%16 Basalt Pasture 10-20%17 Basalt Pasture 20-30%18 Basalt Pasture 30-80%Figure 3.7: Spatial distribution of sub-catchments (SCs) and hydrological response units (HRUs).443.5.3 Meteorological input dataClimate was assumed to be uniform over the two adjacent and relatively small watersheds, as limitedhistorical weather stations were available in proximity to the watersheds. Daily total rainfall data wereobtained from the National Meteorological Institute of Costa Rica (Instituto Meteorológico Nacional, IMN)from January 2005 until December 2013 from the Nicoya Extension Agricola station (station ID#72101),and from January 2014 to June 2014 from the Nicoya Centro Station (ID#72165) (Table 3.5). The NicoyaCentro Station had replaced the Nicoya Extension Agricola station at the end of 2013. Starting in July2014, rainfall was measured at the Eddy Covariance (EC) station (operated by the Futuragua researchproject) which is located at the center of the two watersheds (see Figure 2.9 for station location). Fromthe 30-minute measurements at the EC station, daily totals were generated. These data were gap-filledwith records from the nearby Davies Weatherlink station and were used for the modelling period fromJuly 2014 to December 2016.Best quality measurements of air temperature, relative humidity, and wind speed data were availablefrom the EC station for the duration of the calibration period from 2014-07-04 to 2016-12-31 (Table 3.5).Daily averages of those variables were generated from 30-minute measurements, and gap-filled withWeatherlink station data if needed. For the early period from 2005 to 2007, no meteorological datawere available from meteorological stations in the region. Therefore, I used Climate Forecast SystemReanalysis (CFSR) air temperature, relative humidity and wind speed data (Saha et al. 2010; Dee et al.2014) for this period. From February 2007 to September 2012, Weatherlink station data were availablefor air temperature and relative humidity, and the CFSR wind speed data were used. CFSR data werealso used from September 2012 to the start of our own monitoring in July 2014. Monitoring and dataprocessing of meteorological data from the EC monitoring station was done by Futuragua researchpartners.Cloudiness fraction (the number of bright sunshine hours per day over the total hours of daylight) is con-verted to net radiation in the WEAP model. I estimated the cloudiness fraction based on total incomingshort-wave solar radiation, extraterrestrial solar radiation and the Ångström coecients (Shuttleworth,1993; Equation 3.6):St =Ås + bsnNãS0 (3.6)where St⇥MJm2dy1⇤ is the total incoming short-wave solar radiation, s [] and bs [] arethe Ångström coecients (s is the fraction of extraterrestrial radiation S0 on overcast days (n = 0),and s + bs is the fraction of extraterrestrial radiation S0 on clear days), nN [] is the cloudinessfraction, with n [hor] as the bright sunshine hours per day and N [hor] as total day length, andS0⇥MJm2dy1⇤ is the extraterrestrial radiation.45Total incoming short-wave solar radiation was available from the EC station from 07-2014 to 12-2016,and from the CFSR dataset for the earlier time period. I obtained modelled extraterrestrial solar ra-diation based on the latitude and longitude of the EC station and the day of year (Bojanowski 2016).Measurements of total incoming short-wave radiation and total extraterrestrial solar radiation can beused to estimate the Ångström coecients as and bs (Shuttleworth 1993). I estimated the Ångströmcoecients based on the maximum and minimum of the ratio of the measured incoming short-wavesolar radiation from the EC station and the extraterrestrial solar radiation during that period (as = 0.05;bs = 0.7). I then solved Equation 3.6 for the cloudiness fraction and thus generated this input variable.Table 3.5: Data sources for rainfall (daily total), air temperature (daily average), relative humidity (dailyaverage), wind speed (daily average) and incoming solar radiation (daily total) variables used as input inthe WEAP model for the Potrero and Caimital watersheds. IMN = Instituto Meteorológico Nacional(Costa Rican National Meteorological Institute); CFSR = Climate Forecast System Reanalysis; ECstation = Eddy Covariance monitoring station operated by Futuragua research project; Weatherlink =Weatherlink monitoring station operated by farm owners in close proximity to EC station.Period Rainfall Air temperature Relative humidity Wind speed Solar radiation2005-01-01 to2007-02-01IMN CFSR CFSR CFSR CFSR2007-02-02 to2012-09-18IMN Weatherlink Weatherlink CFSR CFSR2012-09-19 to2014-07-03IMN CFSR CFSR CFSR CFSR2014-07-04 to2016-12-31EC station EC station EC station EC station EC station3.5.4 EvapotranspirationTo model evapotranspiration, the crop coecient Kc was needed as input parameter. Kc representsthe ratio between actual evapotranspiration (ET) and reference evapotranspiration of each land usetype. Five main land use types exist within the Potrero and Caimital watersheds (forest, pasture,melon crops, rice crops, and residential). Mean Kc values for each day of year were estimated formelon and rice crops from actual ET values measured at the EC station at the melon-rice farm andestimated crop reference potential evapotranspiration ETreƒ (as calculated by WEAP using the modifiedPenman-Monteith equation, Equation 3.1). Specifically, Kc was calculated as the daily ratio of measuredET values from 2015 (drier year) and 2016 (wetter year) and ETreƒ over the same time period. ETmeasurements from the start of the monitoring in July 2014 to December 2014 were not included, as2014 was also a drier year. Similar to processing for pasture and forest data (see below), 8-day averageswere calculated to smooth the Kc curve and lessen the eect of daily variability.No ET measurements were available for the other land use types. The Food and Agricultural Organiza-tion (FAO) of the United Nations provides Kc estimates for pasture (Allen et al. 1998), but these values46were developed for the northern hemisphere and do not reflect the strong seasonal behaviour of thewet-dry tropics, and further, no Kc estimates were available for seasonally-dry forest in the literature.Therefore, I estimated Kc for the non-crop land use types using the MODIS evapotranspiration productMOD16 (MODIS 16A2/Terra Evapotranspiration 8-Day (1000m spatial resolution) (ORNLDAAC 2008b).To relate MODIS ET to dierent land use types within the Potrero and Caimital watersheds, I selectedMODIS pixels that were dominated by one land use type (fraction cover > 89% for pasture and forest)(Figure 3.8), according to the SENARA land use classification from 2010 and under the assumptionthat land use changes throughout the study period stayed minimal. ET values were extracted for eachselected homogeneous land use MODIS pixel, and mean values of all homogeneous pixels for eachland use type were calculated for each 8-day period from 2005 to 2014 (MOD16 was not available for2015 and 2016 at the time of the WEAP model setup) (see Appendix B for time series of forest andpasture ET means).Figure 3.8: Selected MODIS ET pixels (1000 m resolution) and SENARA 2010 land use classification.Note that for selected pixels along the watershed boundary, I verified using Google Earth imagery thatsame land use also covered the pixel outside the watershed boundary.As MODIS reports ET as totals of 8-day periods, I calculated the 8-day total of the ETreƒ . Then Icalculated Kc values for the period 2005 to 2014 as the ratio between MODIS ET for each land use47type and ETreƒ (as calculated by WEAP using the modified Penman-Monteith equation), and allocatedthe 8-day Kc values to daily values. Next, to generate a daily Kc value to be used for every year of themodelling period (considering that not all years had available MODIS ET data), I calculated a long-termmean value of daily Kc for each land use type for each day of year based on the daily Kc series from theperiod 2005 to 2014 (Figure 3.9). For fields where only rice was planted in the wet season and fieldslay fallow during the dry season, Kc values derived from measured ET were used during rice growingseason (wet season), and Kc values obtained from MODIS ET for pasture land use type were used forthe dry season.Figure 3.9: Daily Kc values, generated from EC station ET and ETreƒ for melon and rice, and fromMODIS ET and ETreƒ for forest and pasture. Note that Kc values for melon/rice and fallow/rice are thesame, apart from the melon season for Days of Year 30-114. Forest and pasture: mean (dark lines) andstandard deviation (light shading) over the time series from 2005 to 2014. The variability in ET betweendierent forest and pasture pixels is indicated in Appendix B. Melon and rice: mean over time seriesfrom 2015 to 2016 (here, no standard deviation is given as time series only included two years).3.5.5 Soil input parametersSoil input parameters for the WEAP model include eective water holding capacity and saturated hy-draulic conductivity for soil layer 1 (top) and soil layer 2 (lower), runo resistance factor, and preferredflow direction (Table 3.6). Parameters of soil layer 1 can be adjusted for each HRU, while parameters forsoil layer 2 are assigned for the entire SC (see also Figure 3.4). Most of these parameters are typicallyused as calibration parameters (Sieber and Purkey 2015). Further details on calibration parameters arealso provided in section 3.6 of this chapter.The eective water holding capacity (or available water holding capacity) is the dierence betweenvolumetric water content at field capacity and at the permanent wilting point, and is represented in the48WEAP model in mm over the depth of the soil layer (Table 3.6). Garcia-Serrano (2015) measured thewater holding capacity for dierent geology and land use units within the Potrero andCaimital watershedsin 2015 (Appendix B). I assumed that an HRU in the Potrero watershed has the same soil characteristicsas an HRU with the same geology and land use units in the Caimital watershed. Next, I quantified thewater holding capacity in relation to the depth of soil layer 1 (Table 3.6).I assumed that the soil depth in each SC is constant, but that the water holding capacity varies accordingto the HRU unit (Table 3.6). No continuous datasets on soil depth were available for the Potrero andCaimital watersheds, but lithology logs of 29 wells were available in reports from SENARA (Garcia-Serrano 2015; Agudelo 2006), which I digitized and integrated into a database. Soil depth ranged from1 to 10 m, and I associated each SC with a soil depth based on these data (Figure 3.10). However, mostof the lithology logs provided just the total soil depth above the aquifer, and only in some cases moredetailed information on soil or aquifer material. Further, only a few lithology logs existed for the colluvialand basalt zone, and I thus used the soil depth as calibration parameter for these two geological zones.Table 3.6: List of soil input parameters, range of values, and the sources of the values used for thePotrero and Caimital watershed model. Soil layer 1 refers to top soil layer, soil layer 2 to lower soil layer.See also Equation 3.2 & Figure 3.4 for soil rainfall-runo conceptualizations.Parameters Value range SourceSoil layer 1Eective water holding capacity[mm] = eective water holding capacity [] x soil depth [mm]Eective water holding capacity [] (0.18, 0.21) Field measurements by Garcia-Serrano (2015)Soil depth alluvial zone [mm] (1000, 10000) Based on lithology logs (Agudelo 2006;Garcia-Serrano 2015)Soil depth colluvial zone [mm] (500, 7200) Calibration parameterSoil depth basalt zone [mm] (500, 5000) Calibration parameterSaturated hydraulic conductivity⇥mmdy1⇤ (0.0005, 3583) Calibration parameterPreferred flow direction (PFD) [] (0.2, 0.9) Calibration parameterRuno resistance factor (RRF) [] (0.1, 600) Calibration parameter (seasonal variability basedon Leaf Area Index)Soil layer 2 (basalt zone only)Eective water holding capacity [mm] (250, 5000) Calibration parameterSaturated hydraulic conductivity⇥mmdy1⇤ (10, 35) Calibration parameter, initial value (34.6) based onDomenico and Schwartz (1997)49Figure 3.10: Soil depth at lithology logs (from SENARA; Agudelo 2006; Garcia-Serrano 2015), andcategorized soil layer 1 depth for each sub-catchment as input for the WEAP model.The saturated hydraulic conductivities of soil layer 1 were used as calibration parameters, as onlyinfiltration capacities at the root zone existed as field measurements from Garcia-Serrano (2015). High-end value ranges in Table 3.6 (and further below in Table 3.12) depict these infiltration capacities,which were used as initial model parameters, while lower hydraulic conductivities were explored duringcalibration.The preferred flow direction (PFD) parameter, also required for theWEAPmodel, partitions the outflowof the top soil layer between interflow and downward percolation (see also Equation 3.2) and it is relatedto soil, land cover, and topography. It ranges between 0 (100% vertical flow) and 1 (100% horizontalflow) and is generally used as a calibration parameter in WEAP modelling. General trends of PFDincluded higher horizontal flows for pasture, agriculture and residential land cover than for forest, as wellas higher horizontal flows in alluvial and basalt zone than in colluvial zone, and with steeper slopes (forvalue ranges, see Table 3.6). The parameter was adjusted during model calibration for the HRUs.The runo resistance factor (RRF) is related to aspects such as slope and leaf area index (LAI), orland use cover (Sieber and Purkey 2015). Higher/lower values of the RRF lead to less/higher surfaceruno (higher/lower resistance). No equation is provided to estimate RRF from data and it is usually50used as a calibration parameter (Sieber and Purkey 2015). However, LAI and surface resistance willchange over the course of a year, in particular in a region like the wet-dry tropics where greening-up ofvegetation and senescence are so pronounced. Therefore, I used LAI data to estimate the inter-annualvariability of the RRF. I used the MODIS LAI product (MODIS 15A2 Leaf Area Index and Fraction ofPhotosynthetically Active Radiation (FPAR) 8 Day composite MODIS Collection 5 Land Product) for theperiod 2000-02-18 to 2017-01-01 (ORNL DAAC 2008a).In order to capture the long-term annual seasonality of LAI for each land use type (forest, pasture oragriculture), I first assigned a MODIS LAI value to each land use type and then estimated a mean valueof MODIS LAI for each day of the year for the period 2000-2016. To relate LAI to dierent land covers,MODIS pixels with a high fraction cover of one single land use (forest, pasture or agriculture) wereselected and mean LAI was calculated for all pixels of the same dominant land use (fraction cover >89% for forest and pasture; > 50% for agriculture). Next, I estimated the 2000-2016 mean value of each8-day period of the year reported by MODIS for each land use type. Then, 8-day mean values wereallocated to obtain a daily time series of LAI per land use type for the period 2000-2016, and it wassmoothed to reduce the eects of annual variabilities (Figure 3.11). Annual means of LAI per land usetype were also calculated. Daily RRF values were expressed in the WEAP model as the annual meanLAI per land use plus or minus the daily variation of LAI from the annual mean. This facilitated the useof the annual mean of the three land use types as a calibration parameter and allowed for changing thisvalue during calibration while keeping the seasonal variability. The RRF was assumed constant overtime for residential land cover type. RRF values were also dierentiated between the four slope classesfrom the HydroSHEDS DEM, with values adjusted during calibration.Figure 3.11: Mean (line) and standard deviation (shading) of leaf area index (LAI) for (a) forest, (b)pasture, and (c) crops, smoothed, over the period from 2000 to 2016, calculated from MODIS 15A2(ORNL DAAC 2008a).51Soil layer 2 was only eective for the basalt zone, as it was replaced by the Potrero-Caimital aquifer inthe alluvial and colluvial zones (see Figure 3.4). Parameters for soil layer 2 in the basalt zone were usedas calibration parameters (for value ranges, see Table 3.6), with the initial value for saturated hydraulicconductivity based on permeable basalt conductivity reported by Domenico and Schwartz (1997).3.5.6 River reachesThe modelled river reaches were the main stems of the Potrero and Caimital rivers and their maintributaries, based on a detailed river GIS layer obtained from the ACT in Nicoya (ACT 2013). Thesereaches traverse the alluvial-colluvial valley underlain by the Potrero-Caimital aquifer (i.e. where surfacewater and groundwater interact). River reaches were not modelled at higher elevations beyond theboundary of the aquifer, and headwater catchment runo was assumed as inflow into the first modelledriver reach (see Figure 3.5 for modelled river reaches). Based on a study on riparian evapotranspirationfor tropical headwatersheds in Costa Rica (Cadol et al. 2012), evapotranspiration from river reacheswas assumed at 2.5% of daily streamflow.3.5.7 GroundwaterThe Potrero-Caimital aquifer extends through the alluvial-colluvial zone (Figure 3.5, and Figure 2.4), andtherefore, SCs in this zone were connected to the aquifer module in theWEAPmodel (as conceptualizedin Figure 3.4b). The Potrero-Caimital aquifer was modelled as one aquifer storage unit within the WEAPmodel, thus combining the alluvial/colluvial material and the underlying fractured basalt into one storageunit, as it is not possible in the WEAP model to assign dierent properties to dierent material within anaquifer storage unit. WEAP thus describes the total amount of water entering the groundwater systemthrough diuse and focused recharge, as well as withdrawals from the aquifer through anthropogenicpumping and baseflow to streams for each time step. Interaction with surface water is determined byaquifer properties, groundwater storage volume as well as stream water level (see also Section 3.4).The basalt zone on the surrounding hills (see Figure 2.4) is not underlain by the Potrero-Caimital aquifer,and was therefore modelled with two soil layers, as conceptualized in Figure 3.4a. A limitation of theWEAP conceptualization of aquifers is that it is not possible to connect interflow from the lower soil layer2 from hills (here, the basalt zone) as potential lateral recharge to an aquifer (should that occur withinthe system), but that instead it contributes as baseflow to headwater streams (Figure 3.4a), from whichit, in turn, may contribute via infiltration through streambeds as focused recharge to the aquifer.Aquifer characteristics such as specific yield (0.002) and saturated hydraulic conductivity (26mdy1)were based on values reported by the Costa Rican groundwater agency SENARA (Garcia-Serrano 2015;52Losilla and Agudelo 2003). These previous studies estimated the specific yield of the aquifer from fivepumping tests, and found that it varied between 0.0007 and 0.008, with a mean of 0.002 (Losilla andAgudelo 2003). They estimated the total storage volume of the aquifer as 40 Mm3, assuming an area of20 km2, a mean aquifer thickness of 20 m, and a mean specific yield of 0.002 (Garcia-Serrano 2015;Losilla and Agudelo 2003). Losilla and Agudelo (2003) reported that saturated hydraulic conductivities(determined based on transmissivity measurements and aquifer thickness) ranged between 4.5 and 95mdy1, with a reported mean of approximately 26 mdy1. This mean represents the averageacross the dierent material of the aquifer (no details were given in the report on individual measure-ments). Considering these uncertainties and limited details available regarding hydraulic conductivitiesand specific yield, I explored a range of values during the calibration process, but found that ultimatelythe reported mean values represented the aquifer conditions best, and adopted these for the modelling(details in section 3.6 on model calibration).Diuse groundwater recharge is modelled through the rainfall-runo routine (Figure 3.4b) where down-wards percolation through the upper soil layer 1 of each HRU enters the aquifer (if the SC is connected toan aquifer such as in the alluvial and colluvial zones). Focused recharge is modelled through the surfacewater – groundwater interaction, where river reaches can be dynamically linked to an alluvial aquifer.This was done for all modelled stream reaches within the alluvial and colluvial zones that are underlain bythe Potrero-Caimital aquifer. Parameters related to modelling surface water – groundwater interactionwere used as calibration parameters. These parameters included wetted depth in river, groundwaterstorage at equilibrium with the river, and horizontal distance from farthest edge of aquifer to river (seeSection 3.4 for details on parameters and conceptualization).Table 3.7: Groundwater-related parameters.Parameter Range SourceSpecific yield [ - ] (0.002, 0.23) Calibration parameter, initial values based onAgudelo (2006) and Garcia-Serrano (2015)Saturated hydraulic conductivity [mdy1] (1, 600) Calibration parameter, initial values based onAgudelo (2006) and Garcia-Serrano (2015)Total storage volume of aquifer [Mm3] (40) Agudelo (2006) and Garcia-Serrano (2015)Wetted depth [m] (0.5, 1) Calibration parameterHorizontal distance [m] (400, 4000) Calibration parameterStorage volume at equilibrium with river at wetteddepth [Mm3](15, 40) Calibration parameter533.5.8 Water demands by dierent sectorsDomestic and business/municipal water demand in the Potrero and Caimital watersheds includes bothrural villages locatedwithin the watersheds (serviced by ASADAs), as well as water transfers via pipelinesto the towns of Nicoya and Hojancha located outside the watersheds (serviced by AyA). Water demandsin towns and villages consist of domestic (household), municipal (schools, churches, cemeteries, mu-nicipal oces, etc.) and business (stores, etc.) water demands. Most rural households also operatetheir own shallow household well (“artisanal well”) that supplements the water received from the ASADAdistribution system. Agricultural irrigation and cattle ranching constitute further water demands in the wa-tersheds, as are further water license holders who were not included in the above categories. Therefore,to model surface water and groundwater demand from the studied watersheds, I accounted for:• Rural villages within watersheds: Water provided by ASADA’s for domestic andmunicipal/businessuse; Artisanal household pumping for domestic use• Towns outside watersheds with water transfers (Nicoya and Hojancha): Water supplied by AyA fordomestic and business/municipal use• Agriculture: Irrigation of melon crops (rice is rainfed)• Ranching: Water use for cattle• Additional water licensesWater demand of rural villagesMost ASADAs of rural villages in the Potrero and Caimital watersheds operate central groundwaterpumping wells from the Potrero-Caimital aquifer, and pipelines distribute water to households. The ruralvillages of Curime, Caimital, Dulce Nombre, Gamalotal, and Hondores (see Figure 2.6 for locations)each have their own ASADA water distribution system. They were therefore included as demand sitesin the WEAP model. The village of Varillal is part of the ASADA of Curime, and its water demand isincluded therein. The village of La Virginia is located at the outflow of the Caimital watershed and isadministered by its own ASADA. However, as it is further from the Potrero-Caimital aquifer, it relieson water from a small mountainous aquifer (southwest of the Caimital watershed) and on a surfacewater spring from which water is withdrawn, treated with chlorine, and stored in a tank from which waterpipelines are fed via gravity. Both these water supplies are not located within the Potrero and Caimitalwatersheds, and therefore, the La Virginia water demand was not included in the model. The villageof Casitas is located in the Potrero watershed. But as it is close to Nicoya, it is part of the AyA waterdistribution system for the town of Nicoya. Its main water demand was therefore included in the Nicoya54demand site in the WEAP model. Households in Casitas, however, have artisanal wells that draw fromthe Potrero-Caimital aquifer, and these were included as separate demand site in the model.Rural domestic water demand was estimated based on annual population. The Costa Rican censusesfrom 2000 and 2011 (INEC 2001, 2011) include population data at the district and canton level. Pop-ulation data for some of the villages were available in Morataya Montenegro (2004) for 2000, and Iobtained 2014 population data for rural villages from the Ministerio de Salud in Nicoya (Table 3.8).Annual population for the model period from 2005 to 2016 was estimated based on these datasets witha geometric growth model, assuming that the population increased each year by a constant proportion(Table 3.9). This is consistent with the method chosen by INEC (INEC 2014). For Casitas and Hondores,only the combined population was available for 2000, and it was assumed that the ratio of the populationin Casitas versus Hondores was the same in 2000 as in 2014. For Gamalotal and La Virginia, for whichno 2000 population data were available, it was assumed that the mean of the growth rates of the othertowns represents the growth rate for these towns.Table 3.8: Annual population estimates for rural villages in the Potrero and Caimital watersheds, for first(2005) and last year (2016) of modelling period, as well as for 2000 and 2014 when data were available.Based on INEC (2001, 2011) and Morataya Montenegro (2004), and data obtained from the Ministeriode Salud in Nicoya. *Population for Casitas in 2000 includes Hondores.Year Casitas Caimital Curime DulceNombreGamalotal Hondores La Virginia Varillal Total2000 428* 346 656 175 - * - 32 -2005 244 498 738 253 303 204 254 43 2,5372014 266 957 914 490 448 222 377 74 3,7482016 271 1,107 958 568 489 226 411 83 4,113Table 3.9: Geometric annual growth rates in % for population in rural villages, calculated based onINEC (2001, 2011) and Morataya Montenegro (2004), and data obtained from the Ministerio de Saludin Nicoya.Casitas Caimital Curime DulceNombreGamalotal Hondores La Virginia Varillal0.9 7.5 2.4 7.6 4.3 0.9 4.3 6.2While water consumption is metered at the household level and monthly bills are issued by all of theASADAs, most of the water use data are not kept for records. Only the ASADA of Caimital in one of thebigger villages had a full-time employee who maintained records of hand-written monthly water bills foreach household (available from 2012 to 2015), which I had digitized (see Chapter 2, section 2.5) (Figure3.12).55Figure 3.12: Water use at the ASADA Caimital. (a) Total monthly use of the ASADA, and (b) Mean perhousehold use. Number of households were 261, 271, 278 and 292 by the end of the year 2012, 2013,2014, 2015, respectively (see Appendix C).To obtain the annual water use rate per person as input for the WEAP model, total annual domesticwater use in Caimital was divided by the estimated annual population for that year (see Appendix B).Considering that both population data as well as water use data were available for 2014, the annual perperson water use rate of 77m3 from 2014 was assumed as annual water use rate for all years when noother data were available. The mean number of people per household for 2014 (3.6) was also similar tothe reported mean from the census of 2011 for the Nicoya canton (3.67, INEC 2011). The annual perperson water use rate of 77m3 was also assumed for the modelling period from 2005 to 2016 for theother rural villages in the Potrero and Caimital watersheds, as there were no other data available andthe rural villages are in close proximity to each other and similar in structure.In the WEAP model, daily activity levels indicate the variability of water use throughout the course of ayear. They were estimated from the monthly water use data from Caimital for 2012 to 2015, assumingthat daily activity levels stayed constant throughout a month. The mean daily activity levels from 2012 to2015 were assumed for all other years of the modelling period. It was also assumed that the daily activitylevels from Caimital are representative for the other villages in the Potrero and Caimital watersheds.Non-domestic water use for municipal use (such as schools, church, cemetery) and for business use(stores) was indicated on the original water use bills fromCaimital that were digitized, andwere separatedfrom the household water use calculations. Mean annual water use rates for these non-domestic siteswere calculated for 2012 to 2015 for Caimital. It was assumed that the number of non-domestic sites andthe mean annual water use rate stayed constant throughout the modelling period from 2005 to 2016, asmost of the non-domestic use was for basic infrastructure. To estimate the number of sites for all othervillages, the fraction of their population in comparison to the population of Caimital in 2014 determinedthe fraction of municipal/business sites in each town. Daily activity levels were estimated similarly to thedomestic activity levels from the monthly water use data from 2012 to 2015.56While ASADAs provide water for drinking and indoor water needs, many households supplement thiswith water from their own shallow artisanal groundwater pumping well or surface water pump. Peoplemight operate these private artisanal wells due to the lower costs of electricity for pumping rather thanpaying for ASADA water, or use of these wells may provide some independence from a sometimesunreliable water distribution network. The water from these artisanal pumping wells is typically usedfor outdoor applications, such as laundry, watering yards and gardens, cleaning cars and porches, andspraying down dusty roads during the dry season. To my knowledge, no data exist for household levelpumping for non-potable uses as a supplement to domestic drinking water supplied by water utilities.However, I was able to estimate artisanal water use from one of my groundwater monitoring stations(GW1 in Varillal) installed in an artisanal well, where high-frequency monitoring (10-minute intervals)of groundwater levels showed daily groundwater drawdown due to pumping (Figure 3.13). I assumedthe time from start of well drawdown to lowest water level in well to represent the pumping time, andestimated from that the total pumping time per day (in most cases it approximated two hours of pumpingper day). According to the property owners, the mean pumping rate of the pump was 20 liters per minute.Based on this information, I estimated the total daily extraction per household and per person over thetime of monitoring (December 2013 to December 2016), and from this, calculated the mean annual perperson water use rate (48.8m3) and mean daily activity levels. The mean annual per person wateruse rate and mean daily activity levels were used as an assumption across all the rural villages of thePotrero and Caimital watersheds. Not all households have their own artisanal water supply, and basedon conversations with local water utility ocials, an average of 60% of the population was assumed touse artisanal wells in addition to the ASADA-provided water supply.Figure 3.13: Observed water depth (10-minute frequency) at artisanal household well GW1 in Varillal(for location see Figure 2.9), with daily household pumping for outdoor water use.57Water demand in townsWater supplies for the town of Nicoya include the Potrero-Caimital aquifer, the Potrero river, and theNicoya aquifer. Water supplies for the town of Hojancha include the Potrero-Caimital aquifer and theHojancha aquifer. As neither Nicoya nor Hojancha is located within the Potrero and Caimital watersheds,their artisanal water demand was not included in the WEAP model. Population data for Nicoya andHojancha were available at the district level from the 2000 and 2011 census (INEC 2001, 2011). Again,a geometric growth rate between these two data points was used to estimate the annual population from2005 to 2016 (see Appendix C).NicoyaThe district of Nicoya includes rural villages that are not connected to the AyA water distribution system,and therefore, only a fraction of the total district population draws on the water supply of the Potrero andCaimital watersheds. I obtained the number of connection sites (i.e., number of households, municipalinstitutions, and business sites connected to the AyA distribution network) and the average annualwater volume used by each of these connection categories for the year 2016 from the AyA Nicoya (seeAppendix C). Some data were also available for the year 2008 (Fernandez-Sing 2009), which permittedme to estimate change of the number of households and annual water use rates over time.To convert mean annual water use per household into mean annual water use per person, I used meannumber of persons per household for the district of Nicoya, as determined in the 2000 and 2011 censuses(3.87 and 3.41, respectively, INEC 2001, 2011), interpolated between 2000 and 2011 to obtain anestimate for 2008, and assumed it to remain constant after 2011. To estimate number of connectionsites each year between 2005 and 2016, I calculated the geometric growth rate from 2008 to 2016(3.8%), and assumed that the same growth rate could be applied prior to this period (i.e., from 2005 to2008). I calculated the annual change in mean annual water use rate per person between 2008 and2016, and assumed the 2008 rate for 2005 to 2007. I also estimated the percentage of the populationof the district of Nicoya serviced by AyA Nicoya based on connection sites and people per household(see Appendix C). Further, I assumed that the municipal/business annual water use rate per connectionstayed constant at the 2016 rate (see Appendix C).I also obtained monthly extraction volumes for each of the AyA-operated groundwater wells within thePotrero-Caimital aquifer, the Nicoya aquifer, and the Hojancha aquifer, as well as from the Potrero riverextraction plant for the time period from 2005 to 2016 from the AyA in Nicoya. These are the totalextraction volumes that include domestic, municipal/business, as well as unaccounted water uses andwater leakages from the pipeline system. Unaccounted water use and leakage volumes were estimatedas the dierence between total extraction volumes and the sum of domestic and municipal/business58water use volumes. It accounted for 15 to 27% of total water extraction volumes and was added tothe model as water loss. In 2016, domestic water use was 62% of total extraction volumes, whilemunicipal/business water use was 18%, and lost or unaccounted water was 20%.Daily activity levels of water use were based on the monthly total extraction rates, and it was assumedthat each day of the month contributed equally to monthly water use. To represent the extraction ratesfrom the dierent water supplies, I linked the water demand site (Nicoya) to the three dierent watersupply sources (Potrero-Caimital aquifer, Potrero River and Nicoya aquifer) via transmission links (whichrepresent water pipelines in the WEAP model). For each month from 2005 to 2016, I calculated thepercentage of water extracted from the dierent water supply sites in relation to total monthly extractionvolumes, and used this to restrict the maximum flow to the demand site within the transmission link.HojanchaSimilar to the district of Nicoya, the district of Hojancha includes rural villages surrounding the townof Hojancha that are not part of the AyA distribution network, and therefore, were not included in thedemand modelling. I obtained the number of connection sites and the average annual water rates foreach connection type for the year 2016 for Hojancha from the AyA Nicoya. To estimate the populationserviced by AyA, I assumed the mean of people per household from the 2011 census (INEC 2011) forthe year 2016, and multiplied it with the number of total domestic connection sites for each year.For the other years of the modelling period, no data for Hojancha were available on connections sitesand water use rates. Considering that I had obtained the total AyA extraction volumes for each year from2005 to 2016 for Hojancha, I assumed that the fractions of the total extraction volumes for domestic andmunicipal/businesses use, and lost water remained the same over the modelling period (60%, 10%, and30%, respectively). From this, I estimated the domestic and municipal/business water demand for eachyear (see Appendix C). Daily activity levels were based on the monthly total extraction rates, assumingthat each day of month contributes equally to total monthly extraction volumes.Water supplies for Hojancha include the Hojancha aquifer and the Potrero-Caimital aquifer. Within themodel, the Hojancha demand site was connected via transmission links to both supply sources. TheHojancha aquifer was not modelled as an aquifer, but as another supply source where the extractedwater was available. For each month from 2005 to 2016, I calculated the percentage of water extractedfrom the dierent water supply sites in relation to total monthly extraction volumes based on obtaineddata, and used this to restrict the maximum flow to the demand site within the transmission link.59AgricultureTo account for water demand for melon irrigation during the dry season, I used irrigation data obtainedfrom one of the two large melon farms (La Costeña) located within the watersheds (Morillas et al. 2018).I assumed that melon irrigation at the other melon farm in the watershed used the same amount of waterper area of planted melons as both farms are operated similarly, planted similar melon crops, and useddrip irrigation systems. In total, throughout the two watersheds, there are 405 ha of irrigated melons andan additional 185 ha where only rice is planted during the wet season and fields lay fallow during thedry season.Melon plants are started in the greenhouse and transplanted between the beginning of December andthe beginning of February; the dierent fields are planted one after the other. Similarly, harvest startsat the end of January for the early fields, and continues until April. Based on the 2014-2015 and 2015-2016 planting and harvesting dates for the La Costeña fields (147.51 ha and 163.2 ha planted in total,respectively), there is a varying fraction of the total farm area under cultivation at any point of the melonseason (Figure 3.14).Figure 3.14: Total area of melons under production for each day of the 2015 to 2016 melon seasonat melon/rice farm, estimated from transplanting and harvesting dates for each of the fields, and thereported area size of each field.This planting scheme and high spatial resolution for every field would have been dicult to representin the WEAP model, and it also varies between dierent years. Furthermore, evapotranspiration mea-surements (and melon specific Kc values) represent planting and irrigation pattern of fields within thefootprint of the eddy covariance tower at the EC monitoring station. Therefore, to characterize irrigationand evapotranspiration response consistently in the model, I assumed that all melon fields are plantedand harvested at the mean planting and harvesting dates for the 2014-2015 and 2015-2016 seasonsof the four fields within the footprint of the eddy covariance tower. That assumption resulted in a meloncycle of 67 days between planting (January 28) and harvest (April 5). Theoretical irrigation volumes perarea of planted melons over the course of one melon season were estimated based on actual irrigation60data obtained from the La Costeña farm (Laura Morillas, pers. comm.). Based on this dataset (Morillaset al. 2018), I assumed a mean daily water use per area per day for the melon growing period (3.95m3dy1h1). Within the WEAP model, I introduced an irrigation water demand site to each ofthe sub-catchments that had melon agriculture as land use, and connected transmission links to thePotrero-Caimital aquifer as all melons are irrigated with groundwater.The irrigation module within theWEAPmodel applies irrigation based on soil moisture content. However,current irrigation practices at the farms do not take soil moisture into account, but rather use a constantdrip irrigation rate that varies from one to three hours per day depending on crop stage. To representthese practices in the model, I assigned a low soil moisture threshold to initiate irrigation, and a highthreshold to stop watering. I then restricted the water flow within the transmission link (water pipelines)to represent the actual extracted water amounts. Modelling irrigation as part of the catchment routinesallows irrigation water to enter the catchment system, i.e., to model evapotranspiration and infiltrationinto the soil. Irrigation volumes are based on area of planted melon, and I calculated the total dailyirrigation for each sub-catchment. After the harvest, melons are washed within packaging plants. Watervolumes for melon washing were available for 2016. These volumes are based on the hectares of melonsplanted by La Costeña in 2016. Again, I assumed that the same values could be used for all harvestedmelons and over the modelling period. Considering my assumption regarding a harvest date of April 5,I assumed all melon washing takes place over the course of two days (April 5 and 6). An approximatetotal water volume of 5,705 m3 is needed for melon washing of the 405 hectares of planted melons in thetwo watersheds (see Appendix C). I represented the washing of melons as a water demand site withinthe WEAP model.RanchingCattle pastures are an important land use in the watersheds (38% of total area). However, no data onwater use by ranches were available, apart from a few water licenses that did not cover all of the ranchesand also did not provide water use per area. The agricultural census from 2015 gives total number ofranches and cattle for the province of Guanacaste, categorized by ranch size (INEC 2015). I usedthis information to estimate the mean number of cattle per hectare (1.35 cttle h1, see AppendixC). Next, I assumed drinking water demand per animal (35.5 L dy1) from a study from a tropicalenvironment with Zebu steers, which are also the main breeds in Guanacaste (Ayantunde et al. 2002).From that, I estimated total ranching water demand per hectare.The agricultural census (INEC 2015) gives information on the number of farmers that use a specific watersupply in the district of Nicoya. Using this information, I estimated the percentages of groundwater (62%)and surface water (38%) supplies used for ranching, and modelled groundwater as withdrawn from thePotrero-Caimital aquifer. Surface water for ranches in the Caimital watershed was assumed to be all61withdrawn from the Caimital River, and similarly, surface water for ranches in the Potrero watershedwas assumed to be all withdrawn from the Potrero River. Considering that land use stayed relativelyconstant throughout the modelling period and no further data were available on ranching water demand,I assumed the same water demand for each year of the modelling period.Other water licensesAll operators of groundwater wells or surface water pumps have to apply for water licenses at the MINAE.Most of these water licenses were already covered in the demands discussed above (AyA, ASADA,irrigation wells, ranching). However, there were a number of extraction sites registered for domesticwater use that were not part of the AyA or ASADA water distribution system, and some extraction sitesregistered for irrigation that did not belong to the large melon farms.Water licensing data were available online at theMINAEwebsite (Direccion de Agua 2015). The datasetsincluded location, owner, water source, water use type as well as registered pumping rates. I processedthe licenses for all extraction sites from the Potrero and Caimital watershed to a spatial database. Wateruse types included domestic, domestic – business, domestic – irrigation, irrigation and ranch. To addto the WEAP model, I excluded all extraction sites that were already covered through the AyA, ASADA,melon irrigation (all licenses by major melon companies, and irrigation wells located within melon fields),and ranching water use.Based on the reported pumping rate in liters per second, the assumed total daily water use was estimatedfor each category. No data exist on actual water volumes used, and no operating hours are includedin the water licensing, but it is unlikely that the pumps were running for 24 hours. Based on irrigationhours, domestic/artisanal pumping hours, as well as discussion with local water managers, I assumeda conservative estimate of three hours of pumping per day (see Appendix C). All active licenses are forgroundwater use.I added the sites as separate extraction sites into WEAP, and assumed that water use stayed constantthroughout the modelling period. While the water licenses provide some indication of water extractionrates, real water extraction volumes are likely higher than reported, as illegal wells also exist in the areaor operators may pump more than reported. The water use estimates for this study are nonethelessmostly based on monthly extracted volumes documented by the water utility and agricultural companies,or estimated based on land use such as ranching (i.e., not based only on water licenses).623.6 Model calibrationIn the next step, I calibrated the WEAP model to the monitored streamflow data of the DownstreamPotrero River site (SW1), using a combination of manual sensitivity analysis and calibration, and anoptimization algorithm (the Parameter ESTimation PEST algorithm incorporated in WEAP) to estimatethe parameter set of best fit (Doherty 2015; Sieber and Purkey 2015). The period covered by streamflowmonitoring (July 2014 to December 2016) included both drier and wetter years (2015 and 2016, respec-tively), and thus reflected a range of hydrological conditions. No continuous streamflow data existedprior to the start of this research.It should be noted that, while many of the calibrated parameters are named after physical quantities thatcan be measured in the field, these parameters become “eective parameters” within the hydrologicalmodel environment (Beven 2012). Eective parameters refer to a single parameter value assigned toall points within a model unit, with the goal that model output based on this single parameter value hasthe same output as a model based on a heterogeneous field (Blöschl and Sivapalan 1995; Westernet al. 2002). They are adjusted during the model calibration process, and parameterize the sub-unitheterogeneity of a given process by aggregating (up-scaling) the smaller spatial scale variability (Bron-stert et al. 2005; Western et al. 2002). As hydrological models are imperfect simulators of complexenvironmental processes of the real world, often, mathematical representations that try to describehydrological processes in the model have been developed on a laboratory scale (Blöschl and Sivapalan1995; Western et al. 2002). This makes the parameters scale-dependent, and the scale at whichmeasurement techniques of parameter values (such as porosity and hydraulic conductivity) are availableis generally much smaller than what is needed to represent the response of model units (Beven 2012).Thus, on a larger spatial scale and in modelling a natural watershed, their physical validity decreases.In the next sections, I provide a description of the dierent calibration parameters for the WEAP modelof the Potrero and Caimital watersheds, including the explored range and their final values.For the runo resistance factor (RRF), I developed the seasonal variation of RRF based on the LeafArea Index (section 3.5.5) and described this variation relative to the annual mean. I then calibrated theannual mean for the dierent HRUs. As the RRF is related to land cover and slope (Sieber and Purkey2015), I used dierent RRF values for forest, pasture, agriculture and residential in the alluvial-colluvialzone (slope < 10%), and dierent values for forest and pasture in the four slope categories of the basaltzone (Table 3.10). The allowable range of RRF in the WEAP model is between 1 – 1,000 []. Surfaceruno will decrease with higher values of RRF. General trends of RRF indicate that it is higher for forestthan for pasture and agriculture (surface runo is typically lower in forest than in agriculture or pasture,Sieber and Purkey 2015). I tested a wide range of RRF values (Table 3.10), and found that lower RRFvalues represented the system best (similar to the WEAP default setting of 2).63Table 3.10: Annual mean of RunoResistance Factor (RRF). Seasonal variability was determined basedon the leaf area index (section 3.5.5).Hydrological Response Unit (HRU) RRF range RRF final[  ] [  ]Alluvial/colluvial zone – slope < 10%: forest (0.3, 300) 3Alluvial/colluvial zone – slope < 10%: pasture (0.2, 200) 2Alluvial/colluvial zone – slope < 10%: agriculture (0.1, 100) 1Alluvial/colluvial zone – slope < 10%: residential (0.1, 200) 2Basalt zone – slope < 10%: forest (1, 600) 2Basalt zone – slope 10 - 20%: forest (1, 600) 2Basalt zone – slope 20 - 30%: forest (1, 600) 1Basalt zone – slope 30 - 80%: forest (1, 600) 1Basalt zone – slope < 10%: pasture (1, 600) 1.5Basalt zone – slope 10 - 20%: pasture (1, 600) 1.5Basalt zone – slope2 0 - 30%: pasture (1, 600) 1Basalt zone – slope 30 - 80%: pasture (1, 600) 1For the soil depth parameter, I used soil depth values from the lithology logs for each SC of the alluvialzone (see section 3.5.5), while soil depth values for the colluvial and basalt zone were calibrated (Table3.11). Four dierent slope categories existed for the basalt zone, for which dierent soil depths wereexplored in calibration (generally assuming lower soil depth for steeper slopes).Table 3.11: Soil layer 1 (top layer) depth in millimeters.Hydrological Response Unit (HRU) Soil layer 1 depth range Soil layer 1 depth final[mm] [mm]Alluvial zone – slope < 10% based on lithology logs (1000, 10000)Colluvial zone – slope < 10% (500, 7200) 3,000Basalt zone – slope < 10% (500, 5000) 2,000Basalt zone – slope 10 - 20% (500, 5000) 1,000Basalt zone – slope 20 - 30% (500, 5000) 750Basalt zone – slope 30 - 80% (500, 5000) 700For the saturated hydraulic conductivity of soil layer 1, I used field measurements of infiltrationcapacities in the Potrero and Caimital watersheds (Garcia-Serrano 2015) as initial calibration values(high-end values in Table 3.12), but noted during calibration that much lower eective values wereneeded for the WEAP model. Final saturated hydraulic conductivity parameters were in a similar rangeas the WEAP default value (20 mmdy1), or as reported in other WEAP studies (Amato et al. 2006;Yates et al. 2009). For the eective water holding capacity (i.e., the dierence between volumetricwater content at field capacity and at the permanent wilting point) of soil layer 1, field measurementswere used (Appendix B).64Table 3.12: Saturated hydraulic conductivity of soil layer 1 in millimeters per day.Hydrological Response Unit (HRU) Soil layer 1 conductivity range Soil layer 1 conductivity final⇥mmdy1⇤ ⇥mmdy1⇤Alluvial zone: agriculture (0.0005, 101) 7Alluvial zone: forest (0.0005, 290) 9Alluvial zone: pasture (0.0005, 118) 7Alluvial zone: residential (0.0005, 118) 7Colluvial zone: agriculture (5, 629) 10Colluvial zone: forest (5, 3583) 12Colluvial zone: pasture (5, 146) 10Colluvial zone: residential (5, 146) 10Basalt zone: forest (10, 1163) 11Basalt zone: pasture (5, 1000) 10The Preferred Flow Direction (PFD) parameter was calibrated for each HRU classification (Table 3.13).The PFD is a partitioning coecient between horizontal and vertical flow in soil layer 1, and partitionsthe outflow of the top soil layer between interflow and downward percolation (see also Equation 3.2). Itis related to soil, land cover, and topography.Table 3.13: Preferred flow direction of soil layer 1.Hydrological Response Unit (HRU) PFD range PFD final[  ] [  ]Alluvial zone – slope < 10%: agriculture (0.5, 0.9) 0.9Alluvial zone – slope < 10%: forest (0.5, 0.9) 0.8Alluvial zone – slope < 10%: pasture (0.5, 0.9) 0.9Alluvial zone – slope < 10%: residential (0.5, 0.9) 0.9Colluvial zone – slope < 10%: agriculture (0.2, 0.9) 0.6Colluvial zone – slope < 10%: forest (0.2, 0.9) 0.6Colluvial zone – slope < 10%: pasture (0.2, 0.9) 0.6Colluvial zone – slope < 10%: residential (0.2, 0.9) 0.5Basalt zone – slope < 10%: forest (0.3, 0.9) 0.8Basalt zone – slope 10 - 20%: forest (0.3, 0.9) 0.8Basalt zone – slope 20 - 30%: forest (0.3, 0.9) 0.9Basalt zone – slope 30 - 80%: forest (0.3, 0.9) 0.9Basalt zone – slope < 10%: pasture (0.3, 0.9) 0.8Basalt zone – slope 10 - 20%: pasture (0.3, 0.9) 0.8Basalt zone – slope 20 - 30%: pasture (0.3, 0.9) 0.9Basalt zone – slope 30 - 80%: pasture (0.3, 0.9) 0.9Soil layer 2 was only used for the basalt zone in the hills. Initial estimates for the saturated deepconductivity were based on rates for permeable basalt (Domenico and Schwartz 1997), but a range ofvalues was explored during the calibration (Table 3.14).65Table 3.14: Parameters for soil layer 2 in basalt zone adjusted during calibration.Parameter Range FinalEective water holding capacity [mm] (250, 5000) 5000Saturated hydraulic conductivity [mmdy1] (10, 35) 34.6For the aquifer, I used values for the specific yield (0.002) and the saturated hydraulic conductivity(26 m dy1) as reported for the Potrero-Caimital aquifer by Losilla and Agudelo (2003), but alsoexplored a range of values (Table 3.15). Parameters related to surface water – groundwater interaction,including wetted depth in river, groundwater storage at equilibrium with the river, and horizontal distancefrom farthest edge of aquifer to river were adjusted during calibration (Table 3.15, see section 3.4 fordescription of these parameters).Overall, the WEAP model showed in particular sensitivities to soil parameters like the saturated soilhydraulic conductivity and the runo resistance factor. More detailed field data collection on soil datacould help improve the characterization of the soil dynamics in these watersheds, especially for hydraulicconductivities in the soil layer (as currently, only infiltration capacities were available fromGarcia-Serrano2015). The model was also sensitive to parameters relating to surface water - groundwater interaction,where more field investigation on these interaction in these watersheds could improve the capturing andmodelling of this process.Table 3.15: Groundwater related parameters adjusted during calibration.Parameter Range FinalSaturated hydraulic conductivity [mdy1] (1, 600) 26Specific yield [ - ] (0.002, 0.23) 0.002Wetted depth [m] (0.5, 1) 0.5Horizontal distance [m] (400, 4000) 400Storage volume at equilibrium with river at wetted depth [Mm3] (15, 40) 25The crop coecient (Kc) is often adjusted during calibration to account for the specific soil conceptual-ization in the WEAP model (Sieber and Purkey 2015; Yates et al. 2009). The WEAP model was under-estimating total annual ET for the three main land cover types of the studied watersheds (agriculture,forest and pasture) relative to measured values (from the EC monitoring station and MODIS ET) (seeTable 3.16). In the WEAP model, ET is calculated for each HRU based on the potential crop referenceevapotranspiration ETreƒ (meteorological conditions), the Kc factor of the land cover as well as availablesoil moisture which is modelled (see Equation 3.2). The Kc values however were calculated based solelyon the ratio of actual ET and ETreƒ . To account for the soil moisture availability as modelled throughWEAP and the soil conceptualization, I adjusted the Kc values during model calibration by multiplyingwith scaling factors (tested range between 1 to 1.7). A scaling factor was estimated for each land usetype through optimization, so that the error of the modelled annual evapotranspiration was minimal66in comparison to the empirical evapotranspiration measurements (measured for agriculture at the ECstation) and the MODIS evapotranspiration estimates (for forest and pasture) (Table 3.16), and so thatseasonal variability was represented (see Appendix B for figures).For forest land cover, Kc was multiplied by the scaling factor 1.7 (Table 3.16). While total annual ETmodelled by WEAP was still lower than MODIS ET (Table 3.16), this was mostly due to underestimationof ET during the dry season (see Appendix B). The same was found for pasture ET, where a scalingfactor at the higher end (i.e., 1.7) resulted in high overestimation of ET rates during the wet seasonwhile dry season ET stayed low (see Appendix B), and therefore, a mid-range scaling factor of 1.5 waschosen. For the agricultural crop, also a mid-range scaling factor of 1.6 was chosen. However, ETduring the dry season resulting from melon irrigation was not well captured considering that irrigationis applied in the WEAP model to a soil layer of up to 10 meters depth, in contrast to the ‘real’ top soil,where drip irrigation applies water directly to melon plants. Therefore, I used a scaling factor of 3 duringthe melon season, which resulted in good comparison of modelled and measured ET for agriculture.Table 3.16: Total annual evapotranspiration in millimeters per year for dierent Kc scaling factors f. Kcvalues were multiplied by scaling factors (f) to optimize evapotranspiration modelling. For agriculture,two scaling factors were used to represent irrigation during the dry season and rainfed conditions duringthe wet season. Final factors: agriculture (3 dry season, 1.6 wet season), pasture (1.5), forest (1.7).Agricultural ET was calculated as mean from 2015 to 2016 (years of measured data), and forest andpasture ET as mean from 2005 to 2014 (years of available MODIS ET).Land cover MODIS Measured f 1 f 1.3 f 1.5 f 1.6 f 1.7 fƒ nAgriculture n/a 944 578 720 785 821 1009 921Pasture 948 n/a 574 719 786 824 1022 786Forest 1258 n/a 618 760 825 861 895 8953.7 Model evaluationTo evaluated the parameter fit obtained in calibration and the overall model performance, I calculatedgoodness-of-fit measures between observed and modelled streamflow for the four stream monitoringsites (Zambrano-Bigiarini 2017a), including Nash-Sutclie Eciency (NSE) (Nash and Sutclie 1970),Pearson correlation coecient (r), Coecient of Determination (R2), Root Mean Square Error (RMSE)and RSR (the ratio of the Root Mean Square Error over the Standard Deviation) (Table 3.17). Thegoodness-of-fit measures indicate good agreement between observed and modelled streamflow formost sites, where r, R2 and NSE > 0.5 and RSR < 0.7 are considered satisfactory for hydrologicalmodel performance (Moriasi et al. 2007). The Downstream Caimital River site showed slightly loweragreement between modelled and observed streamflow. The goodness-of-fit measures are within therange of other WEAP model performances reported in the literature (e.g., Hunter et al. 2015; Thompsonet al. 2012; Voisin et al. 2013).67Time series plots of observed and modelled daily streamflow provide further information on model per-formance (Figures 3.15, 3.16, 3.17, and 3.18), as well as scatterplots between observed and modelledstreamflow (Figure 3.19). While the model was able to capture the general streamflow dynamics atthe four monitoring sites, some model limitations are evident. One of the main challenges was tocapture the flashy high streamflows following intense rainfall events with a daily model. Since rainfall-runo responses often occurred within an hourly time scale, the model results tended to underestimatethe high stormflows. Similarly, baseflow at the two upstream sites was overestimated by the model,also indicating that the ‘real’ system is flashier in these headwater catchments sites than could berepresented in the daily model. In some cases, the model also indicated a peak in streamflow whilethe observed data did not reflect a similar peak. Here it is important to remember that the observed datawas collected in watersheds that are influenced by human activities. For instance, unregistered surfacewater extraction by households may take place along the stream and could thus reduce an observedpeak in streamflow. Furthermore, rainfall was assumed uniform throughout the two watersheds (due tolimited available data). However, rainfall can be quite localized in these tropical watersheds, and somedisparities between modelled and observed streamflow might also be associated to the assumptionof uniform rainfall in the model. It is further important to recognize that rating curve uncertainties ofobserved streamflow can also lead to uncertainties in model assessment. Other limitations of the modelincluded that land use was assumed constant throughout the modelling period, and that soil dynamicswere conceptualized in a simplified manner in the model.Overall, the model had two main limitations; 1) Groundwater flow was not modelled explicitly, andthe model could therefore not be calibrated to groundwater level data (the decision to use simplifiedgroundwater modelling was mostly based on limited available groundwater data); and 2) the model hada daily time-step while the streamflow response was sub-daily. This limitation might have impacted inparticular the modelling of the relation between surface water and groundwater. The daily time stepreduced peak rainfall intensities, which likely led to lower runo and stormflow in streams in the modelthan observed. Soil infiltration capacities would have been exceeded faster in the real system, generatingmore infiltration-excess overland flow and higher stormflows in streams. Higher stormflows in streamsmight then have lead to higher focused recharge to the aquifer in the real system than could be capturedin the model. The modelling of the surface water - groundwater interaction was further limited by thesimplified groundwater modelling approach.68Table 3.17: Goodness-of-fit measures between observed and modelled daily mean streamflow data forthe four streamflow monitoring sites (site locations in Chapter 2, Figure 2.9).Objective function DownstreamPotrero (SW1)Upstream Potrero(SW3)UpstreamCaimital (SW4)DownstreamCaimital (SW5)Pearson correlation coecient(r)0.78 0.8 0.84 0.63Coecient of Determination(R2)0.61 0.64 0.7 0.40Nash-Sutclie Eciency (NSE) 0.61 0.59 0.64 0.39Root Mean Square Error(RMSE)⇥m3s1⇤ 1.4 0.32 0.1 2.46RSR (Root Mean Square Error/Standard Deviation)0.63 0.64 0.6 0.78Figure 3.15: Observed and modelled daily mean discharge in cubic meters per second at the Down-stream Potrero River site (SW1).69Figure 3.16: Observed and modelled daily mean discharge in cubic meters per second at the UpstreamPotrero River site (SW3).Figure 3.17: Observed and modelled daily mean discharge in cubic meters per second at the UpstreamCaimital River site (SW4).70Figure 3.18: Observed and modelled daily mean discharge in cubic meters per second at the Down-stream Caimital River site (SW5).Figure 3.19: Scatterplots between observed and modelled daily mean discharge in cubic meters persecond, for the Downstream Potrero River site (SW1), the Upstream Potrero River site (SW3), theUpstream Caimital River site (SW4), and the Downstream Caimital River site (SW5).To further assess the model performance, I also compared modelled evapotranspiration with MODIS ET(for forest and pasture) and with field measurements (for agriculture) (Figure 3.20). The WEAP modelcaptured the seasonal evapotranspiration dynamics well, especially for agricultural areas where I wasable to compare daily modelling results to actual daily measurements.71The comparison of modelled ET for pasture and forest to MODIS ET resulted in more variability (Figure3.20), especially during the dry season when the model underestimated ET in comparison to MODISET. However, MODIS ET is also not a direct measurement of evapotranspiration, but rather is basedon an algorithm that combines a range of remotely sensed data, meteorological datasets and assumedland use cover classifications (Mu et al., 2011), and might have overestimated dry season evapotran-spiration. For instance, field measurements of NDVI (Normalized Dierence Vegetation Index) at aneddy covariance station in tropical dry forest in northern Guanacaste also showed seasonal changeswith a deep drop of NDVI during senescence towards the end of the wet season (Castro et al. 2018). Onthe other hand, evapotranspiration is modelled in WEAP based on soil water availability in the top soillayer, while in the natural system, some tree species might also be able to access deeper soil water orgroundwater and thus keep transpiring longer into the dry season than WEAP is able to model. Further,rainfall was assumed uniform across the two watersheds in the model. However, spatial heterogeneityof rainfall in the real system might have led to higher localized rainfall, especially along the hillsideswith forest and pasture land cover. Higher localized rainfall in the real system may have then led to thehigher ET for these land cover types that were captured by MODIS ET. It is also important to rememberthat values compared here for both MODIS and WEAP model output represented means across manypixels (MODIS) and hydrological response units (WEAP) across the two watersheds, thus adding somevariability.72Figure 3.20: Comparison of modelled and measured (agriculture) and MODIS (forest, pasture) evapo-transpiration in millimeters per day.However, overall, the main seasonal dynamics of streamflow and evapotranspiration were capturedwell by the WEAP model, and hydrological model performance was thus considered satisfactory forthe data-limited Potrero and Caimital watersheds. While it is important to keep potential modellinguncertainties in mind (as is typical with hydrological models), the WEAP model will allow synthesizingcurrent socio-hydrological dynamics and exploring future change impacts, as will be discussed in thefollowing chapters.73Chapter 4Assessing currentsocio-ecohydrological vulnerabilitiesto drought to improve water security4.1 IntroductionDroughts and water security challenges aect communities around the world and are caused by acombination of social and hydrological drivers (Van Loon et al. 2016b). These drivers have to beassessed in a holistic framework when aiming to improve water security and inform drought adaptationstrategies.Water security at its most essential means that “every person has access to enough safe water ataordable cost to lead a clean, healthy and productive life, while ensuring that the natural environmentis protected and enhanced” (GWP 2000). This definition also highlights the importance of sustainablewater management and the “balance between resource protection and use” (GWP 2000). Further,recent research has argued that water security should also ensure social, cultural and political relationsbetween societies and their water resources (Wutich et al. 2017; Jepson et al. 2017). As such, watersecurity describes not only a “state of adequate water” but also a “hydrosocial process” that recognizesthe way that “water shapes human lives” and reflects the inter-relation of society and water at multiplescales from households to countries (Jepson et al. 2017). This is based on the concept of a hydrosocialcycle that includes the social aspects of water flows, in contrast to a purely hydrological cycle in whichwater flows are separated from society (Linton and Budds 2014). This is related to the theory ofsocio-hydrology which acknowledges that today most watersheds are dominated by human activitiesand therefore, both hydrologic and social processes and their inter-relations need to be considered inwatershed assessments (Sivapalan et al. 2012). Societal changes may lead to changes in hydrologyand water supplies, and conversely, changes in hydrology may induce societal changes and responses.These two components are considered a coupled and co-evolving socio-hydrological system (Sivapalan74et al. 2012). Water security, then, is formed as an “outcome” of this coupled socio-hydrological system,and as such, is also dynamic and evolves over time (Srinivasan et al. 2017).To improve water security, a first step is understanding how people use water and how that relates towater availability (Srinivasan et al. 2017). Water use is a generic term that can include consumptivewater use or water extraction (Gleick 2003). Here, water consumption specifically refers to water thatis removed from the system and is unavailable for further uses (Gleick 2003). Water extraction orwithdrawal refers to water that is removed (e.g. from an aquifer), some of which may return to thesystem while some is consumptive, and it also includes water that is lost in transfer pipes or is otherwiseunaccounted for (Gleick 2003; Sieber and Purkey 2015). Water demand describes the quantity ofwater that is desired by a user, but which is not necessarily fulfilled by water extraction (Gleick 2003).High water use has already led to depletion of water supplies in many regions of the world, and mightyet increase with population growth and socio-economic changes in the future, thus further increasingsocietal pressures on water resources and potential water security problems (Schewe et al. 2014; Wadaet al. 2014, 2011b).Water security problems often become more pronounced during times of drought when water availabilityis reduced. Hydrologic drought refers to a groundwater or surface water deficit and can be caused by acombination of natural causes (meteorological anomalies) and anthropogenic causes (water extraction)(Van Loon et al. 2016b, 2016a). Traditionally, a meteorological drought (rainfall deficit) is considered topropagate uni-directionally to a soil moisture drought (reduced soil moisture) and hydrological drought(reduced water availability in streams and groundwater), resulting in a climate-induced drought (VanLoon et al. 2016b). However, in human-modified watersheds, anthropogenic activities also impact waterflows and storages, and thus, can change the propagation of drought, or even be the sole cause for adrought (human-induced drought) (Van Loon et al. 2016b). Therefore, drought is not a merely externalnatural hazard but is linked with human influences and societal feedbacks (Van Loon et al. 2016b).The impacts or consequences of a drought hazard on society are determined by the exposure and thevulnerability of the society to this hazard (UN/ISDR 2009). Vulnerability is defined as the “characteristicsand circumstances of a community, system or asset that make it susceptible to the damaging eectsof a hazard” (UN/ISDR 2009). Thus, a drought hazard can lead to water insecurities if a society is notprepared and has a high vulnerability, or, on the other hand, the consequences can be less dramatic ifthe vulnerability of a society is low (Hoekstra et al. 2018). Similarly, water insecurities can arise even ifa drought hazard is low but the vulnerability of the society is high (Hoekstra et al. 2018).Vulnerability of the population and agriculture to drought is high in the seasonally-dry tropics of CentralAmerica due to poverty and livelihood structures that depend on local water resources, and the regionhas experienced water insecurities in recent years (Magrin et al. 2014; Vignola et al. 2018; FAO 2016;Madrigal-Ballestero and Naranjo 2015; Kuzdas et al. 2015b; WWAP 2012). In Costa Rica, about 21% of75the population lives in poverty (6% in extreme poverty), which is higher in rural areas (26%) than in urbanareas (18%) (INEC 2011; WWAP 2012). While most households have access to piped water systems,rural communities especially can experience high frequencies of water delivery interruptions during thelong dry season (Madrigal-Ballestero and Naranjo 2015). For instance, in a survey of households in sixrural communities in seasonally-dry Costa Rica, 18 to 60% of households per community indicated dailyinterruptions in water deliveries during the dry season and 6 to 54% indicated interruptions once per week(Madrigal-Ballestero and Naranjo 2015). Droughts also negatively impact agriculture and ranching, asfor instance during the 2014-2015 drought when estimated losses of more than 15$US million occurredin the agricultural and ranching sectors of the seasonally-dry tropics of Costa Rica (Vignola et al. 2018).In particular, subsistence smallholder farmers are vulnerable to droughts and were found to have a lowadaptive capacity to drought in this region (Holland et al. 2017).In the seasonally-dry tropics of Central America, drought is, on the one hand, driven by high climatevariability, where annual meteorological droughts (prolonged dry season) occur along with recurrent butless predictable multiannual droughts that have been related to El Niño and during which wet seasonrainfall is much reduced (Steyn et al. 2016). However, on the other hand, social aspects related to watermanagement and use also play an important role, as even in years of national drought emergencydeclarations, such as in 2014 and 2015, total annual rainfall was still high (for instance above 1500 mmand 1900 mm, respectively, in Nicoya, Costa Rica). Part of the water challenge is related to the temporalmismatch between water availability during the wet season and water needs during the dry seasonwhen typically no rainfall occurs for months. But increasing water demands of a growing population,tourism and agriculture have also increased pressure on water resources, resulting in water conflictsbetween dierent sectors (Esquivel-Hernández et al. 2017a; Ramírez-Cover 2008). Weak and inecientwater governance structures and related water infrastructure problems have further exacerbated watershortages (Ballestero et al. 2007; Esquivel-Hernández et al. 2017a; Madrigal-Ballestero and Naranjo2015; Kuzdas et al. 2015a). Technical and financial capacity challenges of, in particular, rural ASADAsoften lead to reduced adaptive capacity when responding to droughts, resulting in water insecurities forhouseholds in these communities (Madrigal-Ballestero and Naranjo 2015).To improve water security in the region and inform adaptation to droughts, it is necessary to assessdrivers that emerge from both the hydrological system (such as surface water and groundwater supplies)as well as the social system (focusing here on water management and water use by dierent sectors).Yet so far, such combined explorations of water resources and water use patterns are limited in theregion, likely also due to the scarcity of both hydrological and societal water use data. For instance,Kuzdas et al. (2015b) provided an overview from a water governance perspective, but highlighted theneed for better data on hydrology (water supply) and water use as it often was not clear how much wateris used when and by whom, especially in respect to agricultural and rural domestic water use.76The emerging field of socio-hydrology can provide a framework for a holistic system assessment ofcombined social (water use) and natural (hydrological) drivers with respect to drought. Yet, evapotran-spiration and natural water demands are often not explicitly included in this framework, even thoughthey constitute important water demands in most watersheds. Falkenmark and Folke (2002) first usedthe term of “socio-ecohydrology” that includes aspects of ecosystem hydrology and highlights the factthat humans not only manipulate water in streams and groundwater, but also aect evapotranspirationprocesses through land use management. Blending ecohydrology and socio-hydrology, the term socio-ecohydrology has only been rarely used in relation to the current socio-hydrology literature, such as ina study on evapotranspiration from rainwater harvesting ponds in India (Van Meter et al. 2016) and in adiscussion on urban water challenges (Pataki et al. 2011). However, the term has been mentioned in tworecent ecohydrology publications as a future research direction (Hamel et al. 2017; Wright et al. 2018),and it highlights the diverse water demands and supplies that exist in human-dominated watershedswhich play important roles in particular under drought conditions.In an eort to apply a socio-ecohydrological approach to understanding human and natural water de-mands, this research investigates socio-ecohydrological vulnerabilities to drought with the overarchinggoal to improve water security in seasonally-dry watersheds. While there are many inter-relationsbetween water and humans, this research focuses on societal water use as a first step towards watersecurity. Specifically, in this chapter, I aim to address my overarching research question 1 (see Section1.2), and approach this through three sub-questions.Research Question 1: What are the current socio-ecohydrological dynamics with respect to water use,and what vulnerabilities to drought emerge?• What is the current annual and seasonal domestic, agricultural and evapotransporative water use?• And, on the other hand, what are the key hydrological annual and seasonal dynamics in relationto water needs?• Further, what socio-ecohydrological vulnerabilities to drought emerge from this analysis, and whatkey management issues for drought adaptation and improvement to water security can be identi-fied?774.2 MethodsTo address the research questions for this chapter, I draw on the empirical data obtained in the field (asdescribed in Chapter 2) and support this by hydrological modelling over the historical period from 2005to 2016, as described in Chapter 3. To understand societal water use by dierent sectors, I make useof the water use data results from Chapter 3, and analyze these datasets further. I then synthesize thesocietal and hydrological processes in this chapter to assess for socio-ecohydrological vulnerabilities.4.3 Socio-ecohydrological dynamics4.3.1 Water use dynamicsDomestic water use and water securityWater use data analysis (Chapter 3) showed that domestic water consumption constituted 62% of totalAyA extraction water volumes for Nicoya in 2016, while 18% supplied municipal and business waterconsumption, and an estimated 20% of extracted volumes were unaccounted for or lost to pipelineleakage (Appendix C). In Hojancha, domestic water consumption comprized 60%, municipal/businesswater consumption 11% and unaccounted/lost water 30% of total AyA extraction volumes in 2016. Thedistribution of total extraction volumes to domestic and municipal/business water use and pipeline leak-age is associated with some uncertainties. Only total monthly extraction volumes from all groundwaterwells and the surface water treatment plant were available, and per-capita water use was estimatedbased on household/site water use rates available for a couple of years. While acknowledging theseuncertainties, estimated values indicated a high leakage rate and a potential opportunity for reductionof water extraction rates through fixing pipelines. Future field research could explore pipeline leakageas well as domestic and municipal/business water use rates in more detail to allow better assessmentin the model and for better targeting strategies to reduce total water extraction volumes. Further, whileestimated annual per person water use rates decreased from 80m3 to 61m3 in Nicoya from 2008 to2016 (Appendix C), artisanal (household) water pumping should also be considered on top of theseAyA-provided water use rates. Annual per person water use in Hojancha was slightly lower in 2016 thanin Nicoya (4.1), but similarly, artisanal pumping would likely need to be added.Artisanal household water consumption constituted 39% of total annual per person water consumption inthe ASADA (Table 4.1). High water use activities in a household (such as laundry, watering, and clean-ing) are typically supplied through artisanal pumping, explaining the high artisanal water consumption.While further research into this type of water use is required, this result highlights the importance ofartisanal water extraction rates. However, so far, these water extraction rates have been dismissed as78minor by water ocials and are often not included in water accounting and management activities in theregion.Table 4.1: Water consumption in cubic meters per person per year for ASADA-supplied (in 2014) andartisanal pumping, as well as for AyA-supplied water in Nicoya and Hojancha (for 2016).ASADA Artisanal Total ruralvillagesAyA Nicoya AyAHojanchaAnnual per person water consumption⇥m3⇤77 49 126 61 57% of total annual consumption per person 61 39 100The total annual water consumption per person (126m3) in rural villages is high in comparison tointernational estimates (Table 4.2), and thus indicates a potential for water conservation measures at thehousehold level. While the high annual water use does not seem to implicate any water security issuesin this particular village, these results are based on data from one of the best organized ASADAs in thewatersheds with good financial and personnel capacities. Further, the village’s groundwater pumpingwell is located in the centre of the Potrero-Caimital aquifer where groundwater levels are typically higher(Agudelo 2006), and water supply is thus more reliable. In contrast, in many other villages, watershortages and curtailments caused households to experience direct drought impacts (Vignola et al.2018). For instance, households in the rural village of La Virginia only had access to a few hoursof running water per day in the dry season of 2015, as the groundwater pumping well would run dryafterwards, and people had to rely on water storage containers for their water needs for the rest of the day(as reported by ASADA president and household owners). In the town of Curime, social inequities aroseduring the drought, as due to limited water availability, only households located close to the groundwaterpump were serviced with water, while low pressures in distributions pipelines led to limited water accessfor households located further away and at higher elevation (as reported by ASADA president). Madrigal-Ballestero and Naranjo (2015) also reported similar social inequities due to technical deficiencies andlow water pressures whereby water would reach households at a higher elevation with diculties.79Table 4.2: International comparison of annual domestic water use per person, for selected countriesfrom Gleick et al. (2014). No data were reported for Costa Rica in this database. Artisanal water usefor Nicoya and Hojancha is not known, but should likely be added to AyA-provided water.Country Annual domestic water use per person⇥m3⇤Nicaragua 31Guatemala 37Honduras 44India 46Germany 47Panama 60Greece 76Brazil 83France 91Mexico 102Argentina 176USA 193Canada 260This study (ASADA plus artisanal) 126This study (AyA plus artisanal?) 57 & 61 + artisanal water useWater extraction by sectorSectorial comparison of water extraction volumes from the Potrero and Caimital watersheds showedthat towns and villages accounted for the highest extraction volumes, which included domestic andmunicipal/business water consumptions as well as unaccounted or lost water (Table 4.3). Most extractedwater was transferred out of the watersheds to Nicoya, followed by Hojancha. Total water extraction vol-umes for ASADAs and artisanal pumping within the watersheds were much lower. Agricultural extractionvolumes for irrigation accounted for 28% of total water extraction volumes in 2016, and ranching for only1%. While agricultural irrigation volumes were therefore high, this double cropping system of rainfedupland rice and irrigated melons is considered one of the most water-sustainable cropping systems inthe region (Morillas et al. 2018). Nevertheless, Morillas et al. (2018) also showed that farmers might beover-irrigating, and that there could be some potential for water conservation in agriculture.Table 4.3: Annual water extraction volumes from the Potrero and Caimital watersheds in million cubicmeters and as percentage of total extraction volume for 2016.Nicoya Hojancha Rural TotaltownsAgriculture Ranches Other TotalMm3 1.67 0.38 0.38 2.43 1.14 0.05 0.39 4.01% 42 10 9 61 28 1 1080Water supply sourcesAnalysis of water supply sources showed that reliance on groundwater resources, in contrast to surfacewater resources, increased between 2005 and 2016 (Table 4.4). While Nicoya for instance supplied 50%(0.8 Mm3) of total water extraction volumes from the Potrero river in 2005, its surface water extractionvolumes decreased to 27% (0.56 Mm3) by 2016. Furthermore, reliance on the Potrero-Caimital aquiferas water supply source for the surrounding region has increased over this time period. While Nicoya andHojancha also supply their water demands from the Nicoya and Hojancha aquifers (located beneath therespective towns), their extraction volumes from the Potrero-Caimital aquifer increased (Table 4.4).Table 4.4: Percentages of water supply sources of total extraction volumes for each sector in 2005 and2016. Total PC watersheds = total extracted volumes from Potrero and Caimital watersheds and aquifer(not including Nicoya and Hojancha aquifer); PC aquifer = Potrero-Caimital aquifer; PC rivers = Potreroand Caimital rivers.Year Total PC watersheds Nicoya Hojancha ASADAs Agri-cultureRanchingPCaquiferPCriversNicoyaaquiferPCaquiferPotreroriverHojanchaaquiferPCaquiferPCaquiferPCaquiferPCaquiferPCrivers2005 77 23 11 39 50 13 87 100 100 62 382016 86 14 15 58 27 0 100 100 100 62 38Water extraction over timeI also analyzed changes in water extraction rates over time, and found that rates for all towns increasedby 33% from 2005 to 2016, and specifically, by 21% for Nicoya, 74% for Hojancha, and 64% for theASADAs (Figure 4.1). Seasonal peaks of extraction rates occurred during the dry season, driven mostlyby high agricultural extraction, which was concentrated within these months, but peaks of domesticextraction rates were also observed during the dry season (Figure 4.1).81Figure 4.1: Monthly total water extraction rates in million cubic meters per month for dierent sectorsbetween 2005 and 2016, indicating water supply sources for each sector . These extraction rates do notinclude water extraction from the Nicoya and Hojancha aquifers but focus on the Potrero and Caimitalwatersheds.Seasonality of water useNext, I explored the seasonal domestic water use dynamics in more detail. First, I focused on seasonalwater use in the ASADA Caimital, as I had the most detailed, household level data for this system.Based on the mean water use across households for each month from 2012 to 2015, I calculated thedaily household mean to minimize the eect of dierent days per month on total monthly water use, anddivided the time series into dry season (December to April) and wet season (May to November). I foundthat mean daily household water use was significantly higher during the dry season than during the wetseason, and that mean household water use decreased with the beginning of wet season rainfall (Figure4.2). Daily household water use in the dry season (847 liters) was on average 18% higher than duringthe wet season (720 liters). Daily water use rates for artisanal water use were also significantly higherduring the dry season than the wet season (Table 4.5; Figure 4.2a).82dry.mean wet.mean60070080090010002 4 6 8 10 12202224262830IndexCaim.rain.wetdry$Use● ● ●●●●●●●●●● 01002003004005002 4 6 8 10 12MonthDry season Wet seasonDaily Household Water Use [L day-1]607080901,00020Mean Household Water Use [m3  month-1]2242628300100200300400500Mean Total Rainfall [mm month-1]Water UseRainfall(a) (b)Figure 4.2: ASADA Caimital seasonal household water use from 2012 to 2015. (a) Daily householdwater use (mean across all households) from 2012 to 2015 for dry season (Dec – Apr) and wet season(May – Nov), and (b) Mean monthly household water use (from 2012 to 2015), and mean monthly totalrainfall (from 2012 to 2015).For water use in Nicoya and Hojancha, no household level water use rates were available, but only totalextraction rates and estimates of household numbers and population. Over the time series from 2005 to2016, water extraction volumes had significantly increased, and annual per person water use rates alsovaried. To minimize eects of these changes in this seasonal analysis, I used daily activity rates (i.e.,the percentage of water use per day relative to total annual personal water use). Water use in the dryseason was only slightly higher in Nicoya and Hojancha than during the wet season (Figure 4.3; Table4.5). Nicoya and Hojancha also used periodic water curtailments during the dry season to reduce waterconsumption, in contrast to the ASADA where no curtailments occurred.83dry wet0.10.20.30.40.5dry wet0.10.20.30.40.5dry wet0.10.20.30.40.5Daily Water Use Rate [% of Annual Total](a)Dry Season Wet Season Dry Season Wet Season Dry Season Wet Season(b) (c)0.10.20.30.40.5Figure 4.3: Daily water use rates as percentage of annual total for dry season (Dec – Apr) and wetseason (May – Nov), for (a) artisanal household well, (b) AyA Nicoya, and (c) AyA Hojancha.Table 4.5: Increase of water use in dry season (Dec – Apr) relative to wet season (May – Nov), forASADA based on mean daily household water use between 2012-2015, and for artisanal use and AyAsbased on daily activity levels.ASADA Artisanal AyA Nicoya AyA HojanchaDry season increase [%] 18 270 2 4While the dierences in domestic water use between wet and dry season were less pronounced forNicoya andHojancha, the household level data from the ASADA indicates an interesting socio-hydrologicalfeedback between the drier climate of the dry season and increased water use. Many high wateruse activities such as showers, laundry, watering and cleaning typically increase in the dry season,and if no water restriction and conservation measures are in place, this may lead to increased waterextraction. Thus, in already water-limited months, increased water extraction rates can cause human-induced propagation of a hydrological drought. I also attempted to compare dry year versus wet yearwater use in the historical record of annual water extraction rates. However, this was somewhat limitedby data availability with short time-series (especially for Caimital) as well as the increasing populationand number of households over the course of the year (for which no monthly data were available forNicoya and Hojancha). Yet, a high-level comparison of total annual rainfall and total extraction volumesfor the town of Nicoya showed that extraction rates were higher in recent low-rainfall (El Niño) yearsversus subsequent wetter years despite the increasing population, i.e. a drier 2012 (annual extractionof 2.04 Mm3) versus a wetter 2013 (1.95 Mm3), and drier 2014 (2.00 Mm3) and 2015 (2.05 Mm3)84versus a wetter 2016 (1.99 Mm3). This trend is however not as clear in earlier years and warrantsmore detailed research. While the higher water use during the dry season (or a drier year) would notnecessarily translate into higher water use in a drier climate under climate change, it is important toconsider this potential socio-hydrological feedback when exploring future scenarios. Many of the typicaldry season activities (such as, irrigation of household gardens, and higher laundry-washing, cleaningand showering) might increase, without water conservation eorts or behaviour changes, if dry seasonsbecome longer or the mid-summer drought extends as climate change projections indicate (Grossmannet al. 2018). A similar feedback between a drier climate and increased water extraction has oftenbeen associated with agriculture, with increased irrigation in a drier climate. For instance in India, aprecipitation deficit led to higher groundwater extraction for irrigation agriculture (Asoka et al. 2017).Ecohydrology - evapotranspirative water demandsIn addition to water extraction from surface water and groundwater (’blue water’) to supply irrigationagriculture and ranching, land use/land cover in the Potrero and Caimital watersheds also result in evap-otranspirative (’green water’) demands. According to Morillas et al. (2018), the total agricultural greenwater demand for rainfed rice was measured as 384 and 382 mm for 2015 and 2016, respectively, in alarge melon-rice farm of the Potrero and Caimital watersheds. Modelled total annual evapotranspirationranged from 741 to 1045 mm (forest), 679 to 879 mm (pasture) and 790 to 950 mm (agriculture) from2005 to 2016. The ratio of total annual evapotranspiration relative to rainfall ranged from 0.39 to 0.47per land use unit (Table 4.6). When taking the land use coverage of the watersheds into account,the high forest cover (52% of watersheds area) is reflected in the highest volume-based rainfall ratio,while agricultural evapotranspiration is relatively low due to lower agricultural land use coverage (8%)(Table 4.6). In drier years, evapotranspiration ratios are higher (ratio of 0.54 for year of lowest annualtotal rainfall between 2005 and 2016) in comparison to wetter years (ratio of 0.29 for year of highestannual total rainfall). Thus, the proportion of the remaining water that is available for societal use afterfulfilling evapotransporative demands is lower in drier years. This indicates the importance of consideringevapotranspirative water needs especially in drought years.Therefore, the incorporation of evapotranspirative (green) water demands into water resources manage-ment are important in particular in drought-prone areas. Society influences evapotranspirative waterdemands through land management decisions, as partitioning of rainfall into green water (soil mois-ture) and blue water in streams and groundwater is influenced by land cover, as are transpirationrates influenced by vegetation type (Falkenmark and Folke 2002; Falkenmark and Rockström 2006;Lathuillière et al. 2016). Socio-ecohydrological aspects are also important regarding agricultural watermanagement. While blue water demands of irrigation agriculture are traditionally managed for (suchas melon irrigation in the study watersheds), green water demands of rainfed agriculture (such as non-85paddy rice) are also important. Falkenmark and Rockström (2006) suggested for instance that greenwater can be conserved in rainfed agriculture by reducing (non-productive) evaporation and increasingproductive transpiration (plant growth) to improve agricultural productivity while lowering overall waterdemands. Further, if early rainfall in Guanacaste may become less reliable in the future under climatechange (Grossmann et al. 2018; Imbach et al. 2018), more farmers might start to increase their bluewater extraction rates for agricultural irrigation and rely less on rainfed (green water) supplies, indicatingthe importance of considering both in a socio-ecohydrological approach. While evapotranspiration ratesare high, evapotranspired water also plays an important role by contributing to vapour supply of theatmosphere and recycling rainfall over land (Ellison et al. 2012). Thus, extending socio-hydrologytowards socio-ecohydrology more explicitly calls for the inclusion of ecohydrological water demands(and ecosystem water needs) in socio-hydrological analysis.Table 4.6: Mean and range of the ratio of total annual evapotranspiration of dierent land use/landcover relative to total annual rainfall, given per land unit, and for the total of the Potrero and Caimitalwatersheds. The second value reflects the land use coverage in the watersheds. Mean and range arebased on modelled evapotranspiration from 2005 to 2016.Land Unit WatershedsForest Pasture Melons/rice Fallow/rice Forest Pasture AgricultureMean 0.44 0.39 0.47 0.41 0.22 0.14 0.03Range (0.32,0.59)(0.27,0.54)(0.31, 0.65) (0.27, 0.56) (0.17,0.30)(0.10,0.20)(0.02, 0.05)Modelled evapotranspiration was also characterized by high seasonality, due to high availability of waterduring the wet season and low water availability during the dry season (Figure 4.4). Evapotranspirationwas low during the dry season for forest, pasture and fallow rice fields, while in contrast, agriculturalirrigation of melon fields led to high evapotranspiration rates. Along with the start of wet season rainfall,green-up of vegetation and higher evapotranspiration began in April/May, increased to the highestevapotranspiration rates during vegetation maturity between June to October, and decreased again withdecreasing water availability starting in November (senescence).Similar seasonal cycles have been reported by Castro et al. (2018), who showed how seasonal rainfallpatterns influenced the timing of green-up, maturity and senescence of tropical dry forest in CostaRica, and importantly, how the drought of 2014 and 2015 led to changes in the phenological cycleof these forests. High seasonality of evapotranspiration rates was also found for pasture and reforestedecosystems in the wet-dry tropics of Panama (Wolf et al. 2011). In line with Castro et al. (2018), Esquivel-Hernández et al. (2017b) found that the ecohydrological resilience of the wet-dry tropical biomes islow (resilience defined by the authors as degree to which a catchment can return to its normal ecohy-drological functioning after a disturbance), potentially due to a combination of ecohydrological factors86(sensitivity of ecosystems to changes in rainfall seasonality) and societal factors (introduction of highwater demanding commercial teak plantations and high water extraction rates for agriculture).Figure 4.4: Mean total monthly modelled evapotranspiration (ET) and standard deviation between 2005and 2016 for dominant land use types in the Potrero and Caimital watersheds. Melons are irrigatedduring the dry season, explaining the peak in evapotranspiration for melon agriculture during this period,and are harvested between March and April, leading to a drop in evapotranspiration in April.4.3.2 Hydrological dynamicsDomestic and agricultural extraction in the Potrero and Caimital watersheds is supplied from the Potrero-Caimital aquifer and the Potrero and Caimital rivers and their tributaries. Total annual rainfall contributeddominantly to streamflow and evapotranspiration in these watersheds, while also recharging the aquifer(Table 4.7). Anthropogenic extraction also constituted an important component of the water balance.The remainder of the rainfall contributed to changes in soil moisture storage, and other minor flows.Importantly, surface water and groundwater interact, and groundwater contributed to streamflow viabaseflow, and surface water contributed to groundwater via focused recharge.87As is typical for the Guanacaste climate, high inter-annual rainfall variability characterized these water-sheds, and led consequently to high inter-annual variability for water flows (Table 4.7). The inter-annualrainfall variability in Guanacaste is mostly driven by the El Niño Southern Oscillation (ENSO) (Steynet al. 2016). Lower annual rainfall during El Niño years led to reduced evapotranspiration, streamflowand groundwater recharge (Table 4.7). On the other hand, increased rainfall during La Niña yearsshowed much higher water flows throughout the system. These impacts of ENSO on water flows in thePotrero and Caimital watersheds will be discussed in more detail in Chapter 5.Table 4.7: Annual mean and range of water flows in million cubic meters over modelled time series from2005 to 2016 for the Potrero and Caimital watersheds, as well as for years classified as El Niño (6 years)and La Niña (5 years) (classification based on Steyn et al. (2016); see Appendix D for details). ET =evapotranspiration; GW = groundwater.Rainfall ET Streamflow GW recharge Extraction Baseflow⇥Mm3⇤ ⇥Mm3⇤ ⇥Mm3⇤ ⇥Mm3⇤ ⇥Mm3⇤ ⇥Mm3⇤Mean 164 64 73 5 4 4Range (100, 247) (54, 73) (34, 124) (3, 8) (3, 4) (2, 6)El Niño Mean 128 61 48 4 4 2El Niño Range (100, 169) (54, 72) (34, 68) (3, 5) (3, 4) (2, 3)La Niña Mean 202 67 98 7 4 5La Niña Range (172, 247) (64, 73) (78, 124) (6, 8) (3, 4) (4, 6)GroundwaterObserved groundwater levels fluctuated seasonally (Figure 4.5). Groundwater recharge during the wetseason typically increased groundwater levels, and groundwater levels decreased again during the fol-lowing dry season due to anthropogenic extraction and baseflow contribution (Figure 4.5). The importantrole of seasonal groundwater recharge for water resources management is explored in Chapter 6, andbelow, results from the three groundwater level monitoring stations are discussed.Monitoring well GW1 was located in Varillal (see Chapter 2 for field site map and site description), andis used for artisanal household pumping. GW1 is a shallow groundwater well only five meters deepwithin the centre of the Potrero-Caimital aquifer, where groundwater levels are typically high (Agudelo2006), and it is in close proximity (3 m) of a small stream. Monitored water levels in this well wereinfluenced by daily pumping and surface water – groundwater interaction. While seasonal fluctuationswere less pronounced in this well and it never ran dry, a decrease in groundwater table depth occurredduring the dry seasons (Figure 4.5a). Monitoring well GW2 was located in Dulce Nombre, also closeto the centre of the Potrero-Caimital aquifer, and as it was not influenced by pumping for most of thetime, it provided a more robust representation of groundwater table fluctuations in the Potrero-Caimitalaquifer (Figure 4.5b). Monitoring well GW3 was located in Gamalotal, within the southwestern portionof the aquifer where the water table is typically deeper. This well was sometimes pumped, and due to88leakage of rainwater and ponding overland flow into the artisanal borehole, the measured water tabledepth was also influenced by occasional rainfall events (Figure 4.5c). While the observed water tabledepth was thus impacted by natural and anthropogenic factors, the well ran dry for many months duringeach dry season. This highlights the water challenges that the owner, a smallholder subsistence farmer,faced for meeting water needs during the dry season, similar to other smallholder farmers in the region.Both GW2 and GW3 showed a delayed rise of groundwater levels during the drought of 2014 and 2015,when groundwater levels stayed low until September/October, while in 2016, the higher early rainfallcontributed to an early rise of groundwater levels (Figure 4.5).Figure 4.5: Monitored daily mean groundwater table depth for (a) GW1 in Varillal, (b) GW2 in DulceNombre, and (c) GW3 in Gamalotal (see Figure 2.9 for locations). GW2 was dry during the dry seasonin 2016, and GW3 during all dry seasons (2014, 2015, 2016).89StreamflowHigh proportions of rainfall in these watersheds contributed to streamflow generation (Table 4.7). Theseasonality of streamflow was high, as shown in daily streamflow from the four monitoring stations(Figure 4.6). Baseflow during the dry season was typically low, and the upstream portions of the tworivers (SW3 and SW4) almost ran dry during this time. Even during the wet season, relatively lowbaseflow conditions typically persisted, which were only interrupted by high stormflows in response tointense rainfall events. These stormflows typically only lasted for a few hours, until baseflow conditionsresumed (Figure 4.7), indicating the flashy character of these watersheds. Flow duration curves for thelocation of the four monitoring stations show the seasonal variability in streamflow between low baseflowduring the dry season, and high stormflows in the wet season (Figure 4.8).90Figure 4.6: Daily mean streamflow (observed and modelled) at the Potrero and Caimital rivers and dailytotal rainfall from 2014 to 2016. See Figure 2.9 for monitoring site locations.91Figure 4.7: Examples of observed stormflow response to rainfall at the Potrero River Downstream site(SW1). Total rainfall andmean streamflow are shown in 30-minute intervals. Arrows indicate the durationof the rainfall and stormflow event.Figure 4.8: Seasonal flow duration curves based on modelled daily streamflow from 2014 to 2016 forSW3 (Potrero River Upstream), SW1 (Potrero River Downstream), SW4 (Caimital River Upstream), SW5(Caimital River Downstream), locations in Figure 2.9). Dry season = Dec - Apr; early wet season = May- Jun, mid-summer drought = Jul - Aug, late wet season = Sep - Nov.92The high seasonality of observed streamflow is typical for streams in the wet-dry tropics (Warfe et al.2011 (Australia); Ogden et al. 2013 (Panama); Recha et al. 2012 (Africa)). No comparable data werefound for the wet-dry tropics of Costa Rica. One of the only streamflow monitoring sites in the region,the humid tropical watershed of Sarapiqui in the mountains of the Costa Rican Cordillera, also showedseasonality in streamflow; however, seasonality was less pronounced than in the wet-dry tropics withlonger dry seasons (Birkel et al. 2012). Land use cover also plays an important role in runo generation,which is typically higher in non-forested (pasture, agriculture) watersheds (Ogden et al. 2013; Rechaet al. 2012; Birkel et al. 2012). Almost half of the Potrero and Caimital watersheds is characterized bynon-forested land use, likely contributing to high annual runo coecients (Table 4.8). Annual separatedruno coecients also showed the high contributions of rainfall events to stormflow, while baseflowswere much lower.Total annual runo coecients were similar to agricultural catchments in the tropics of Brazil, but muchhigher than observed in forested catchments (Dias et al. 2015). They were also higher than observed inforested watersheds in Honduras (Caballero et al. 2012), but lower than in the Sarapiqui watershed,which has a 66% forest cover, yet rainfall through the entire year (Birkel et al. 2012). The runocoecients were also higher than in wet-dry tropical watersheds in Kenya, where it was found thatthe runo coecient increased with deforestation and conversion to agriculture (Recha et al. 2012).Table 4.8: Mean and range of annual runo coecients (RC; ratio of total annual runo to total annualrainfall) of the Potrero and Caimital watersheds, based on modelled 12-year time series (2005-2016) fortotal streamflow (RC), baseflow (RCbƒ ) and stormflow (RCsƒ ).Caimital PotreroRC RCbƒ RCsƒ RC RCbƒ RCsƒMean 0.45 0.02 0.43 0.41 0.02 0.39Range 0.36 - 0.56 0.01 - 0.02 0.34 - 0.53 0.31 - 0.49 0.02 - 0.03 0.29 - 0.46These results indicate a key finding for water resourcesmanagement, as high streamflows show potentialfor increased use of surface water in the Potrero and Caimital watersheds. Currently, high volumes ofwater are ‘lost’ from the watersheds through outflows in streams, while downstream anthropogenic andecosystem water needs also have to be recognized and managed. Furthermore, high seasonality limitssurface water use to the wet season, as dry season streamflows are already low and do not oer anybuer for anthropogenic extraction (even though, at the moment, surface water extraction continuesthroughout the dry season for the town of Nicoya). Another challenge for surface water use is theflashiness and high stormflow runo. Extraction of surface water during stormflows (that constitute themajority of streamflow) can be challenging due to fast flows and high sediment loading.934.3.3 Emerging vulnerabilitiesComparison of total annual groundwater recharge and total annual extraction volumes from the Potrero-Caimital aquifer highlighted one of the main vulnerabilities of this socio-ecohydrological system. Duringyears with reduced rainfall, which occurred typically in El Niño years, groundwater recharge was muchreduced (Table 4.9). In these years, extraction volumes were higher than, or close to, the total annualgroundwater recharge (Table 4.9). As total annual groundwater recharge also contributes the baseflowsin streams, not all recharge is available for human extraction, and these are the years when manycommunity pumping wells were running dry and people faced water insecurity. Extraction rates beyondthe sustainable yield of the aquifer could also impact the long-term sustainability of groundwater use inthese watersheds. Currently, drier years are often followed by wetter years, which allow the groundwaterstorage to recover. However, increasing groundwater extraction rates due to population growth and apotentially drier climate under climate change could change these dynamics (potential future changeswill be explored in more detail in Chapter 5, and groundwater recharge will be explored in Chapter 6).Table 4.9: Annual total rainfall, groundwater recharge (for the Potrero-Caimital aquifer), total extractionrates from the Potrero-Caimital aquifer, as well as dierence between recharge and extraction in millioncubic meters for the modelled time series from 2005 to 2016. The ENSO classification is based on Steynet al. (2016); Appendix D). ENSO = El Niño Southern Oscillation, PC aquifer = Potrero-Caimital aquifer.Year ENSO Rainfall Rainfall GW recharge (PCaquifer)PC aquiferextractionGW Recharge -Extraction[mm]⇥Mm3⇤ ⇥Mm3⇤ ⇥Mm3⇤ ⇥Mm3⇤2009 El Niño 1,310 99.95 3.14 2.79 -0.532006 El Niño 1,384 105.58 3.27 2.57 -0.072014 El Niño 1,551 118.36 3.41 3.24 -0.622012 El Niño 1,623 123.83 3.83 3.40 -0.322015 El Niño 1,953 149.05 4.97 3.39 0.822013 La Niña 2,215 169.01 5.16 3.70 0.742011 La Niña 2,259 172.36 5.98 3.36 1.982016 La Niña 2,295 175.16 6.06 3.46 2.052005 ENSONeutral2,578 196.70 6.00 2.61 2.592008 La Niña 2,625 200.30 6.22 2.68 2.682007 La Niña 2,823 215.47 6.75 2.59 3.382010 La Niña 3,235 246.84 8.05 3.25 4.13Overall, the combined assessment of social and hydrological drivers in this chapter highlighted a numberof key vulnerabilities to drought in the Potrero and Caimital watersheds that can be used as basis todevelop and target adaptation strategies and improve water security, and can be informative for othersimilar watersheds in the region (Figure 4.9).94Figure 4.9: Socio-ecohydrological vulnerabilities to drought that emerged from the combined social(water use) and hydrological analysis.Population growth over the last decade has led to constantly increasing water extraction rates for townsand villages (Figure 4.1). Population growth projections for the next decades (INEC 2014), combinedwith climate change projections of a drier climate (Hidalgo et al. 2013), highlight the importance ofimplementing water management strategies that reduce the pressure on water resources and improvewater security under drought. Analysis of water use rates per person indicated that domestic wateruse was high by international standards. This demonstrates an opportunity for household level waterconservation measures. These could include, among others, household fixture changes such as lowflow showers and grey water reuse for outside water applications.The Potrero and Caimital watersheds act as water supply for many surrounding towns in the region, andout-of-watershed transfer via pipeline accounted for highest extraction rates. According to AyA waterocials, additional communities are starting to rely on the water supplies of the Potrero and Caimitalwatersheds during droughts, and water transfers out of the watersheds are continuously increasing. Ontop of high extraction volumes, these transfers also reduce the potential for water to be recycled within thewatersheds and contribute to groundwater recharge through infiltration. Estimates for water lost throughleakage, or otherwise unaccounted for, indicated a fairly high percentage of total extraction volumesfor Nicoya and Hojancha, and while further research into pipeline leakage is needed, there might bepotential for water conservation by fixing leakage in the distribution system. Limited data existed onrural water extraction volumes likely due to technical, personal and financial capacity challenges by95ASADAs. This highlights the need for improving water governance structures especially at the rurallevel, as was indicated by previous research in the Potrero and Caimital watersheds focusing on watergovernance (Kuzdas et al. 2015b; Madrigal-Ballestero and Naranjo 2015). Estimates of artisanal waterextraction volumes constituted 39% of total annual per person extraction volumes, indicating the need toincorporate this water use type in water management and to conduct further research on it. Agriculturalirrigation was another major water use in the watersheds, and as research by Morillas et al. (2018) hadshown a current over-irrigation for an agricultural farm of the Potrero and Caimital watersheds, theremight be also potential to reduce the agricultural water footprint .Currently, a temporal mismatch between water needs and supplies exists, as both agriculture anddomestic water users tend to have highest blue water demands during the dry season. I also identifieda potential socio-hydrological feedback between a higher water use and the drier climate conditions ofthe dry season.Both agriculture and domestic water users depend primarily on the single water supply of groundwater.As groundwater levels fluctuate seasonally and decrease towards the end of the dry season due toanthropogenic extraction and baseflow contribution, groundwater pumping wells often run dry towardsthe end of the dry season leading to water security concerns for communities. Groundwater rechargeduring the wet season is thus important for replenishing the aquifer. However, during El Niño years withlow rainfall, total annual groundwater extraction rates often exceeded total annual groundwater recharge.As surface water and groundwater resources are interacting, groundwater pumping can also negativelyimpact baseflow. This is a concern especially during the dry season when baseflows are already low.Considering high contributions of rainfall to streamflow, there is potential to diversify water resourcesmore and use more surface water during the wet season. Currently, large volumes of streamflow leavethe watersheds, and while their importance for downstream water users and ecosystems has to beconsidered, there may be potential to increase surface water use. However, high-frequency monitoringdata also showed the flashiness of streamflow with brief, but intense stormflow events and low baseflow,which can make surface water withdrawal and treatment challenging.Natural water demands through evapotranspiration constituted the second-highest water outflow fromthe watersheds after streamflow. This highlights the importance of integrating green water fluxes intowater resources management.964.4 ConclusionA combination of social and hydrological drivers has led to droughts and water security concerns in manywatersheds of the seasonally-dry tropics of Guanacaste, Costa Rica. A comprehensive assessmentof these drivers via a socio-ecohydrological approach can highlight key vulnerabilities to drought andhighlight foci to target adaptation strategies. Yet, both social water use and hydrological data arescarce in this region. This research brought together a range of diverse datasets obtained throughfield monitoring, work within communities and digitizing of hand-written records, as well as hydrologicalmodelling. The social part of this analysis focused on water use by dierent sectors, while there is alsoa range of water governance concerns which have been extensively researched for instance by Kuzdaset al. (2015b), Kuzdas (2014), Madrigal-Ballestero and Naranjo (2015), and Vignola et al. (2018).Key vulnerabilities to drought that emerged in this research included high domestic water use that tendedto increase during the dry season, as well as total annual anthropogenic groundwater extraction ratesthat exceeded groundwater recharge in low rainfall years. Further, artisanal domestic water use ishigh and while currently not receiving much attention in the region, it should be further researched andincluded in water management. Agricultural extraction accounted for almost 30% of total extraction vol-umes in 2016 and was concentrated during the water-limited dry season. While currently, communitiesand agriculture dependmostly on groundwater, there is potential to increase surface water use during thewet season, as high proportions of rainfall are ‘lost’ from the watersheds as streamflow. This combinedanalysis of societal water use dynamics and hydrological dynamics provided a first assessment ofthe seasonally-dry Potrero and Caimital watersheds, and can provide substantial information to watermanagers to improve water security under drought, and can be informational for similar watersheds inthe region.97Chapter 5Impacts of ENSO-driven climatevariability, future climate change andgrowing water demands on waterresources5.1 IntroductionThe seasonally-dry tropics in Costa Rica are already facing droughts and water shortages in the currentsystem, as discussed in the previous chapters, but these challenges may increase in the future withincreasing pressures on water resources by a growing population, intensifying agriculture and tourism(Imbach et al. 2017; INEC 2014; KC and Lutz 2017), combined with changes in rainfall amounts andseasonal cycles under climate change (Magrin et al. 2014). The climate has already been warming overthe last decades in Central America (Aguilar et al. 2005; Hidalgo et al. 2017), and is projected to risean additional 3º to 4ºC in Guanacaste by the end of the 21st century (Hidalgo et al. 2013; Karmalkaret al. 2011; Maurer et al. 2009). While general circulation models (GCMs), and statistical and dynamicaldownscaling of GCMs, have challenges in reproducing the complex climate of the region, they overallindicate that the climate will likely be drying in Guanacaste in the future (AlMutairi et al. 2018; Grossmannet al. 2018; Hidalgo et al. 2013; Imbach et al. 2018, 2012; Karmalkar et al. 2011; Maurer et al. 2009;Rauscher et al. 2008).However, the translation of these climate change projections of temperature and rainfall into implicationsfor water resources is limited, and projections have been applied mainly on a regional level across all ofCentral America. For instance, Hidalgo et al. (2013) modelled impacts of climate change scenarioson runo for the end of the 21st century using the Variable Infiltration Capacity (VIC) macro-scalehydrological model (spatial resolution of 0.5° x 0.5°) for Central America, and found a runo reductionof ~ 10% for southern Central America (the wider region to which Guanacaste belongs). Imbach et al.98(2012) used a vegetation model (Mapped Atmosphere Plant Soil System, MAPSS) to assess climatechange impacts on vegetation states and runo for the end of the 21st century, and also found a runoreduction across Central America, with at least a ~20% reduction in runo for Guanacaste. Maureret al. (2009) modelled climate change impacts on the Rio Lempa basin in El Salvador, Honduras, andGuatemala, also using the VIC model, and found that a 10% reduction in annual precipitation caused a24% reduction in annual average streamflow.While regional studies provide important information for general runo trends under climate change, de-cisionmakers needmore detailed information on a watershed level to develop climate change adaptationstrategies. Furthermore, the regional models of Central America typically do not address the specifichydro-climate of Guanacaste, and generally do not include any streamflow stations from Guanacaste.For instance, due to data scarcity Hidalgo et al. (2013) only included one streamflow monitoring stationfromGuanacaste in the development of their Central American hydrological model, and found insucientmodel performance for this station and thus did not include it in their final runo results. Jiménez-Rodríguez et al. (2015) presented one of the only studies on a watershed level. They investigatedpotential climate change impacts on runo in two mountainous headwater catchments in the easternpart of Guanacaste within the volcanoes of the Central American cordillera, using the two hydrologicalmodels HYLUC (Hydrological Land Use Change) and NAM (Nedbor Afstromnings Model). They didnot apply GCM-based climate change scenarios, but rather a generic rainfall reduction factor of 10%and a temperature increase factor of 2°C to simulate climate conditions in 2050. They found annualdecreases in discharge between 15 and 20%. To my knowledge, this study provides the only researchon climate change impacts on streamflow on a watershed scale within Guanacaste. However, it focusedon forested headwater catchments within the mountainous volcanoes where few people live, while muchof Guanacaste is characterized bymixed land use lowlands and surface watersheds that exchange waterwith underlying alluvial aquifers.Communities in Guanacaste depend by almost 80% on groundwater to meet their water demand (Guz-man 2015), as do many communities throughout the seasonally-dry tropics of Central America (Balles-tero et al. 2007). Yet, to my knowledge, only a few studies have explored the impact of climate changeon groundwater supplies in Central America. Garcia-Serrano (2015) applied a 12% reduction in rainfallto represent climate conditions in 2035, and used this to empirically estimate groundwater recharge asinput for a hydrogeological model.Thus, only a limited number of studies in Central America and specifically in Guanacaste exist on climatechange impacts on surface water and groundwater resources. This may be due, among other causes,to the sparse hydrological data in the region that can be used for hydrological modelling and the largeuncertainties that are often associated with the data that are available (Westerberg 2011). The limitedknowledge on potential impacts of climate change on water supplies makes it dicult for water managers99to develop climate change adaptation strategies, and more climate impact studies are needed to guidedecision making.Yet, future change is not only driven by changes in climate, but also by the drivers of population growthand increasing water demand (Vörösmarty et al. 2000; Wada et al. 2017; Wada and Bierkens 2014).Analyses of future water resources need to consider these multiple, interacting drivers. On a globallevel, Wada and Bierkens (2014) showed that domestic water demand alone might more than doubleby the end of the 21st century assuming a business-as-usual water use scenario. Most countries inCentral America will likely see further population growth towards the end of the 21st century, with atotal increase by 2080 between 1 and 145% (depending on type of population scenario) relative to2010, with a ‘middle road’ scenario increase of 54% (KC and Lutz 2017; Riahi et al. 2017). For CostaRica, population scenarios suggest an increase between 23 and 80% by 2080, with a 49% increase inpopulation for a ‘middle road’ scenario (KC and Lutz 2017; Riahi et al. 2017). The National Statisticsand Census Institute of Costa Rica estimates a 33% population increase by 2050 relative to 2011 for allof Costa Rica, and an increase of 23% by 2025 for Guanacaste (INEC 2014). If no substantial changesin water demand management were to occur, this population growth will be accompanied by significantincreases in water demand, which has to be accounted for in any future change impact modelling.Furthermore, communities in the region draw from both surface water and groundwater supplies. Assurface water and groundwater interact, they should be managed as one single water resource (Winteret al. 1998), and both need to be considered in climate impact studies. Yet, so far, most climate impactstudies in the region have focused on surface water alone.Most watersheds in Guanacaste are influenced by humans. Rural towns, cities, agriculture and tourismextract water and interact with the hydrological system in many ways (socio-hydrological systems, Siva-palan et al. 2012). Therefore, changes in the climate and hydrological system will also aect the socialsystem, and vice versa. While interactions between social and hydrological systems are complex andcan includemany dierent aspects, such as changes in human behaviour and societal norms and values,one important aspect is the potential change in domestic water demand in response to changes inclimate. In Chapter 4, I showed that current water demand in drier months can be ~18% greater incomparison to months with higher rainfall. This, in turn, leads to the question if a drier future climatemay increase domestic water demand (in case there are no pro-active water conservation schemes)and how that would impact water resources. On the other hand, the implementation of eective waterconservation eorts by society could also lead to decreased per-capita water use in comparison to thecurrent system.Ideally, drought adaptation strategies to future change impacts should include a set of short-term andlong-term strategies. Many adaptation strategies take time to implement, and to be eective in futuredecades with a drying climate, planning may have to start within the next decade. For this, long-term100climate change impact studies, such as until the year 2100, are useful. Furthermore, future climatechange projection can also explore ‘worst-case’ scenarios following high emissions scenarios, whichallow gauging the outer boundaries of potential impacts for communities and ecosystems.However, decision makers also require information on potential climate impacts within the next decade,and many water managers operate on these shorter timescales. Yet, current climate change trends aredicult to distinguish for a region such asGuanacaste that is characterized by high inter-annual variabilityin rainfall and where current rainfall variability is mostly driven by variability in the ocean-atmosphericinteraction ENSO (Steyn et al. 2016). Thus, exploring implications of dierent expressions of El Niñoand La Niña on water resources can provide important information for water managers for short-termplanning and preparation.The overall goal of this research is to investigate the impacts of the multiple drivers of climate variabil-ity/change and growing water demands onwater resources in seasonally-dry watersheds in Guanacaste,Costa Rica. Specifically, in this chapter, I address my second overarching research question (seeSection 1.2), and approach it via three sub-questions.Research Question 2: What are the impacts of current annual climate variability (i.e., El Niño), futureclimate change and growing water demands on streamflow and groundwater, and howdo those impacts change if socio-hydrological feedback between climate and water useis considered?• How does inter-annual climate variability driven by ENSO impact water resources of the presentand near future (~next decade)?• What are the implications of end of the 21st century climate change scenarios on streamflow andgroundwater recharge?• What eects do the multiple drivers of climate change and a growing population have on waterresources?1015.2 Methods5.2.1 Climate context of GuanacasteThe hydrological, geological, land use and water use context of the study site, the Potrero and Caimitalwatersheds, are described in Chapter 2. This section provides additional details on the climate inGuanacaste as background information for the climate scenarios that were developed in this research.Rainfall seasonalityCosta Rica lies between the Pacific and the Atlantic Oceans, and drivers of various geographic scalesinfluence its climate (Waylen et al. 1998). One of these drivers are the Northeast Trade Winds thataccount for moisture convergence in the tropics and convective activity near the Inter-tropical Conver-gence Zone (ITCZ) (Amador 1998). Imbedded in the Northeast Trade Winds is the Caribbean Low LevelJet, a region of about 500 km in width (north to south) with strong zonal winds, which extends from theCaribbean to the Lesser Antilles (Hidalgo et al. 2015a). The Caribbean Low Level Jet dominates theclimate in Costa Rica from December to March and causes intense rainfall due to orographic lifting in theEast of the high mountains of the Central American Cordillera. In contrast, Guanacaste in NorthwesternCosta Rica lies within the rain shadow and has dry and warm subsidence climate conditions when thetrade winds are predominant (Waylen et al. 1998). The trade winds tend to jet through local passesin the Cordillera forming gap winds, and when the main gap winds along the northern border of CostaRica reach the Pacific Ocean, they redirect oshore winds and cause upwelling of cooler water alongthe coast (the Costa Rica Dome) (Waylen et al. 1998). This increases atmospheric stability and reducesrainfall in Guanacaste from December to March (Waylen et al. 1998).The northward migration of the ITCZ in April interrupts atmospheric stability and dryness in Guanacaste,and marks the start of the wet season. The northern position of the ITCZ in the eastern equatorial Pacificreaches to about 10°N, and thus, lies within Guanacaste, extending from 9.5° - 11.5°N. The ITCZ bringsatmospheric instability and heavy convective storm events. In July and August, a reduction in rainfalltypically occurs in Guanacaste (mid-summer drought or “Veranillos de San Juan”). One explanation forthe occurrence of this mid-summer drought is that the Caribbean Low Level Jet strengthens temporarily,causing convection and a rainfall maximum in the Caribbean, and rain shadow eects (Amador 1998;Hidalgo et al. 2015a; Waylen et al. 1998). During this period, the ITCZ might also move southwardin the Pacific due to the Caribbean Low Level Jet (Hidalgo et al. 2015a). However, the mechanismsresponsible for the generation of the mid-summer drought in Guanacaste are complex and still debatedin the literature.102From September to November, tropical cyclones enter the Caribbean, interrupt the dominance of thetrade winds, and reverse the pressure gradient between the Atlantic and Pacific Oceans (Waylen et al.1998). The maximum tropical storm activity in the tropical North Atlantic enhances seasonal west windsfrom the Pacific, which bring moisture to Guanacaste (Waylen et al. 1998). Thus, the second part of thewet season from September to November (late wet season) is typically wetter than the first part fromApril to June (early wet season) (Steyn et al. 2016).Rainfall inter-annual variabilityThe gradient in sea surface temperature and atmospheric pressure between the equatorial Pacific andthe tropical North Atlantic Ocean determines the intensity of the trade winds, and thus, of the rainfallin Guanacaste (Waylen and Quesada 2002). Anomalies of sea surface temperatures and atmosphericpressures lead to inter-annual variability of rainfall in Guanacaste.During a positive sea surface temperature anomaly in the eastern Pacific (i.e., El Niño), the gradientbetween the colder equatorial North Atlantic and the warmer Pacific increases and the trade windsintensify. A negative (cold) temperature anomaly in the Atlantic enhances this eect (Waylen andQuesada 2002). Warmer than normal waters in the central Pacific can also displace the ITCZ in theeastern Pacific to the south and west of its mean seasonal position, which further strengthens thenortheastern trade winds. Stronger trade winds during an El Niño increase the rain shadow eect ofthe Cordillera and cause a reduction in rainfall and a drier mid-summer drought in Guanacaste (Waylenet al. 1998). Strong gap winds and atmospheric stability of the Costa Rica Dome can further intensify themid-summer drought (Waylen et al., 1998). Often, these conditions persist into September and October,potentially due to fewer tropical cyclones in the North Atlantic during years with a warmer Pacific andmore intense trade winds. Under normal (non-El Niño or La Niña) conditions, North Atlantic cyclonesinterrupt the mid-summer drought with the dominance of the trade winds by September and October.ENSO is a major control on the rainfall regime in Guanacaste (Aguilar et al. 2005; Altman et al. 2018;Steyn et al. 2016; Waylen et al. 1998; Waylen and Quesada 2002), and rainfall is significantly lowerduring an El Niño (Altman et al. 2018; Steyn et al. 2016). This happened for instance during the El Niñofrom 2014 to 2016, which was one of the three strongest El Niños on record and annual total rainfallwas ~46% lower than average across the Central American Pacific coast (Sánchez-Murillo et al. 2016).Climate change - Historical trendsOver the last decades, air temperatures in Central America have increased (Aguilar et al. 2005; Hidalgoet al. 2017). Aguilar et al. (2005) analyzed data from 48 temperature stations in Central Americaand northern South America over the period 1961 to 2003; however, the analysis did not include any103meteorological stations in Guanacaste. Temperature trends showed general spatial coherence betweenstations, and both mean temperatures and daily high temperature extremes increased (Aguilar et al.2005). Hidalgo et al. (2017) analyzed temperature data from 1970 to 1999 for Central America, andfound warming trends for most of Central America (for Guanacaste between 0.02° and 0.04°C per year).Total annual rainfall averaged over 105 precipitation stations in Central America and northern SouthAmerica did not have a significant trend over the period from 1961 to 2003 (Aguilar et al. 2005). Manystations had a positive trend, while some stations in northern Mexico and along the southwestern sideof the Central American isthmus had negative trends (Aguilar et al. 2005). Daily precipitation extremesshowed positive trends for most stations and also a significant positive trend for the region-wide timeseries (Aguilar et al. 2005). Similarly, Hidalgo et al. (2017) found no statistically significant annual trendsof total precipitation over the period 1970 to 1999 for Central America for most regions in Central America.For Guanacaste, Altman et al. (2018) also found no overall statistically significant trend of total annualrainfall.Climate Change – Future projectionsStudies on climate change projections indicate increases in air temperature over the next decades (Hi-dalgo et al. 2013; Imbach et al. 2018; Karmalkar et al. 2011; Maurer et al. 2009). For instance, Karmalkaret al. (2011) used dynamical downscaling via the regional circulation model (RCM) PRECIS (ProvidingRegional Climates for Impacts Studies) for Central America, and found a temperature increase of 4.3°Cfor the wet season and 3.8°C for the dry season by the end of the century (2070-2100) for westernCosta Rica and Nicaragua. Hidalgo et al. (2013) used statistical downscaling (Bias Correction andSpatial Downscaling, BCSD) and found temperature increases of up to 3°C for Central America northof 10°N, and 4°C for regions south of 10°N for the period of 2050-2099. Maurer et al. (2009) bias-corrected and statistically downscaled 16 GCMs for Central America, and found a temperature increaseof 2° to 4°C for Central America (~3° to 4°C for Guanacaste) for the period from 2070 to 2099. A recentstudy by Imbach et al. (2018) used dynamical downscaling with the high resolution (8-km) Eta RCM toanalyze climate change projections for the period 2021 to 2050, and found a temperature increase forthe region of Guanacaste of ~1.6° to 2°C for the 2021-2050 period in comparison to the baseline period(1961-1990).Most climate change studies for Central America have projected a drying pattern, however with largeuncertainties (AlMutairi et al. 2018; Grossmann et al. 2018; Karmalkar et al. 2011; Hidalgo et al. 2013;Neelin et al. 2006; Rauscher et al. 2008). Using PRECIS, Karmalkar et al. (2011) projected decreasesin rainfall for the end of the 21st century. A southward shift of the ITCZ resulted in a decrease in wetseason rainfall along the Pacific coast (between 24 and 48% for Guanacaste), as well as a decreasein dry season rainfall. Hidalgo et al. (2013) projected rainfall reductions of 5 to 10% for the 2050-1042099 period across Central America, a southward movement of the ITCZ, and a stronger mid-summerdrought. Similarly, Rauscher et al. (2008) also projected an overall decrease in annual rainfall based onGCM simulations (no downscaling), with rainfall reductions mostly during the first part of the wet seasonand during the mid-summer drought. Some of these changes in rainfall patterns may be due to an ElNiño-like warming of the tropical eastern Pacific (Rauscher et al. 2008). They also projected a gradientof increasing rainfall from north to south, with the largest reduction in rainfall between Guatemala tonorthwestern Nicaragua, and a shift from decrease in rainfall to increase in rainfall at about 7°N, justsouth of Guanacaste. Imbach et al. (2012) also indicated that the decrease in rainfall is more certain forthe north of Central America than for Costa Rica and Panama, which are in the transition zone betweenareas of decreasing rainfall to the north, and areas of increasing rainfall to the south. Imbach et al. (2018)projected decreases in rainfall for the Guanacaste region for the 2021-2050 period, especially duringthe mid-summer drought (reduction between 0.5 and 1 mm day1). Imbach et al. (2018) also assessedpotential changes in rainfall extremes, but Guanacaste did not show a consistent trend. Instead, changesranged between an increase in positive rainfall extremes in southern Guanacaste and a decrease innorthern Guanacaste.To date, studies have highlighted challenges of GCMs to reproduce the complex Central Americanclimate with its small-scale variabilities due to local topography and the combined influence of Pacificand Atlantic atmospheric and oceanic circulations. Most GCMs underestimated historical rainfall forthe 20th century (by up to 60%) when compared to observed data, although they captured the generalannual rainfall cycle with wet and dry seasons, and most models also captured the mid-summer drought(Rauscher et al. 2008). For Guanacaste, Rauscher et al. (2008) showed a historical rainfall underesti-mation between 20 to 40%. While Karmalkar et al. (2011) used an RCM to avoid some bias of GCMsdue to their low spatial resolution, they still encountered rainfall bias, with rainfall underestimation duringthe wet season. This may also be due to the severe rainfall underestimation by the GCM that drives thePRECIS RCM. Similarly, Imbach et al. (2018) also encountered rainfall underestimation when using ahigh resolution RCM.In contrast to the above studies that focused on all of Central America, AlMutairi et al. (2018) tested theskill of a selection of CMIP5 GCMs to reproduce the local climate of Guanacaste. They used a statisticalmethod and fit a Gaussian mixture model, which had been developed specifically for the Guanacasteclimate (Steyn et al. 2016), to observed and GCM monthly mean rainfall from 1979 to 2005 for the widerGuanacaste region (9°-15°N and 85°-86°W). They then assessed GCM results based on critical featuresof the Guanacaste rainfall regime (early and late wet season rainfall maxima, rainfall minimum duringthe mid-summer drought, and seasonal length and transitions times). Only six GCMs reproduced theclimate in Guanacaste, but rainfall variability was still high between these six GCMs.105Given the challenges of GCMs, and of statistical and dynamical downscaling, to reproduce the historicalclimate in Guanacaste, Grossmann et al. (2018) used a systematic expert elicitation protocol to assessGCM shortcomings and develop a range of future rainfall scenarios. According to this analysis, thechallenge of most GCMs to reproduce the current bimodal rainfall regime of Guanacaste and the frequentrainfall underestimation could be due to an overestimation of the northeast trade winds in the early wetseason, which would suppress early rainfall in the GCMs (Grossmann et al. 2018). The expert elicitationalso highlighted challenges of future GCM projections. For instance, the spatial and temporal changesin the gradient of sea surface temperatures between the Atlantic and the Pacific vary between GCMs.This gradient determines the strength of the northeast trade winds, and therefore the rain shadow eectand amount of drying in Guanacaste. Timing of the early wet season rainfall also is not well resolved inthe GCMs, and potentially, early rainfall could be delayed or weakened. Further, Guanacaste lies withinthe transition zone between the wet tropical region to the south (which will become wetter under climatechange) and the seasonally-dry region to the north (which will become drier) (Grossmann et al. 2018).As the transition zone does not resolve well in the GCMs, there are uncertainties if Guanacaste will bewithin the wetter or drier zone in future.Grossmann et al. (2018) combined the results of the expert elicitation with the GCM model assessmentby AlMutairi et al. (2018) to describe a range of possible future rainfall scenarios. These scenariosincluded a reduction of rainfall and an intensification of the mid-summer drought, a delayed start ofthe early rainfall, as well as one scenario that explored the possibility that, as Guanacaste lies in thetransition zone between wetter and drier tropics, it might experience an increase in rainfall in the future. Iused these climate change scenarios based on expert elicitation for my climate change impact analysisdescribed in the following sections.5.2.2 Overview of scenarios and modellingTo address my research questions and explore climate and water use impacts on water resourcesin seasonally-dry tropical watersheds, I applied a range of climate and water use scenarios to thehydrological model of the Potrero and Caimital watersheds. I described details of the Potrero andCaimital watershed model in Chapter 3. This section includes a brief overview of the scenarios andoverall method, and more details follow in the next sections.To address my first research question on how inter-annual climate variability driven by ENSO impactswater resources, I applied five climate scenarios ranging from an extreme La Niña to an extreme El Niño(in the current climate) to the hydrological model (Figure 5.1). The rainfall scenarios were based on amonthly rainfall model driven by ENSO indices developed specifically for Guanacaste (Steyn et al. 2016)and describe the monthly total rainfall over the annual cycle for each of the five ENSO expressions. As106climate change trends within Guanacaste are not yet discernible, these ENSO scenarios are likely stillapplicable for the near future (~next decade) (Altman et al. 2018; Steyn et al. 2016).Next, to address my second research question regarding future climate change impacts on streamflowand groundwater recharge, I applied four dierent climate change scenarios to the hydrological model,three of which were developed for Guanacaste (Grossmann et al. 2018), and one was based on GCMoutputs for the region (following AlMutairi et al. 2018). These climate scenarios describe monthly climateconditions over the annual cycle, as a mean of the period from 2074 – 2100. The period from 2074 to2100 was chosen based on the analysis by AlMutairi et al. (2018), who had analyzed the Guanacasteclimate for this period, and further, this period oered the possibility to explore potential ‘worst-case’climate change impacts. To compare climate change impacts to current conditions, I used a baselinescenario that described mean monthly climate conditions over the annual cycle of the historical periodfrom 1980 to 2016.For my third research question regarding the multiple eects of climate change and population growth,I compared impacts on water resources between the four climate change scenarios in combination witha low and a high population scenario. For this, I first assumed a business-as-usual water demand perperson (i.e., water demand based on historical data), and then also explored potential socio-hydrologicalfeedbacks between climate and increased/reduced water demand. The focus of the scenarios was onchanges in climate and domestic water demand. Recent land use change within the Potrero and Caimitalwatersheds has been minimal, aecting less than 4% of total area from 2005 to 2016 (see Chapter 3).Therefore, I did not include any land use change scenarios in this analysis while acknowledging that landuse and land cover changes can significantly impact evapotranspiration and rainfall-runo processes,and could be explored in future research.Both ENSO and climate change scenarios provided mean monthly values of rainfall and other climatevariables over the annual cycle. However, the WEAPmodel of the Potrero and Caimital watersheds runsat a daily time step. Therefore, I used a weather generator (LARS-WG; Semenov and Barrow 2002) togenerate a daily time series for each scenario for hydrological model input. For the generation of dailyscenarios through the weather generator, daily historical meteorological data are needed. For this, I useddaily historical meteorological data from 1980 to 2016. To develop the climate scenarios, I then applied amonthly change according to the dierence between the historical monthly mean and the monthly meanof the climate scenario. I then generated daily time series of rainfall and other meteorological variablesfor each of the climate scenarios as input into the hydrological model.For each ENSO scenario, I generated a daily time series over two years (for which themonthly mean overthe two years would be similar to the monthly value of the scenario) and applied these to the hydrologicalmodel. As two years are relatively short to capture variability in daily weather and potential impacts for107instance on rainfall-runo generation, I created six dierent two year time series for each of the ENSOscenarios, and applied each of them to a new model run.For each of the climate change scenarios, I created a daily time series over 26 years (similar to the periodfrom 2074 – 2100). The length of this daily time series allows capturing variability in daily weather, aswell as potential long-term impacts such as groundwater level declines over many years. To comparethe climate change scenarios to the historical climate, I also created a daily baseline scenario, also over26 years, using the same weather generator to minimize eects of the weather generator model onclimate comparisons. I applied each of the daily climate scenarios to the hydrological model, and wroteVisual Basic scripts to automatically export results from the WEAP model, and R scripts to automatepost-processing of results. More details on the methods follow in the next sections.Figure 5.1: Overview of climate and water use impact modelling approach. GCM = General circulationmodels.1085.2.3 Baseline dataAs daily time series input for the weather generator, I used a daily composite dataset from 1980 to 2016based on observed data from Nicoya in Guanacaste, the Eddy Covariance (EC) monitoring station data(see Chapter 2), and the Climate Forecast System Reanalysis (CFSR) dataset (Dee et al. 2014; Sahaet al. 2010). For rainfall, the most sensitive and spatially-variable parameter, I had access to a full recordof daily observed rainfall data: a composite dataset from 1980 to 2014 from the Instituto MeteorólogicoNacional (IMN) Nicoya station, and from 2014 to 2016 from the EC monitoring station. However, priorto 2007, no air temperature, relative humidity, solar radiation and wind speed data were available in theregion. Therefore, I used the CFSR data corresponding to the location of the EC monitoring site. The2007 to 2016 meteorological dataset is a composite of the EC monitoring station, the CFSR data, andthe Weatherlink station that is nearby the EC monitoring station. I describe these monitoring stationsand the dierent datasets in Chapter 3.5.2.4 ENSO scenariosIn the near future (~ next decade), the high inter-annual variability of rainfall due to ENSO will likelyto continue to overshadow any detectable climate change trends in rainfall (Altman et al. 2018; Steynet al. 2016). Given the importance of assessing near future climate impacts for local water managers,I explored a range of ENSO scenarios. Knowledge of potential ENSO impacts can also help watermanagers to prepare for hydrological extremes such as floods or droughts, once ENSO forecasts havebeen made public, which is usually 3 to 6 months in advance. While I evaluated the impact of ENSOon water flows for specific years of the 2005 to 2016 historical record in Chapter 4, the analysis in thischapter aims to provide more generalized ENSO scenarios for dierent strength of El Niño and La Niña,that are based on a common index used for ENSO forecasts.Steyn et al. (2016) developed a double Gaussian model that captures the monthly total rainfall of thebimodal rainfall regime of Guanacaste. The double Gaussian model provides separate representationsof the early and late wet season rainfall, and their superposition forms the aggregate model. One of thecovariates of the double Gaussian model is the Oceanic Niño Index (ONI), an indicator for the strengthof El Niño and La Niña (see Appendix D for details on ONI and the double Gaussian model).To explore the generalized impact of dierent ENSO expressions, I used a constant monthly ONI valuefor each of the ENSO scenarios. While ONI typically vary throughout the year, this allows relatingchanges in water resources to dierent strength of ENSO conditions. Hydrological processes are con-tinuous and specific eects of changing monthly ONI would be dicult to distinguish with the hydrologicalmodel. Further, most El Niño and La Niña conditions persist between one to two years (NOAA 2018),and the use of a weather generator to generate daily time series for hydrological model input will create109daily and monthly variability (the monthly mean across the generated time series will reflect the ENSOscenario, but monthly totals will vary), thus not assuming constant atmospheric conditions.I applied five dierent ONI values to the double Gaussian rainfall model to generate ENSO rainfallscenarios, which included ENSO neutral (ONI = 0), moderate (ONI -1) and extreme (ONI -2.5) LaNiña, and moderate (ONI +1) and extreme (ONI +2.5) El Niño (Figure 5.2). The El Niño of 2014-2015 was one of the strongest recorded in history (Sánchez-Murillo et al. 2016), and recorded seasurface temperatures were above an ONI of 2.5 in 2015 for several months (see Appendix D for atable of historical ONI). As input for the weather generator, I determined the monthly dierence betweenthe monthly mean of the baseline period (1980-2016) and each of the ENSO rainfall scenarios (seeAppendix D).Figure 5.2: ENSOmonthly total rainfall scenarios, calculated with the double Gaussian model developedby Steyn et al. (2016).For the WEAP model, I also needed to determine further meteorological input for the ENSO scenarios(i.e., air temperature, solar radiation/cloudiness fraction, relative humidity and wind speed). Steyn et al.(2016) developed an annual classification specifically for Guanacaste to assign ENSO categories (LaNiña/El Niño/ENSO neutral) to each year of the historical record (Appendix D). For each of the timeseries of air temperature, solar radiation, wind speed and relative humidity, I calculated the monthlymean for years of the baseline time series classified as an El Niño or a La Niña (Figure 5.3), and fromthat determined the dierence relative to the baseline mean as input for the weather generator (AppendixD).110Figure 5.3: Meteorological inputs for ENSO scenarios, based on historical baseline (1980-2016) andannual El Niño/La Niña classifications after Steyn et al. (2016): (a) Air temperature, (b) incoming short-wave solar radiation, (c) wind speed, and (d) relative humidity.5.2.5 Climate change scenariosThe climate change scenarios explore impacts of the most extreme pathway of greenhouse gas emis-sions developed in the Intergovernmental Panel on Climate Change (IPCC) Fifth Assessment Report(AR5), i.e. the Representative Concentration Pathway (RCP) 8.5, under which projected greenhousegas emissions result in radiative forcing levels in the year 2100 of 8.5 Wm2 relative to pre-industrialtimes (IPCC 2014; Moss et al. 2010). I chose the RCP8.5 as it allows exploring a ‘worst-case’ scenarioand can help to gauge the outer boundaries of climate change impacts. As time period for the climatechange scenarios, I chose the end of the 21st century also with the reasoning to explore ‘worst-case’impacts, and specifically, I modelled over the time period from 2074 to 2100, based on AlMutairi et al.(2018). I generated the monthly mean for each climate variable to describe the mean climate over thisperiod.Future changes in inter-annual rainfall variability, as caused by possible changes of ENSO under climatechange, were not explicitly explored in these monthly climate change scenarios, as the specific eects111of climate change on ENSO (and its teleconnections) are still debated in scientific literature (Power et al.2013; Maher et al. 2018; Vecchi and Wittenberg 2010). However, the high inter-annual rainfall variability(due to historical ENSO) is reflected in the historical baseline time series, and is reproduced throughthe weather generator for the daily time series for hydrological model input. Thus, inter-annual rainfallvariability is still incorporated over the synthetic time series for the future scenarios.The section below describes the generation of the monthly climate change scenarios, while the nextsection (section 5.2.6) describes the generation of the synthetic daily time series based on thesemonthlymodels for hydrological model input.RainfallI used four dierent rainfall scenarios under climate change, one of which was based on a monthlymean of GCMs under RCP8.5, and three of which were based on climate change scenarios developedby Grossmann et al. (2018) for Guanacaste. I describe the development of these rainfall scenariosbelow.AlMutairi et al. (2018) analyzed GCMs from the most recent Coupled Model Intercomparison Project 5(CMIP5) to assess their skill in reproducing the distinctive rainfall regime of Guanacaste, and concludedthat the following six GCMs best represent the local climate: CanESM2, CNRM-CM5, MIROC5, MPI-ESM-LR, MPI-ESM-MR, MRI-CGCM3 (see Appendix D for details on GCMs). For these six GCMsand the CMIP5 Multi-model mean (which is the mean of 39 GCMs), I obtained mean monthly rainfall,near-surface air temperature, relative humidity and incoming short-wave radiation data over the regionof Guanacaste (274°-275°E, 9.5°-11°N) for the historical period and the RCP8.5 until 2100 from theKNMI Climate Explorer (Koninklijk Nederlands Meteorologisch Instituut) (Oldenborgh 2018). If severalensemble members were available for one GCM, I calculated the mean of these ensembles for eachmonth of the time series. I then compared monthly means from the observed data over the baselineperiod from 1980 to 2016 to monthly means from each GCM over the same historic period from 1980 to2016 (Figure 5.4a), as well as to monthly GCM means over the future period from 2074 to 2100 underRCP8.5 (Figure 5.4b) (see also Appendix D for additional figures).112Figure 5.4: Monthly mean of total rainfall, near-surface air temperature, relative humidity, and incomingshort-wave solar radiation for the observed data composite in Nicoya (dashed line), the CMIP5 Multi-model mean, and the monthly mean of the six selected GCMs for Guanacaste (GCM6 Mean) for (a) thebaseline period from 1980 to 2016 (left panel), and (b) for the period from 2074 to 2100 (RCP8.5) forthe GCMs (right panel). Grey shading indicates the standard deviation (SD) of the 6 selected GCMs.113The monthly mean of the six selected GCMs (GCM6) represented the historical climate much better thanthe CMIP5 Multi-model mean (which is the mean of 39 dierent GCMs), as can be expected consideringthat the six GCM were selected for Guanacaste (AlMutairi et al. 2018). The CMIP5 Multi-model meanstrongly underestimated rainfall in Guanacaste. Other climate change studies for the region have alsoreported systematic underestimation of rainfall of GCMs (both statistically and dynamically downscaled)for the region (Imbach et al. 2018; Karmalkar et al. 2011; Rauscher et al. 2008). While the mean ofthe six selected GCMs (GCM6) captured the rainfall regime in Guanacaste better than the CMIP5 Multi-model mean, it did not capture all the features of the climate (such as late wet season rainfall peak,and timing of early rains and the mid-summer drought), and still underestimated rainfall. Further, thestandard deviation between the six selected GCMs was high.Given the GCM challenges in capturing the Guanacaste climate, I also used the climate change rainfallscenarios developed by Grossmann et al. (2018) for Guanacaste based on expert elicitation discussedabove in section 5.2.1. Specifically, Grossmann et al. (2018) developed dierent monthly rainfall sce-narios, which were expressed as factors of change of the monthly mean between historical and futureclimate (Table 5.1, Appendix D): Scenario 1 explores the potential of amore intensemid-summer droughtunder climate change, with similar annual total rainfall (Scenario 1a), and with a lower second rainfallpeak similar to what is observed during El Niño (Scenario 1b). Scenario 2 explores the possibility that thegeographic location of Guanacaste would be in the transition zone towards the wet tropics and rainfallin Guanacaste would increase. Scenario 3 explores the uncertainty related to the onset of the early wetseason convection, and has reduced early rainfall. In Scenario 3a, the reduced early rainfall are followedby a deeper mid-summer drought and result in a one-peak pattern, while Scenario 3b includes reducedearly rainfall and follows otherwise the historical baseline rainfall. Of these rainfall scenarios, I selectedthe Scenario 1b, Scenario 2 and Scenario 3b for my research, as these capture most of the potentialchanges in rainfall dynamics. I then applied the factors of change developed by Grossmann et al. (2018)(Table 5.1) to the monthly mean of the baseline rainfall time series from 1980 to 2016 (Figure 5.5).114Table 5.1: Factors of change between total monthly historical rainfall and total monthly rainfall underclimate change for scenarios developed by Grossmann et al. (2018) for Guanacaste, Costa Rica.Month Scenario 1b: Intensemid-summer drought withreduced late wet seasonrainfallScenario 2: Transition to wettropicsScenario 3b: Reduced earlywet season rainfall1 1.00 1.00 1.002 1.00 1.00 1.003 1.00 1.00 1.004 1.00 1.10 0.825 1.00 1.20 0.746 1.00 1.20 0.737 0.67 1.40 1.008 0.67 1.20 1.009 0.67 1.20 1.0010 0.67 1.20 1.0011 0.75 1.20 1.0012 0.95 1.10 1.00For hydrological modelling, I used the climate change Scenario 1b, 2 and 3b, as well as the GCM6scenario that described the monthly mean of the six selected GCMs (Table 5.2; Figure 5.5). I did notinclude the CMIP5 Multi-model mean of 39 GCMs in the hydrological modelling, as the comparison tothe local historical climate was poor.Table 5.2: Climate change scenarios selected for hydrological modelling.Abbreviation Description SourceScn 1b Scenario 1b: Intense mid-summer drought with reduced latewet season rainfallGrossmann et al. (2018)Scn 2 Scenario 2: Transition to wet tropics (increased rainfallthroughout wet season)Grossmann et al. (2018)Scn 3b Scenario 3b: Reduced early wet season rainfall and no otherchanges compared to historical baselineGrossmann et al. (2018)GCM6 Monthly mean of six selected GCMs for the region ofGuanacasteMean of CanESM2,CNRM-CM5, MIROC5,MPI-ESM-LR, MPI-ESM-MR,MRI-CGCM3, followingAlMutairi et al. (2018)115Figure 5.5: Mean monthly total rainfall of observed rainfall in Nicoya (1980-2016) and climate changescenarios used for this research (2074-2100): the GCM6 scenario (mean of selected 6 GCMs) andthree rainfall scenarios (Scenario 1b, 2, 3b) developed by Grossmann et al. (2018) for Guanacaste,Costa Rica. Note that scenarios 1b, 2 and 3b were based on change factors (see Table 5.1) and areconsistent with historical rainfall for some months of the year, depending on scenario. Scn 1b = Intensemid-summer drought & reduced late rainfall; Scn 2 = transition to wet tropics; Scn 3b = reduced earlyrainfall; GCM6 = mean of 6 selected GCMs.Wind speedInput for the hydrological model also required other meteorological datasets (wind speed, air temper-ature, relative humidity, and solar radiation data). I developed wind speed scenarios for the futureclimate scenarios that are consistent with the rainfall scenarios, as one of the major drivers for changesin the rainfall scenarios are changes in the northeast trade winds (Grossmann et al. 2018). Duringincreased easterly winds, rainfall in Guanacaste is lower due to increase of the rain shadow eectof the Central American Cordillera. Typically, the increased easterly winds are also accompanied byhigher wind speed. This can be seen by the increase in wind speed during the dry season when thenortheastern trade winds are dominant, in contrast to the wet season when wind speed is generally muchlower (Figure 5.3c). Similarly, during El Niño, the northeastern winds typically intensify, wind speeds arehigher in particular during the mid-summer drought (Figure 5.3c), and rainfall is reduced in Guanacaste.In contrast, during La Niña with higher rainfalls, wind speed is lower during the wet season (Figure5.3c). Considering these observations, wind speed can serve as a surrogate for wind direction, whichis related to rainfall quantities. I explored the relationship between mean total monthly rainfall and meanmonthly wind speed of the historical baseline data, and found a good correlation in a linear model for thewet season when rainfall is > 0 mm (R2 = 0.86; Figure 5.6a). I used the relationship between monthlyrainfall and wind speed to estimate monthly wind speed for the rainfall scenarios for the wet season, andassumed there are no changes in wind speed during the dry season (Figure 5.6b). Wind direction is not116necessary for WEAP model input considering that it is a watershed model. It is important to highlighthere that this linear relationship between wind speed and rainfall is valid only for the specifics of thisstudy site and is not applicable for other regions.Figure 5.6: (a) Linear model of observed monthly mean rainfall and wind speed over the period 1980to 2016. (b) Wind speed scenarios developed for each of the rainfall scenarios, based on linear model.Note that wind speed scenarios were developed based on rainfall scenarios, and for months duringwhich rainfall did not change in comparison to historical rainfall, wind speed also stayed consistent withobserved historical wind speed. Scn 1b = Intense mid-summer drought & reduced late rainfall; Scn 2 =transition to wet tropics; Scn 3b = reduced early rainfall; GCM6 = mean of 6 selected GCMs.Air temperatureNear-surface air temperature is governed by larger-scale processes than the convective rainfall regimeof the region. Historical annual mean temperatures of the GCM6 mean and the observed temperaturesover the period from 1980 to 2016 only varied by 0.02°C (Figure 5.7). Seasonal temperature fluctuationswere also low in this tropical region, ranging from monthly means of 25.8°C in October to 29.1°C in April(observed data, 1980 to 2016).The annual mean temperature showed an increase of 3.5°C from 27.2°C (observed, 1980 to 2016)to 30.7°C at the end of the century (2074-2100, RCP8.5, GCM6 mean). This temperature increasetowards the end of the century is consistent with other climate change studies from the region (Hidalgoet al. 2013; Karmalkar et al. 2011; Maurer et al. 2009).The seasonal temperature cycle of the historical GCM6 mean varied slightly from the observed cycle(Figure 5.7). Therefore, I calculated the temperature dierence for each month between historical andfutureGCM results, and then applied thismonthly dierence to the observedmonthlymean temperaturesto obtain a projection for future monthly temperatures (Figure 5.7).117Figure 5.7: Mean monthly near-surface air temperature scenario for 2074 to 2100, for which the monthlydierence between historical GCM6 mean and future GCM6 mean was determined, and added toobserved historical monthly temperatures.Relative humidity and solar radiationGCM results for relative humidity and solar radiation also showed a slightly dierent seasonal cycle incomparison to historical baseline data. Similarly to the temperature scenarios, I calculated the monthlydierence between historical and future GCM results, and added that monthly dierence to the historicalobserved baseline data to correct for any biases in the seasonal cycle (Figure 5.8). Overall, the monthlydecrease in relative humidity is low, with amean of 0.013 and amaximum of 0.048 during the wet season.The LARS weather generator allowed processing of solar radiation, air temperature and rainfall data togenerate a consistent daily time series (see section 5.2.6). However, relative humidity could not beprocessed with the weather generator. Instead, monthly means were used, as generally, the strongseasonal dynamics of relative humidity between wet and dry season overshadowed daily variability.118Figure 5.8: (a) Relative humidity and (b) solar radiation scenarios, for which the monthly dierencebetween historical GCM6 mean and future GCM6 mean was determined, and added/subtracted toobserved historical monthly values.5.2.6 Weather generatorTo generate the daily meteorological time series for hydrological model input from the monthly ENSOand climate change scenarios, I used a stochastic weather generator, the Long Ashton Research StationWeather Generator (LARS-WG; Semenov and Barrow 2002). LARS-WG is one of the commonly usedweather generators, and has been shown to provide comparable results to other stochastic weathergenerators (Mehan et al. 2017).Synthetic weather generation with LARS-WG consists of three main steps, where first the statisticsof observed daily weather data are determined and with these, the LARS-WG model is calibratedautomatically within the LARS-WG environment (Semenov and Barrow 2002). During this step, LARS-WG uses historical records of daily observed rainfall to fit parameters to semi-empirical distributionfunctions for each month of the year. Probability functions are determined for both the length of theseries of wet and dry days, as well as for the amount of precipitation on a wet day (Semenov and Barrow2002). Next, the model is evaluated, where statistical characteristics of observed and synthetic weatherdata are analyzed and compared for statistically significant dierences. And third, synthetic weatherdata are generated that can either have the same statistical characteristics as the observed data (butdierent daily expressions), or monthly change factors can be applied to account for changes in climate.A synthetic daily rainfall time series is created based on the fitted distributions and the new monthlymean. Air temperature and solar radiation data can be processed along with the rainfall datasets, andsimilarly, changes to the monthly means can be applied.119I used the historical 1980 to 2016 baseline dataset as input for calibration of the LARS-WG model. Per-formance of LARS-WG was good, with no statistical significant dierences between observed and gen-erated data (see Appendix D). I then applied monthly change factors to the historical monthly mean, asdeveloped for the ENSO and climate change scenarios described in the previous sections. I processedthe rainfall time series together with the air temperature and solar radiation time series to preserve thedaily relations between these variables.Using LARS-WG, I then generated synthetic daily weather time series for each of the ENSO and cli-mate change scenarios. For the ENSO scenarios, I generated daily time series of two years each forhydrological model input. I generated six (two year long) runs of each scenario to represent the dailyvariability, that would be likely not captured well in one short two year time series. However, it wouldbe unrealistic to generate a time series that is longer than two years, as atmospheric ENSO conditionswould not persist for that long. For each of the six runs of one scenario, the daily and monthly totalrainfall vary (representing some variability in atmospheric conditions), however, the monthly mean overthe two year time series represents the ENSO scenario. Applying an even shorter time series of forinstance only one year to the hydrological model would be dicult with the hydrological model, due tothe importance of storage terms and continuity of hydrological processes between years.For each of the climate change scenarios, I generated a 26-year time series (similarly to 2074 to 2100 forthe climate change scenarios) for climate change scenarios. The lengths of the time series permittedrepresentation of the variability within the daily rainfall distributions and to exploration of longer-termeects on the hydrological system. To allow comparison to the historical climate, I also generated abaseline synthetic time series (for the historical climate) over 26 years to minimize eects of the LARS-WG model on comparison between historical and future climate.Solar radiation is expressed as cloudiness fraction in the WEAP model, and I processed the generatedsynthetic solar radiation time series using the same relation as during model setup and calibration(Chapter 3).Wind speed and relative humidity time series cannot be processed with LARS-WG to obtain a daily timeseries. The daily variability for these two parameters is relatively low, and they are mostly characterizedby seasonal fluctuation. I therefore assumed the constant monthly mean wind speed and relativehumidity as developed above for each scenario for input into the WEAP model.1205.2.7 Population scenariosDomestic water demand is driven by population. For the baseline and ENSO scenarios, I used 2016population data for all towns that depend on the water supplies of the Potrero and Caimital watersheds(Chapter 3 and Appendix C).Future population estimates were needed for modelling interactions within the socio-hydrological sys-tem under climate scenarios for the 2070-2100 period. As part of the CMIP5, five dierent plausiblenarratives of global socio-economic development were developed that accompany the climate changescenarios and form a framework for future climate impact, vulnerability and adaptation analysis (Riahiet al. 2017). These five narratives of global socio-economic development are called the Shared Socio-economic Pathways (SSP), and include sustainable development (SSP1), middle of the road (SSP2),regional rivalry (SSP3), inequality (SSP4), and fossil-fueled development (SSP5) (Riahi et al. 2017).Each SSP includes population estimates for each country for each decade from 2010 to 2100 (KC andLutz 2017).I compared the population estimates for each SSP for Costa Rica (International Institute for AppliedSystems Analysis, IIASA 2018) with estimates from the Costa Rican Statistical and Census Institute(INEC 2014) (Appendix D). Since INEC estimates do not extend beyond 2050, I used IIASA populationestimates based on SSP scenarios for future modelling. As I am modelling the mean climate over the2074-2100 period, I also used mean population estimates over this period. Specifically, I was interestedin exploring a low and a high population scenario to assess potential impacts of population growth. TheSSP3 results in a high mean population for Costa Rica over the 2070 to 2100 period, while the SSP5scenario results in a relatively low population (that is slightly higher than the Costa Rican population in2010). Therefore, I chose the SSP3 and SSP5 as my high and low population scenarios, respectively(Table 5.3).Table 5.3: Population estimates from Shared Socio-economic Pathways (SSP) for Costa Rica. Datasource: IIASA (2018).Time period SSP3 (high population) SSP5 (low population)2010 4,658,887 4,658,887Mean over 2070 - 2100 8,591,072 5,808,025Based on the Costa Rica-wide population mean over the 2070-2100 period (Table 5.3), I estimatedyearly growth between 2010 and 2085 (mid-point of 2070 to 2100). I assumed that population growthis uniform across Costa Rica (while acknowledging that there might be also shifts in population, asfor example from rural regions to urban centers). I then estimated the mean population for each townwithin the study area for the 2070-2100 period based on the population growth rate and the 2016 townpopulation.121Rural villages within the Potrero and Caimital watersheds and the two nearby towns of Nicoya andHojancha depend on the water supplies of the Potrero and Caimital watersheds. Population estimatesfor Nicoya and Hojancha included the entire district (i.e., urban areas and rural villages). Yet, not theentire district is served with water drawn from the Potrero and Caimital watersheds. For the futurescenarios, I assumed that the percentages of the Nicoya and Hojancha district population that receivedwater from the Potrero and Caimital watershed stayed constant at the 2016 level. For rural villages, Iused population estimates from 2016 (Chapter 3) to apply population growth rates. I also assumed thatthe number of municipal and business service connections grew proportionally with the population in alltowns.5.2.8 Water demand scenariosThe WEAP model calculates water demand based on the annual population and the water demand perperson. First, I assumed a business-as-usual scenario, under which annual water demand per personstayed constant at the 2016 level. WEAP allows modelling variation in seasonal demand as percentageof total annual demand. For Nicoya and Hojancha, I applied a monthly mean (of seasonal variabilityin water demand) over the historical period for which I had water extraction data (i.e., 2005 to 2016).While water demand may vary with climatic conditions, this period included both wetter and drier years,and thus, the mean represented average climate conditions. For the rural villages, I applied the monthlymean from 2012 to 2015, over the period for when data were available. Again, these years includedboth wetter (2012, 2013) and drier (2014, 2015) years.Water extraction volumes and water consumption volumes diered by about 20% for Nicoya and 30% forHojancha in 2016, which corresponds to the water lost in pipeline transfer or otherwise unaccounted for. Iassumed the 2016 percentage for the scenarios as the variation in unaccounted water was low betweenyears. I also assumed that the same percentage of the total population continued to use artisanalhousehold pumping wells. Furthermore, I assumed that the additional water users (i.e., water usersnot already covered through domestic/business/municipal water use, agricultural irrigation, or ranching)stayed constant. For this business-as-usual scenario, I also assumed that supply fractions stayed thesame as in 2016. This meant that the rural villages drew 100% of their demand from the Potrero-Caimitalaquifer, Hojancha drew 100% from the Potrero-Caimital aquifer, and Nicoya drew 58% from the Potrero-Caimital aquifer, 15% from the Nicoya aquifer, and 27% from the Potrero River.These first sets of scenarios assumed that per-capita water demand remains constant into the future.However, there might also be socio-hydrological feedback processes between a changing climate andwater use. For instance in Chapter 4, I found that domestic water demand was typically higher duringthe dry season and in some drier years. Thus, there might be the potential for increased domestic wateruse in a drier climate with longer dry seasons, a longer mid-summer drought and overall reduced wet122season rainfall. On the other hand, a drier climate may also induce behavioural changes and waterconservation strategies, thus reducing the domestic water demand. To explore these two possibilities ofincreased/decreased water use, I also analyzed two “socio-hydrological feedback” scenarios, in whichI considered an increase/decrease of 18% in annual per person water demand for two of the climatechange scenarios (GCM6 and Scenario 3b). The 18% increase/decrease was based on the analysis inChapter 4.5.2.9 Hydrological modelling & post-processingI applied the rainfall and meteorological time series, as well as the population and water demand sce-narios to the WEAP model. For each of the ENSO scenarios, the WEAP model was run for two years foreach of the six generated daily time series. For the baseline and climate change scenarios, the WEAPmodel was run for 26 years. I modelled in total 22 scenarios (Figure 5.9).Figure 5.9: Overview of modelled hydro-climate and social water demand scenarios. ONI = OceanicNiño Index. Green shading and “X” indicate modelled scenario.The WEAP model provides a multitude of results for dierent aspects of the hydrological and socialsystem. For the hydrological system, this includes a daily time step of evapotranspiration values for eachhydrological response unit (HRU) within each sub-catchment (79 sub-catchments in total), infiltration andruno generation for each sub-catchment, percolation from sub-catchments to the aquifer, surface water– groundwater flows for each river reach, groundwater storage, and other components. For the socialsystem, it includes water demand, extraction from dierent supply sources, unmet demands and othercomponents.To export and process these results, I wrote visual basic scripts to automate result data exports fromthe WEAP model. Once exported, I developed R functions to automatically process the results into123daily, monthly, and yearly databases. My main foci of interests were streamflow, groundwater recharge,evapotranspiration, as well as water demand and extraction.For pre-processing of the scenarios and post-processing of the modelling results, I used the R language,with the following R packages: lubridate (Grolemund andWickham 2011), openair (Carslaw andRopkins2012), tidyr(Wickham 2016), ggplot2 (Wickham 2009), hydroTSM (Zambrano-Bigiarini 2017b).5.2.10 UncertaintiesIn most systems, uncertainties in future change impact modelling are high, but especially so when mod-elling a socio-hydrological system with potential interactions and feedback processes between hydrolog-ical and social system components (Figure 5.10). Uncertainties have always been part of hydrologicalmodelling and have been subject of much previous research (Beven 2012; Blöschl 2001; Blöschl andSivapalan 1995; Bronstert et al. 2005; Western et al. 2002), and I will only briefly discuss it here.Uncertainty can arise at the level of hydrological field data collection, and can propagate during pro-cessing for instance when monitored water depth is converted to streamflow via a rating curve model(Westerberg et al. 2011). Hydrological model setup typically necessitates further assumptions and spa-tial and temporal simplifications, and uncertainties can arise due tomodel structure and parameterization(Beven 2012; Blair and Buytaert 2016). Climate change impact modelling then leads to another set ofuncertainties, including the concern of non-stationarity of future climate (Milly et al. 2008). Some ofthese uncertainties are also associated with the scale of GCMs (IPCC 2014; Kiem and Verdon-Kidd2011) and the challenges in reproducing a local climate such as in Guanacaste through both dynamicaland statistical downscaling (AlMutairi et al. 2018; Grossmann et al. 2018; Hidalgo et al. 2013; Imbachet al. 2018; Rauscher et al. 2008). Furthermore, future climate scenarios include a range of dierentfuture trajectories, represented for instance through the representative concentration pathways (IPCC2014; Moss et al. 2010). There are also uncertainties related to changes to ENSO under climatechange (Maher et al. 2018; Vecchi and Wittenberg 2010), which drive the inter-annual rainfall variabilityin Guanacaste (Steyn et al. 2016).For the social system with focus on water extraction, uncertainties also start at the level of field datacollection, where in a region like Guanacaste, only limited data are available, and therefore, assumptionsneed to be made regarding population and water extraction rates. Model setup also necessitatessimplification, and during scenario development, again a range of future trajectories, such as the SharedSocio-economic Pathways (KC and Lutz 2017; Riahi et al. 2017), is possible. There might also be shiftsfrom rural regions to urban centers, which might be challenging to predict.In a socio-hydrological system, further uncertainties can arise through the interaction and co-evolutionof the hydrological and the social system components (Blair and Buytaert 2016; Levy et al. 2016; Mount124et al. 2016; Wagener et al. 2010). This may start at the level of field data collection in a human-used,non-natural watershed, where monitored streamflows for instance may be impacted by unknown surfacewater extraction. Furthermore, water demandmay increase in a drier future climate or, on the other hand,communities may increasingly respond, adapt their behaviour and reduce their water demand. Socio-hydrological changes may occur at the level of ’blue water’ management in streams and groundwaterfor domestic water use. But human-induced land use change could also lead to significant changes, asit will impact ’green water’ availability in the soil and evapotranspiration processes. This in turn can havesignificant impacts on regional water recycling through precipitation (Ellison et al. 2012).In this complex socio-ecohydrological system, changes in societal norms and behaviour can be highlyunpredictable and also surprising, and past observations may no longer be sucient for future guidance(Blair and Buytaert 2016; Wagener et al. 2010). Feedback and interaction between coupled socio-hydrological system bring propagative uncertainties (Blair and Buytaert 2016; Wagener et al. 2010),and socio-hydrology has thus also been described as a “wicked problem” (Levy et al. 2016).Therefore, the scenarios and results presented in this analysis can only provide an idea of possiblefuture trajectories under climate change and population growth in Guanacaste. Yet, they can helpto encompass the range in which change may happen in Guanacaste. This may then support watermanagers and the general public to prepare for a potentially drier climate with substantial impacts onwater resources, and adapt the societal response to change trajectories towards a more resilient future.Figure 5.10: Sketch of uncertainties in modelled socio-hydrological system. GCM = General CirculationModel; RCP = Representative Concentration Pathway; SSP = Shared Socio-economic Pathways.1255.3 Climate and water use impacts5.3.1 ENSO impacts on water resourcesModelling results of ENSO scenarios showed significant impacts on water supplies in the Potrero andCaimital watersheds (Figure 5.11). The scenarios included both moderate and extreme La Niña and ElNiño. Overall, rainfall, groundwater recharge and streamflow were all highest under extreme La Niñaconditions and lowest under extreme El Niño conditions. While there was consistency across thesethree hydrological fluxes, streamflow showed the highest variability with ENSO conditions (Figure 5.11).La NiñaMean total annual rainfall for amoderate and an extreme LaNiñawere 28%and 78%higher, respectively,than for ENSO Neutral (Figure 5.11). For these high rainfall events, annual groundwater recharge alsoincreased by 29% and 70% for a moderate and an extreme La Niña respectively. This is important giventhat episodic extreme groundwater recharge, as related to La Niña, can interrupt multi-year recessionof groundwater levels caused by over-pumping (Taylor et al. 2012), which plays an important role forrenewing community water supplies. Streamflow also increased substantially by 39% and 118% for amoderate and an extreme La Niña, respectively. Thus, while groundwater recharge increased during LaNiña, the majority of the intense rainfall contributed to streamflow.The eect of high rainfall on streamflow was also evident in monthly flows (Figure 5.12), where bothbaseflows in the dry season and stormflows in the wet season were higher during La Niña than duringENSONeutral. Higher groundwater recharge and resulting higher groundwater storage likely contributedto increased baseflow, while high rainfall intensities during the wet season primarily led to stormflow.The flashy response of streamflow to intense rainfall events was also observed in high-frequency fieldmonitoring of streamflow, where streamflow response typically occurred within a few hours of a rainfallevent and lasted only about six hours, indicating that much of the water supplied by rainfall left thewatersheds as streamflow within hours of the rainfall event. This may be caused by the high intensityrainfall exceeding soil infiltration capacities and leading to infiltration-excess overland flow, or soils mightbe saturated, leading to saturation-excess overland flow. High sediment loading observed in streamsafter rainfall events also indicated the importance of surface runo for streamflow generation duringstorm events. The high stormflows in response to intense rainfall events during La Niña can be a causeof concern, as they may lead to flooding with impacts on communities. Watersheds along the Pacificcoast of Costa Rica tend to experience more floods during La Niña then during ENSO Neutral or El Niño(Waylen and Laporte 1999), and have a high probability of abnormally high flow (flood hazard) duringLa Niña (Emerton et al. 2017).126Figure 5.11: Mean percentage dierence of annual totals for ENSO scenarios relative to ENSO Neutralfor rainfall, groundwater recharge and streamflow for dierent Oceanic Niño Indices (ONI).127Figure 5.12: (a) Total monthly rainfall, evapotranspiration, groundwater recharge and streamflow (totalstreamflow combines Potrero and Caimital flows) presented as monthly mean of the modelled timeseries (26 years) in millimeters (relative to total watersheds area), and (b) Flow duration curves of dailymean streamflow (of all model runs for one scenario) of the Potrero and Caimital rivers as mean dailystreamflow in cubic meter per second. ONI for each scenario is given in brackets in legend.El NiñoIn contrast to La Niña, rainfall is much lower during El Niño conditions, with consequences for waterresources. Total annual rainfall decreased by 20% for a moderate and 41% for an extreme El Niño incomparison to ENSO Neutral (Figure 5.11). Groundwater recharge also decreased substantially duringEl Niño (~46% for an extreme El Niño) (Figure 5.11), which may impact community water supplies.The combination of low groundwater recharge and continued high groundwater extraction rates maycause groundwater recession and groundwater levels that are lower than the depth of many communitypumping wells, leading to wells running dry and causing water shortages, as was observed during the2014-2015 El Niño (Vignola et al., 2018). Modelled annual total streamflow decreased by 56% foran extreme El Niño, and both baseflows and stormflows were much lower during El Niño than duringENSO Neutral and La Niña (Figure 5.12). Reduced streamflows may impact environmental flow needsand communities such as Nicoya that withdraw from the Potrero River for their domestic water supply.Seasonality of rainfall and the response of the hydrological system are especially important in the wet-drytropics with their distinct seasonal changes. The early wet season rainfall decreased during El Niño, andin the case of an extreme El Niño, were not much dierent than monthly rainfall during the mid-summerdrought period (Figure 5.12a). This resulted in delayed and reduced streamflow and groundwater128recharge, with potential implications for communities that await the rise of groundwater levels after longdry seasons. The modelled El Niño scenario with ONI = 2.5 represented an extreme case scenario. Yet,a similar El Niño was observed during the fall of 2015 when ONI values increased above 2.0 and up to2.6 (NOAA 2018). The 2014-2015 El Niño was one of the three strongest El Niño recorded (Sánchez-Murillo et al. 2016), and resulted in water conflicts, stress on agriculture, communities and the economy,and declaration of drought emergencies (Vignola et al. 2018).Using ENSO scenarios for water managementWhile ENSO forecasts are available and some water managers in Costa Rica use them to preparefor El Niño related droughts (Vignola et al. 2018), uptake of forecasts diers widely between dierentstakeholders and national and local government (Babcock et al. 2016; Zebiak et al. 2015). Limitedtranslation of ENSO-related rainfall has been made so far to water resource planning, even though manystakeholders identified this as an important need (Babcock et al. 2016). To my knowledge, no otherquantification of ENSO-related impacts on streamflow and groundwater recharge have been previouslyprovided for Guanacaste, or any other region of the wet-dry tropics of Central America. Considering thatENSO is one of the dominant drivers of current climate variability in the region (Steyn et al. 2016), thequantification, and presentation such as in Figure 5.11, can allow water managers to translate ENSOforecasts into impacts on water resources, and thus support better early preparation in the near future.While the percentage changes reported here are specific to the Potrero and Caimital watersheds, theyallow an indication of what could occur in similar watersheds in the lowlands of the wet-dry tropics thatare characterized by a shallow alluvial aquifer in connection with a riverine system.5.3.2 Climate change impacts on water resourcesIn contrast to the ENSO scenarios that focused on climate variability in the current and near futureclimate, the climate change analysis in this section focused on the end of the 21st century (2074 –2100) under the RCP 8.5 emission scenario. I explored a range of possible rainfall scenarios, of whichone scenario was based on the monthly mean of six selected GCMs (Scenario GCM6), and three otherscenarios (Scenarios 1b, 2 and 3b) were based on Grossmann et al. (2018) and specifically presentfuture climates for Guanacaste (Figure 5.13 and Figure 5.14). I describe impacts of climate changerelative to the modelled baseline scenario (Table 5.4). The discussion of climate change impacts onwater resources in this section focuses on results from the low population scenario (SSP5); the combinedimpacts of climate change and higher population growth are explored in the next section.Over the modelling period of 26 years (representative for 2074 to 2100), inter-annual rainfall variabilitydrove high variability of water flows in the Potrero and Caimital watersheds, such as shown for example129for Potrero River streamflow in Figure 5.13b. While it is important to remember that this is a synthetictime series and does not provide a year-by-year prediction over the 2074-2100 period, it indicates someof the water challenges, should for instance multiple dry years occur in sequence.Figure 5.13: (a) Total daily rainfall and (b) mean daily Potrero River streamflow for climate changescenarios. Scn 1b = Intense mid-summer drought & reduced late rainfall; Scn 2 = transition to wettropics; Scn 3b = reduced early rainfall; GCM6 = mean of 6 selected GCMs.Table 5.4: Percentage change of water flows for future climate scenarios (mean over 2074-2100)relative to modelled baseline (mean over 1980-2016). ET = evapotranspiration, PET = potentialevapotranspiration, GW = groundwater. Scn 1b = Intense mid-summer drought & reduced late rainfall;Scn 2 = transition to wet tropics; Scn 3b = reduced early rainfall; GCM6 = mean of 6 selected GCMs.Scenario Rainfall ET PET GW Recharge StreamflowPotreroStreamflowCaimitalScenario 1b -23 -8 9 -26 -45 -42Scenario 2 22 10 3 23 36 35Scenario 3b -7 -1 7 -11 -15 -14Scenario GCM6 -23 -7 9 -28 -42 -40130Figure 5.14: Climate change implications for water resources for a range of possible scenarios forthe Potrero and Caimital watersheds, Costa Rica. Bar plots indicate annual mean and error bars thestandard deviation over the modelled time series. Streamflow is represented relative to watershed area.Scn 1b = Intense mid-summer drought & reduced late rainfall; Scn 2 = transition to wet tropics; Scn 3b= reduced early rainfall; GCM6 = mean of 6 selected GCMs.Drying climates: Scenarios 1b and GCM6Scenario 1b and GCM6 showed the largest decreases in mean annual rainfall, which declined by 23% incomparison to the historical baseline (Figure 5.14; Table 5.4). Warmer air temperatures and increasedsolar radiation increased the potential evapotranspiration for these two scenarios; however, due to limitedwater availability, actual evapotranspiration decreased.Annual mean streamflow also decreased by 40-45% for these scenarios relative to the baseline. Thisdecrease is higher than what other climate change studies have shown for Central America. For in-stance, Hidalgo et al. (2013) indicated a decrease of ~ 10% for southern Central America, a regionwhich included the Caribbean coast with its wet tropical climate and the decrease is thus not specific toGuanacaste. Further, only one streamflow station in Guanacaste was included in the study, and modelcalibration was considered insucient for that station and its results were not included in the runoresults. Imbach et al. (2012) estimated annual runo reductions for the region of Guanacaste of ~ 20%131for the end of the 21st century compared to the historical baseline, with reductions in the 75th and 50thpercentiles (i.e., high flows) of up to ~40%.Annual total groundwater recharge also decreased substantially by 26 and 28% for Scenario 1b andGCM6, respectively. Lower groundwater recharge and resulting lower groundwater levels is linkedto lower baseflow levels for streamflow. To my knowledge, no comparable studies of climate changeimpacts on groundwater recharge exist for the region.The monthly rainfall means of the two climate change scenarios Scenario 1b and GCM6 are similar to amoderate El Niño in the current climate, while inter-annual variability around these monthly means couldstill be expected. Similar to an El Niño, the climate change Scenario 1b showed a dip in rainfall during themid-summer drought in July/August, while the GCM6 scenario did not show a mid-summer drought, butlower early wet season rainfall. The monthly cycle of groundwater recharge and streamflow reflected thechange in the seasonal rainfall cycle, with overall lower recharge and streamflow for each month (Figure5.16). In particular, the historical peak of recharge and streamflow during late wet season rainstormsremained diminished. In contrast in the current climate, the rainfall peaks of the late wet season played amajor role in groundwater recharge generation (Sánchez-Murillo and Birkel 2016). The consequences ofthe reduced recharge, and continued anthropogenic extraction, led to a dramatic decline in groundwaterstorage for the two drying scenarios (Scenario 1b and GCM6) (Figure 5.15), which could have significantconsequences for the population and agriculture, that rely on groundwater and could lead to reducedbaseflows in streams.Monthly mean streamflow during the mid-summer drought and late wet season was reduced by up to60% and up to 50% under Scenario 1b and GCM6, respectively (Figure 5.17). Importantly, baseflowduring the dry season was also significantly reduced by up to 60% under Scenario 1b, and early wetseason flows by up to 70% under Scenario GCM6, and daily flow duration curves showed consistentlylower flows for Scenarios 1b and GCM6 (Figure 5.17). Baseflow is typically already low during the dryseason under the current climate, and a further baseflow reduction with climate change could havenegative consequences for environmental flow needs and ecosystems. Overall, the more persistentmeteorological and hydrological drought conditions and reduced streamflow and groundwater recharge,such as suggested in these two climate change scenarios, could have negative consequences forcommunities and ecosystems.132Figure 5.15: Groundwater volume in storage for each of the climate scenarios over the modelled timeseries. Scn 1b = Intense mid-summer drought & reduced late rainfall; Scn 2 = transition to wet tropics;Scn 3b = reduced early rainfall; GCM6 = mean of 6 selected GCMs.133Figure 5.16: Monthly mean (line) and standard deviation (shading) for (a) actual evapotranspiration (ET),(b) groundwater recharge, (c) streamflow at the Potrero River, and (d) streamflow at the Caimital River.Scn 1b = Intense mid-summer drought & reduced late rainfall; Scn 2 = transition to wet tropics; Scn 3b= reduced early rainfall; GCM6 = mean of 6 selected GCMs.134Figure 5.17: Monthly percentage change for climate change scenarios relative to historical baseline forstreamflow (a) at the Potrero River, and (b) at the Caimital River. Flow duration curves based on dailymodelled flows (c) for the Potrero River and (d) for the Caimital River, separated into the main seasons(Dry season: Dec – Mar; Early wet season: Apr – Jun; Mid-summer drought: July, August; Late wetseason: Sept – Nov). Scn 1b = Intense mid-summer drought & reduced late rainfall; Scn 2 = transitionto wet tropics; Scn 3b = reduced early rainfall; GCM6 = mean of 6 selected GCMs.Delayed early rainfall: Scenario 3bWhile the decrease in total annual rainfall and evapotranspiration was low in climate change Scenario3b in comparison to the historical baseline, the changes in seasonality and the delayed rainfall ledto decreased annual streamflow and groundwater recharge (Figure 5.14; Table 5.4). The delay ofearly season rainfall translates to monthly streamflows that remained low (with reductions of up to13545%) until the late wet season (Figure 5.16; Figure 5.17). This reduction of streamflow for much ofthe year (including reduced baseflow during the dry season) could have detrimental consequences forecosystems if environmental flow needs are no longer met. Further, the delayed increase in streamflowswould only allow the use of surface water during the late wet season.Groundwater recharge was also delayed from the early wet season to the start of the late wet seasonin this scenario while groundwater storage volume recovered, typically during the late wet season rain-fall, and no overall decline in groundwater storage was observed. However, the delayed groundwaterrecharge could impact communities and farmers that await aquifer recharge after the dry season whengroundwater levels are typically low and pumping wells may have run dry. Further, the early rainfallis critical for farmers as it provides the needed soil moisture to start seeding rice crops and to fullyuse the relatively narrow window of the wet season for crop growth. Thus, delayed early rainfall couldimpact agriculture, and also induce farmers to use irrigation in the beginning of the wet season (as iscurrently already done in some parts of Guanacaste, for instance for sugar cane in the Tempisque Rivervalley). This in turn could further contribute to groundwater depletion, especially if groundwater rechargeis delayed and water tables are already low.Transition to the wet tropics: Scenario 2In contrast to the drying climate change scenarios discussed above, there is also the minor possibilitythat Guanacaste will be within the transition zone to the wet tropics and become wetter (Grossmann et al.2018; Rauscher et al. 2008; Imbach et al. 2012). Scenario 2 describes this possibility. The increasein rainfall in this scenario was found to lead to an increase in actual evapotranspiration, groundwaterrecharge and streamflow (Figure 5.14; Table 5.4; Figure 5.16). In particular, the higher intensity rainfallled to increased streamflow (Figure 5.17), similar to what was observed during LaNiña scenarios. Hydro-climatic projections for the end of the 21st century for Panama (Fabrega et al. 2013), just south ofCosta Rica, show results on impacts of climate change south of the transition zone to the wetter tropics.Fabrega et al. (2013) projected an increase in rainfall for most regions including the northwestern Pacificcoast, with related increases in runo (5-10%). Parts of northern Panama, however, also showed adecrease in rainfall and related decrease in runo, displaying the high climate variability and uncertaintyof the region.1365.3.3 Multiple interacting driversAs discussed in the previous section, climate change will likely lead to impacts on both surface wa-ter and groundwater resources. However, both climate change and population growth are drivers ofchange, especially in a country like Costa Rica where high population growth rates are predicted for the21st century (KC and Lutz 2017). This section explores the combined eects of climate change andpopulation growth on water resources.With a business-as-usual approach in water management, population growth towards the end of the21st century will likely lead to an increased domestic water demand (Table 5.5). Specifically, the twopopulation growth trajectories led to a water demand increase of 23% and 78% for the low and highpopulation scenarios respectively.Table 5.5: Annual total domestic water demand in million cubic meters for baseline (2016 population),low and high population (mean over 2070-2100). Note that these are water demand values includinglosses in the distribution systems, and depending on water availability, these demands might not befulfilled in scenario.Scenario Nicoya Hojancha Rural villages Total domestic⇥Mm3⇤ ⇥Mm3⇤ ⇥Mm3⇤ ⇥Mm3⇤Baseline (2016 population) 1.67 0.38 0.38 2.43Low population (SSP5) 2.05 0.47 0.47 2.98High population (SSP3) 2.96 0.68 0.68 4.32This demand growth is particularly problematic as water demands are typically not spread uniformlythroughout the seasons. With no changes in water use behaviour (business-as-usual), water demand ismuch higher during the dry season (Figure 5.18). This is partly due to the agricultural water demand forirrigation during the dry season. However, as discussed in Chapter 4, domestic water demand is alsocurrently higher during the dry season and relies on both groundwater and surface water supplies. Theincreased water demand in the climate change scenarios Scenario 1b and GCM6 under both a low andhigh population scenario, and in climate change Scenario 3b under a high population scenario could notbe fully supplied (unmet demand), as dry season streamflows were too low to meet the surface waterdemand in the dry season and groundwater levels had declined.137Figure 5.18: Monthly water demands for water supply sources from the Potrero and Caimital watersheds,for (a) baseline (2016 population), (b) low population (SSP5) scenario and (c) high population scenario(SSP3) for the end of the 21st century (mean over 2070-2100).High groundwater demands by society exceeded groundwater recharge in most years in a drying climateunder climate change (Scenarios 1b and GCM6) for the low population growth scenario, but even moreso when considering high population growth (Figure 5.19). Groundwater storage dropped each yeartowards the end of the dry season due to continued anthropogenic extraction and streamflow contri-bution, before groundwater recharge of the wet season increased groundwater storage again (Figure5.20). This seasonal fluctuation became more pronounced under the high population scenario. Forthe drying climate scenarios, where demand much exceeded recharge and where extraction rates wereabove the sustainable yield of the aquifer, groundwater storage did not recover during the wet seasonunder both low and high population scenarios, leading to groundwater depletion. While occasionalyears of high rainfall and high groundwater recharge temporarily increased groundwater storage, theydid not reverse the overall trend of declining groundwater storage under continued high extraction rates.Due to the seasonal fluctuation and overall decline of groundwater storage (Figure 5.20), societal waterdemands were not fulfilled in most years for the drying climate scenarios, in particular towards the endof the modelled time series when groundwater storage was low. Further, not all groundwater recharge138is available for human water extraction, but also contributes to baseflow that is essential for aquaticecosystems. For Scenario 3b (with reduced early rainfall, but a total annual rainfall that is similar tothe historical mean), groundwater demand exceeded recharge in some years for the low populationscenario, but groundwater storage could recover during high rainfall (high recharge) years (Figure 5.19and Figure 5.20). In contrast for the high population scenario, demand exceeded recharge in mostyears, and groundwater storage declined over time.For the wetter Scenario 2, recharge was higher than societal demands in most years. However, even inthis wetter climate, there are years of lower rainfall, due to the high inter-annual rainfall variability, whendemand exceeded recharge and groundwater storage dropped low towards the end of the dry season.139Figure 5.19: Dierences between total annual groundwater recharge and groundwater demand bysociety (all sectors) in million cubic meters over the modelled time series. Note that demands were notnecessarily fulfilled every year, due to seasonal fluctuations of water availability and needs, decreasinggroundwater storage and baseflow contributions. Scn 1b = Intense mid-summer drought & reduced laterainfall; Scn 2 = transition to wet tropics; Scn 3b = reduced early rainfall; GCM6 = mean of 6 selectedGCMs.140Figure 5.20: Monthly groundwater storage in million cubic meters over the modelled time series of 26years for the climate change scenarios 1b, 2, 2b and GCM6, considering both low population (solid lines)and high population (dashed lines) scenarios. Scn 1b = Intense mid-summer drought & reduced laterainfall; Scn 2 = transition to wet tropics; Scn 3b = reduced early rainfall; GCM6 = mean of 6 selectedGCMs.The previous scenarios assumed that water demand per person continues at a business-as-usual level,but potential socio-hydrological feedback processes between climate and water demands may leadto dierent outcomes. In Chapter 4, I showed that domestic water demand was about 18% higherduring the dry season. If no water conservation or adaptation strategies are employed, a similar trend ofincreased water demands may occur in a drier climate under climate change. On the other hand, a pro-active respond of society to a drying climate with increased water conservation eorts might also lead toreduced water demand. To explore the impacts of such a potential increase/decrease of water demandunder a drier climate, I applied an 18% increase/decrease in annual domestic water demand. Agriculturalwater demand could also increase in a drier climate, asmore irrigation would likely be needed. Increaseddomestic water demand was found to shift this system even faster to a non-sustainable system underthe drying Scenario GCM6, where groundwater extraction volumes exceeded the sustainable yield ofthe aquifer, and irreversible groundwater depletion occurred within few years (Figure 5.21). Even underScenario 3b where total annual rainfall was only reduced by 7% in comparison to the historical baseline,the combination of high population growth and a socio-hydrological response led to a rapid declinein groundwater storage. Reduction of per-capita water demand was able to slow down the decline ofgroundwater storage (Scenario GCM6), and for the high population scenario under climate scenario 3b,even reverse the decline in groundwater storage (Figure 5.21).141Figure 5.21: Changes in groundwater storage in million cubic meters under climate change scenarios(a) GCM6 and (b) Scenario 3b (2074-2100) under low and high population growth for business-as-usualwater demand, and considering socio-hydrological feedback of increased domestic water use (+18%per-capita water use), and reduced domestic water use (-18% per-capita water use). GCM6 = mean of6 selected GCMs; Scn 3b = reduced early rainfall.1425.4 ConclusionCommunities in the wet-dry tropics of Guanacaste currently suer from recurrent droughts, often relatedto El Niño. With climate change, droughts may become even more frequent in the future. To increase theresilience of communities to drought, knowledge on the impacts of climate variability on water resourcesis needed. Yet, translation from ENSO forecasts and climate change scenarios to implications on waterresources remains limited. Most studies only provide analysis on a regional level for all of CentralAmerica, which does not capture the specific and complex climate dynamics of Guanacaste, nor doesit capture socio-hydrologic feedbacks.In this chapter, I quantified potential ENSO impacts on streamflow and groundwater recharge for twowatersheds in the wet-dry tropics of Central America using a hydrological model. The scenarios coveredmoderate to extreme La Niña and El Niño. Modelling results showed that extreme El Niño can reducestreamflow and groundwater recharge by ~60% in comparison to ENSO Neutral. Considering that thecurrent and likely the near future inter-annual climate variability is dominated by ENSO, these results cansupport water managers’ eorts to assess the potential impacts of forecasted ENSO on water supplies.To consider a longer-term planning horizon towards the end of the 21st century, I also analyzed theimplications of a range of future climate change scenarios on water supplies in the two watersheds.The climate of Guanacaste is complex and most global climate models (and dynamical and statisticaldownscaling) have challenges in reproducing it. I used a novel approach and applied future climatescenarios developed specifically for Guanacaste through an expert elicitation protocol. These scenarioscovered several potential future climates, ranging from a drier climate with an increased mid-summerdrought to delayed early rainfall, as well as one scenario in which Guanacaste is considered in thetransition zone towards the wetter tropics. Modelling results showed that a drying climate may leadto new annual means of streamflow and groundwater recharge that are reduced by 40-45% and 26-28%, respectively, in comparison to the historical baseline, while inter-annual variability around thesemeans can still be expected. Importantly, groundwater storage was found to decline under a dryingclimate with continued extraction, which could lead to detrimental consequences for towns, agricultureand ecosystems.Further, climate change is not the only driver of change in the region, and population growth alongwith a growing water demand may also impact water resources. Therefore, I analyzed the impact ofthe combined drivers of climate change and a growing water demand. Under business-as-usual watermanagement, high population growth can lead to groundwater depletion where extraction rates areabove the sustainable yield of the aquifer and groundwater levels continue to drop. Even under a wetterclimate, high groundwater extraction may lead to a substantial drop in groundwater levels towards theend of the dry season, when water levels could fall below the level of pumping wells. The system143holds a fine balance between extraction and groundwater recharge, which may be perturbed in thecase of a drying climate with multiple years of lower rainfall, high population growth, or significant socio-hydrological feedback processes (such as increased water demand in a drier climate), and groundwaterdepletion may occur.The climate and population growth scenarios presented in this research present a range of possiblefuture trajectories, and highlight potential sensitivities to climate and a growing water demand. Thetranslation of climate scenarios to implications on water supplies can support water managers and thegeneral public to prepare for a potentially drier climate, and adapt their societal response to change theirtrajectories towards a more resilient future.144Chapter 6Groundwater recharge indicator astool for decision makers to increasesocio-hydrological resilience todrought6.1 IntroductionInter-annual variability of seasonal rainfall and groundwater recharge can pose water challenges forcommunities in regions with extended dry seasons that are reliant on shallow groundwater to meet theirwater demand. These challenges will likely increase in the future with projected increases in seasonaland inter-annual rainfall variability in Central America (Magrin et al. 2014) and increasing water demands(Wada and Bierkens 2014). Seasonality of rainfall defines many climate systems around the world. ThePacific coast of Central America is one of the hotspots of rainfall seasonality, together with northeasternBrazil, western sub-Saharan and central Africa, northern Australia and parts of Southeast Asia (Fenget al. 2013). In these regions, a temporal mismatch between water availability and water needs can leadto technical and natural water shortages and conflicts between dierent sectors (i.e., agriculture anddomestic water demand), and between human and natural system needs (Ballestero et al. 2007). Dueto the high inter-annual rainfall variability and the complex interplay of many local, regional and globalclimatic forces, future climate projections have high uncertainty for the region, but will likely include morefrequent dry conditions (Chapter 5).As in many regions with seasonal rainfall regimes, communities in the wet-dry tropics of Costa Ricadepend primarily on groundwater, which supplied 78% of the total domestic water demand in 2015(Province of Guanacaste, CTI 2017; Guzman 2015). Many of the water conflicts reported for theregion evolved around groundwater extraction, groundwater allocation between domestic, tourism andagricultural sectors, and weak water governance structures (Kuzdas et al. 2015a, 2015b). Groundwater145typically provides the primary water source during the dry season when streamflow is low (see Chapter4, Ballestero et al. 2007). Therefore, seasonal groundwater recharge during the wet season is key forreplenishing aquifers (Jasechko et al. 2014) and assuring water supply. Yet, groundwater recharge inthe tropics is strongly dependent on the intensity of rainfall during the wet season (Jasechko and Taylor2015), and is more pronounced in years of extreme seasonal rainfall (Owor et al. 2009; Taylor et al.2012). This in turn means that in years of low seasonal rainfall, groundwater recharge tends to be lower.For communities depending on renewable shallow groundwater for their livelihoods and agriculturalirrigation, this can lead to significant water stress as happened during the dry season following the ElNiño 2014-2015 in the dry corridor of Costa Rica (Vignola et al. 2018).Therefore, groundwater has to be managed carefully to ensure water supply lasts throughout the dryseason, and baseflow and environmental flow needs are respected. Yet, groundwater managementis a challenge in the region due to scarce hydrological data and aquifer studies, limited oversight ofgroundwater pumping by government agencies, and inecient water governance (Ballestero et al. 2007;Vignola et al. 2018; Kuzdas et al. 2015b).The impacts of droughts on communities depend not only on the meteorological events themselves (i.e.,the hazard), but also on exposure and vulnerability of communities to these extreme events (Magrinet al. 2014). Historically, Central America has been considered as a region with high vulnerabilities ofcommunities to natural disasters, due to the levels of poverty and livelihood structures (Magrin et al.2014). Agriculture is one of the most important livelihoods in the Central America and in particular,smallholder and subsistence farmers can be severely impacted by droughts (Hannah et al. 2017). Inthe wet-dry tropics of Costa Rica, most farmers use groundwater-based irrigation agriculture in the dryseason, and rainfed or irrigation (both surface water and groundwater-based) agriculture in the wet sea-son (INEC 2015). The rural and agricultural communities depend directly on their local water resources,and especially rural communities with weak water management structures often suer the direct eectsof droughts (Kuzdas et al. 2015a). In the agricultural watersheds, social and hydrological processes aretightly connected, and processes from one system component will aect the other (coupled human-watersystems, socio-hydrology) (Sivapalan et al. 2012). This has to be considered when aiming to increaseresilience to drought.Here, resilience refers to the capacity of the system to absorb and adapt to recurrent disturbances withoutmajor system shifts and to develop with a changing environment (Folke 2006; Folke et al. 2016; Maoet al. 2017). Hydrological resilience (i.e., focusing on the response of the water subsystem to extremeevents or anthropogenic impacts) should not be considered separately from the social system (e.g.,social resilience to hydrological hazards), but rather the combined socio-hydrological resilience shouldbe considered (e.g., the coupled human-water system) (Mao et al. 2017). To support the developmentof socio-hydrological resilience to droughts, it is crucial for scientists to integrate their research within146communities and communicate scientific results to local decision makers in a way that they can directlyapply and act upon.Therefore, in this chapter, I address my overarching research question 3 (see Section 1.2), and aim touse scientific information to support water managers to increase the resilience of their communities toseasonal drought. I focus hereby on groundwater recharge, as groundwater recharge in the wet seasonis the key process for determining water supplies in the dry season when communities and agriculturerely on groundwater as their primary water source. I approach my research question 3 through threesub-questions.Research Question 3: How can water managers increase the resilience of their communities to sea-sonal drought and prepare in years of reduced rainfall for the oncoming dry season?• What is the relation between seasonal rainfall and groundwater recharge for the Potrero-Caimitalaquifer?• How can the socio-hydrological resilience of the system to seasonal drought be characterized?• And what decision support tool could support local water managers for increasing the socio-hydrological resilience of communities and prepare in years of reduced rainfall for the oncomingdry season?6.2 MethodsField site description is provided in Chapter 2. Hydrological monitoring and model setup is describedin Chapter 2 and 3, respectively. The development of the groundwater recharge indicator is describedbelow.Total groundwater recharge to the Potrero-Caimital aquifer was calculated as the sum of modelledrecharge from sub-catchments (diuse recharge) and from surface water (focused recharge) for eachyear of the modelling period from 2005 to 2016. To explore the relation between rainfall and groundwaterrecharge, I fit a linear regression model to total annual rainfall and total annual groundwater recharge.For comparison to modelled recharge, I also attempted to estimate groundwater recharge empiricallyusing the groundwater level fluctuation method as well as by residual of the water budget. However,due to too many missing components and uncertainties of the empirical data (for instance, groundwaterlevels influenced by pumping, uncertainties of baseflow), neither method could be applied successfully.Next, I calculated the cumulative rainfall and cumulative modelled groundwater recharge throughouteach year based on daily data, and used this relationship to develop a groundwater recharge indica-tor. Here, I calculated the 0, 25, 45, 55, 75 and 100 percentiles for each day of year based on the147daily distribution of the 12 years of the modelling period for both cumulative rainfall and cumulativegroundwater recharge. Based on the percentiles, I determined five categories: 0 to < 25, 25 to < 45,45 to < 55, 55 to < 75, and 75 to 100. To quantify how well groundwater recharge can be predictedbased on cumulative rainfall throughout the wet season, I calculated confusion matrices (Kuhn 2008)for each day of year for each of the five percentile categories. Confusion matrices 1) count the numberof times that a predicted value matches the ‘true’ value, and 2) report the number of true positives,false positives (type I error), false negatives (type II error) and true negatives. For each day of year,I determined the percentile category into which the cumulative rainfall fell. From this, I predicted thepercentile category into which the cumulative groundwater recharge would be expected to fall at the endof the year (predicted category). The percentile category of the cumulative groundwater recharge atthe end of the year represented the ‘true’ category. I calculated a confusion matrix for each day of theyear for each of the 12 years of our modelling period, and extracted the overall accuracy. I then usedlocally weighted smoothing (Wickham 2009) to show the pattern of the overall accuracy throughout thewet season, and fit the curve through the origin after the first month of the wet season.6.3 Groundwater rechargeTo addressmy research questions, I first explored the relation between groundwater recharge and rainfall(research sub-question 3.1), for which I looked at seasonal fluctuations of observed groundwater levels,and then assessed quantitatively the relation between annual total rainfall and groundwater recharge.Next, I characterized the socio-hydrological resilience of the system (research sub-question 3.2), with thegoal to identify shortcomings and thus target the decision support tool to increase resilience to seasonaldrought, that I developed in the following section (research sub-question 3.3).6.3.1 Observed groundwater levelsObserved groundwater levels in the Potrero-Caimital aquifer fluctuated seasonally as illustrated in thelong-term record (2005-2015) for well X29 in the Potrero-Caimital aquifer (Figure 6.1; Data obtained fromSENARA, well X29 is located close to Casitas, see Appendix Figure A.31). Groundwater levels rose inresponse to recharge during the wet season, and declined during the dry season due to groundwater dis-charge to streams (as baseflow) and pumping, with an overall range of annual variation of approximately7 m (Figure 6.1). Similar seasonal groundwater level fluctuations have been shown for the Nimboyoresaquifer, an alluvial aquifer that is also located within the wet-dry tropics of Guanacaste (north-west ofSanta Cruz towards the coast, Figure 2.1) (García and Arellano 2012), and these seasonal fluctuationsmay be a typical feature of alluvial aquifers in this region.148Figure 6.1: Observed water table depths for two wells in the Potrero-Caimital aquifer and monthlytotal rainfall: (a) Monthly water table depths for well X29 in Casitas (~ 50 m from nearby creek, seeAppendix Figure A.31), from 06-2005 to 09-2015. Data obtained from SENARA; (b) Mean daily watertable depths, calculated from observed 30-minute water level data from the monitoring station at GW2in Dulce Nombre(~190 m from nearby creek). In Spring 2016, water levels were below sensor depth(depth > 9.4 m) and the well ran dry.6.3.2 Modelled groundwater rechargeOver the modelled time period from 2005 to 2016, high seasonal rainfall resulted in high seasonalgroundwater recharge with a range between 98 mm a1 and 252 mm a1 (Table 6.1). Total annualrainfall (measured) and total groundwater recharge (modelled) had a statistically significant linear rela-tionship (R2 = 0.97, p < 0.001; Figure 6.2). The percentage of total annual groundwater recharge to totalannual rainfall stayed relatively constant for both low and high total annual rainfall at 7 to 8%. In contrast,the percentage of total annual streamflow to total annual rainfall increased from 34% to 50% from low tohigh annual rainfall, suggesting the increased generation of overland flow during high intensity rainfallevents. Overland flow runo was also evident by high sediment loading in rivers after intense rainfallevents, as was observed in the field.149Table 6.1: Observed total annual rainfall and modelled groundwater recharge in mm from 2005 to 2016.Year Rainfall Groundwater Recharge[mm] [mm]2005 2,578 1882006 1,384 1022007 2,823 2112008 2,625 1952009 1,310 982010 3,235 2522011 2,259 1872012 1,623 1202013 2,215 1612014 1,551 1072015 1,953 1562016 2,295 190Figure 6.2: Modelled total annual groundwater recharge versus observed total annual rainfall in mm, forthe Potrero-Caimital aquifer, Costa Rica. Line indicates linear regression model (R2 = 0.97, p < 0.001).Similar to other tropical aquifers (Jasechko and Taylor 2015; Mileham et al. 2009; Owor et al. 2009;Sánchez-Murillo and Birkel 2016; Taylor et al. 2012), intense seasonal rainfall is important for rechargingthe Potrero-Caimital aquifer. Sánchez-Murillo and Birkel (2016) showed the importance of the intenserainfall of the second wet season peak in September and October for groundwater recharge for aquifersin the wet-dry tropics of Costa Rica (Pacific lowlands). This is reflected in the increased groundwaterrecharge that I found for the Potrero-Caimital aquifer in years of high annual rainfall, in which monthlyrainfall is typically highest during the fall months (second peak of the wet season, Steyn et al. 2016).Similarly, Jasechko and Taylor (2015) showed that intensemonthly rainfall contributed disproportionatelyto groundwater recharge for 14 tropical aquifers across dierent continents. While Owor et al. (2009)showed that the daily amount of rainfall and groundwater recharge were linearly related for an aquifer in150Uganda, with better coecients of determination when only considering daily total rainfall larger than 10mm, a good linear correlation between rainfall and groundwater recharge is not always found. Studiesfrom large aquifers in East Africa (Mileham et al. 2009; Taylor et al. 2012) found a nonlinear relationbetween rainfall and groundwater recharge, where episodic recharge events resulting from abnormallyhigh seasonal rainfall interrupted multiannual groundwater level declines (Taylor et al. 2012). In contrastto the majority of the aquifers mentioned above, the Potrero-Caimital is a fairly small aquifer with adominance of clayey-loamy soils, where flashy surface runo is the dominant response to intense rainfallevents, which helps to explain the good agreement found between annual rainfall and groundwaterrecharge.It is important to remember that the groundwater recharge estimates carry some uncertainty due tomodelling assumptions and available input data, and the linear relationship between rainfall and ground-water recharge is based solely on modelling results. Some limitations of the modelling included the dailytimesteps, which were not able to capture sub-daily rainfall-runo processes in these flashy watersheds.Infiltration capacities might be exceeded by high intensity rainfall events, leading to more overland flowand higher stormflows in the real system than can be modelled with a daily model. Higher stormflowsof the real system could then also increase focused recharge from streams to groundwater. The sim-plified representation of surface water – groundwater interactions in the model, and considering thatgroundwater flow was not modelled, likely introduced further uncertainties regarding recharge. Othersimplifications of the model related to the definition of the HRUs, land use and soil may also impactmodelled diuse recharge. For example, Toohey et al. (2018) showed that soil hydraulic conductivityand percolation was higher at a forest site than at a pasture site for a watershed in the wet tropics inCosta Rica.6.3.3 Socio-hydrological resilienceIncreasing water use and seasonal fluctuations of groundwater levels have led to water shortagesin the dominantly groundwater dependent communities of the studied region. Thus, droughts andwater conflicts often manifest themselves in dry seasons following wet seasons with lower than normalrainfall, as often occurs in relation to El Niño conditions in western Central America. The current socio-hydrological system is not resilient to these external impacts from reduced rainfall (Figure 6.3, left), andcommunities and ecosystems experience socio-hydrological drought as a consequence.To move towards a more drought-resilient system, I must first define what ‘socio-hydrological resilience’means in the context of seasonal droughts in this system (conceptualized in Figure 6.3). Resilience,or the capacity of a system to absorb disturbances without significant challenges to its functioning orstructures, is composed of absorptive, adaptive and transformative capacities (Mao et al. 2017; Milleret al. 2010). Here, absorptive capacities refer to inherent system components that can help to lessen151impacts or shocks of an external event, such as reduced rainfall, on social and hydrological subsystems.Adaptive capacities are the ability of system components to respond to a disturbance, again with the goalof reduced impacts. Transformative capacities also describe the ability to respond, but more radicallythan adaptive capacities (Mao et al. 2017). Thus, transformation would constitute a system shift towardsa new desired state, should current ecological, economic or social structures become unsupportable dueto an external impact (Mao et al. 2017).I recognize that what I am defining here as external impacts for this analysis (i.e., rainfall variability) is infact not external to the socio-hydrological system, as human-caused climate change (Magrin et al. 2014)and land use change (with related changes in evapotranspiration and precipitation recycling; Ellison et al.2012) influence the temporal and spatial variability of rainfall.Figure 6.3: Conceptual non-resilient and resilient socio-hydrological systems, with focus on groundwaterpumping and impacts in response to external force (reduced rainfall).In the studied, non-resilient system, reduced rainfall (i.e., a meteorological drought) leads to reducedgroundwater recharge, such that many wells become dry and groundwater extraction becomes insuf-ficient to meet the needs of the rural communities. The adaptive capacity of the social subsystemto reduced rainfall is low, as groundwater pumping continues at high rates until groundwater levelsdrop below the depth of many wells, and water shortages begin to impact rural communities as well asecosystems (i.e., there is no adaptive response to reduced rainfall via water management). Absorptivecapacities are also low, as communities rely mostly on the single resource of groundwater, and are152not using other water supplies, which could help to dampen the shock of reduced groundwater levels.While annual renewal rates of the aquifer are relatively high (mean of 13%, for an estimated total aquiferstorage of about 40 Mm3), current borehole depths often do not reach lower groundwater levels, leadingto technical water shortages.In contrast, in a more resilient system (Figure 6.3, right), communities would respond adaptively toreduced groundwater recharge during a wet season with reduced rainfall and decrease their groundwaterpumping as one measure to lessen impacts on groundwater storage for the subsequent dry season (thehydrological subsystem). One of the barriers to this is information, i.e., that communities become awareof pending challenges early enough to adapt their behavior. For wet-dry tropical systems, the mainimpacts from reduced rainfall in the wet season (i.e., meteorological droughts) often occur later in thefollowing dry season.This delay provides an opportunity to make the system more resilient to oncoming impacts. An adaptiveresponse and increased absorptive capacity of the system can help to make the socio-hydrologicalsystem more resilient towards reduced rainfall and seasonal droughts. For example, once it becomesapparent that total recharge is likely to be low or very low for a given wet season, there is the potentialto shift to other water sources such as rainwater harvesting or increased surface water use duringthe remaining wet season. This would also reduce the reliance of communities on the single waterresource of groundwater. Further adaptive responses could include reducing water permits for irrigationof dry season crops after wet seasons with low recharge, and improving agricultural soil and watermanagement. Considering the relatively high total storage volume of the Potrero-Caimital aquifer andgroundwater reserves, there may also be some buer capacity for droughts. Deepening well screendepths of boreholes could reduce technical water shortages and increase resilience to drought. How-ever, it is a more expensive approach that could lead to equity issues, as drilling of deeper boreholesmay be restricted to communities or individuals with more financial capacities. Additionally, extractingdeeper groundwater may lead to extraction rates above the sustainable yield of the aquifer (i.e., abovethe withdrawal rate at which no adverse impacts on ecosystems, land subsidence or other aspects of thehydrologic system occur, Healy 2010), and it could lead to a long-term decrease in groundwater levels.1536.3.4 Groundwater recharge indicatorThe key to making the socio-hydrological systemmore resilient to droughts is to increase both absorptiveand adaptive capacities. In case these responses would not be sucient, a more radical shift could bea transformation towards a new desired system state. The eectiveness of dierent, specific adaptationand water management strategies should be explored in future research under dierent climate changescenarios. But in general, these could include long-term strategies, such as diversifying water sourcesfrom pre-dominantly groundwater use. On a seasonal time-scale, adaptive responses would need to bemade in time for them to be eective in the wet-dry tropical climate where drought impacts often occuras a delayed response to the highly variable external driver of rainfall. For this, adequate information isneeded for local decision makers to both assess the situation in time, and also, to communicate to thegeneral public the need for adaptive response (such as, water restrictions).To address this need, I developed an indicator-based tool that allows local decision makers to assessthe likely status of groundwater reserves in order to provide a framework to react adaptively before majorimpacts of drought hit. One of the main challenges of groundwater is that it is often poorly monitoredand inadequately managed as it is less visible than surface water (Famiglietti 2014). On the other hand,rainfall is monitored in most countries of the wet-dry tropics and most decision makers have access torainfall data through their national meteorological institutes. I used the strong relation between rainfalland groundwater recharge (Figure 6.2) to develop a groundwater recharge indicator (Figure 6.4) thatrepresents a) cumulative rainfall and b) cumulative groundwater recharge over the wet season. The 12years of the modelling period represented a wide range of wet and dry conditions, from 1,310 mm to3235 mm for total annual rainfall and from 98 mm to 252 mm for total annual groundwater recharge.For each day of year, I determined the daily distribution of cumulative rainfall and groundwater rechargewhich were then placed into five percentile categories of rainfall and groundwater recharge, respectively:very high (dark blue), high (light blue), medium (yellow), low (orange), very low (red). The trajectoriesof cumulative rainfall and groundwater recharge for each year showed that, after some variability at thebeginning of the wet season, they tended to stay within one category for the remainder of the wet season.154Figure 6.4: Groundwater recharge indicator developed based on cumulative daily (a) observed rainfalland (b) modelled groundwater recharge for the Potrero-Caimital aquifer from 2005 to 2016. Colorscheme indicates percentiles of daily distributions. Grey arrows indicate example (August 1). Colorsbased on Brewer (2018).155To use this indicator tool, one determines the cumulative rainfall to date (e.g. August 1 in Figure6.4a) and the corresponding percentile category. This percentile category is then directly related tothe projected groundwater recharge category for the same date to determine the projected cumulativerecharge (Figure 6.4b). Figure 6.5 provides step-by-step instructions on how to use the groundwaterrecharge indicator.Figure 6.5: Step-by-step instructions for water managers for use of the groundwater recharge indicator(Figure 6.4), including an example. Colors based on Brewer (2018).156To explore the validity of the groundwater recharge indicator, I assessed the overall accuracy of theprojected end-of-year-groundwater-recharge-category estimated from the day-of-year-rainfall-categoryusing confusion matrices. Confusion matrices count the number of times that a predicted value matchesthe ’true’ value (Kuhn 2008), where here, the ’predicted’ value is the groundwater recharge category aspredicted by day-of-year-rainfall-category, and the ’true’ value is the groundwater recharge category asindicated by modelled total annual groundwater recharge. By the end of May to the end of June, theaccuracy of the projection is high (> 75% accuracy) in most cases, in particular for drought projections(Figure 6.6). For high and very high categories, overall accuracy lessens somewhat in July and August,which coincides with the timing of the mid-summer drought.It is important to remember that any model is an abstract and simplified representation of reality (Beven2012), and as such, there are uncertainties associated with its predictions. Uncertainties within themodelling can be caused by uncertainties in field measurements (such as streamflow and soil mea-surements), in model assumptions, and model procedures. This should be kept in mind when usingthe groundwater recharge indicator. I tried to limit uncertainties by using the five broad categories ofthe percentile classes, instead of trying to project exact groundwater recharge numbers. To improvethe tool in the future, water managers could also use observation wells and develop a relation betweengroundwater level and rainfall more directly than using recharge, which is dicult to measure or model.For this, water managers would need to ensure however that monitored groundwater levels are notinfluenced by any nearby pumping, whichmight be challenging in these watersheds with many registeredand unregistered wells. Further, previous wet season rainfall and dry season pumping, as well asincreasing trends of anthropogenic pumping throughout the aquifer would also influence monitoredgroundwater levels. Yet, it would provide an additional indication of the current state of the aquiferand could support the use of the groundwater recharge indicator.157Figure 6.6: Overall accuracy (from confusion matrices, 1 = full accuracy, 0 = no accuracy) for predictionof groundwater recharge category by the end of the year based on rainfall category for each day of year,starting with first rainfall in April. Daily confusion matrices were calculated for each category: (a) veryhigh, (b) high, (c) medium, (d) low, and (e) very low. Dots represent daily accuracy, line representslocally weighted smoothing (LOESS) curve (Wickham 2009).When using the groundwater recharge indicator, it is important to assess cumulative groundwater rechargeearly on during the wet season, i.e., while it is still raining and there is still water available, and not towait until the end of the wet season, when adaptive responses are limited and negative impacts togroundwater storage may already have occurred. Should the groundwater recharge indicator point tolow or very low groundwater recharge, immediate actions should be initiated with the overall goal to“bank” groundwater for the next dry season when it is the only water supply available.To this end, overall water use could be reduced, rainwater could be harvested during the wet season, andsince high percentages of rainfall leave the watersheds as streamflow during the wet season, increaseduse of surface water during the wet season could reduce the overall impact on groundwater. Surfacewater cannot be used much during the dry season as baseflows are low and it could have negativeimpacts on ecosystems. Furthermore, water managers need to be aware that net recharge at the endof the wet season is likely much lower than total recharge, as groundwater pumping and groundwatercontributions to baseflow continue throughout the wet season. For instance for the modelling periodfrom 2005 to 2016, total annual baseflow (from the Potrero-Caimital aquifer to the Potrero and Caimitalrivers and tributaries) ranged from 54 to 190 mm (between 55 to 85% of annual recharge).158The groundwater recharge indicator can allow water managers to plan ahead and make informed deci-sions, and importantly, to communicate the need for adaptive responses to the public. It should be usedas a tool for a short-term (within a wet season) adaptive response to rainfall, and be integrated within aset of more long-term water management strategies that account for increasing water extraction rates,land use change and climate change. Specific sets of short- and long-term adaptation strategies shouldbe explored in future research to assess their eectiveness under a changing climate and changingwater demands.While rainfall and climate forecasts are available in the region, so far, limited translation to consequenceson water supplies has been made, even though many water managers, farmers and general publicmembers identified this as an important need (Babcock et al. 2016). I presented the groundwaterrecharge indicator during stakeholder workshops in November 2017 in Nicoya, Guanacaste, Costa Rica,where it was well received, and I refined the tool through the feedback provided in these workshops.The concept of the groundwater recharge indicator is potentially adaptable to many regions around theworld, though it is particularly relevant to regions that experience high rainfall seasonality. Hotspotsof rainfall seasonality include for instance also northeastern Brazil, western Sub-Saharan and centralAfrica, northern Australia and parts of Southeast Asia (Feng et al. 2013). In these regions, a temporallag between water availability and water needs necessitates an early adaptive response during thewet season to prepare for the following dry season. To adapt the groundwater recharge indicatorconcept to another aquifer, groundwater recharge would need to be measured or modelled, and thelocal relationship with cumulative rainfall determined. Importantly, the length of the time series availablewill impact the robustness of the indicator.6.4 ConclusionCommunities in regions with seasonal rainfall face a temporal lag between water availability and need.This often leads to water shortages and water scarcity during long dry seasons. Most seasonally-dry regions depend on groundwater as their primary water source during the dry season. However,if not carefully managed, groundwater recharge during the wet season may be insucient to meet waterdemand during the subsequent dry season.In this study, I developed a ‘groundwater recharge indicator’ for an aquifer in the wet-dry tropics of CostaRica in order to provide a tool that can help communities prepare in a timely manner for dry seasonsfollowing wet seasons with low rainfall. The tool allows early prediction of the likely groundwater rechargeat the end of the wet season based on accumulated seasonal rainfall to date. This tool will allow watermanagers to assess the likely state of groundwater during the wet season when water is still available,and implement adaptive responses to low groundwater recharge, such as using rain and surface water159(supply management) or water conservation (demand management) to ‘bank’ groundwater for the dryseason when it is the only water supply. This idea can be applicable to other groundwater-dependentcommunities in regions with seasonal rainfall regimes. The tool provides a way to communicate scien-tific results to decision makers to support them in increasing the socio-hydrological resilience of theircommunities to seasonal droughts.160Chapter 7ConclusionsIn a changing world with more frequent hydrological extremes and increasing pressures on limitedwater resources, ensuring water security is becoming more and more of a problem for many ruralcommunities in developing countries. Scarce hydrological data and poor information on potential futureclimate change impacts can further impede adaptive water management, along with limited knowledgeon current water demands of dierent sectors. The objective of this thesis was to assess current andfuture socio- and hydrological dynamics with respect to water use and impacts on surface water andgroundwater supplies with the goal to inform drought adaptation. Focusing on two drought-prone ruralwatersheds (the Potrero and Caimital watersheds) in the seasonally-dry tropics of Central America, Iundertook a detailed assessment of current socio-hydrological vulnerabilities to drought. I then analyzedfuture climate and water use change impacts on a groundwater dependent socio-hydrological system,and developed an adaptation tool to increase resilience to seasonal drought (Figure 7.1).Data on streamflow and groundwater were sparse in the focus watersheds of this research, as is commonin many watersheds in developing countries. To address this need, I implemented a hydrological moni-toring network throughout the watersheds using innovative data loggers (Chapter 2). The low-cost dataloggers that were developed in this research can provide a powerful means for extending hydrologicalmonitoring networks, and their accessible and open-source nature shows potential for their integrationinto socio-hydrological and community-based applications.Based on the hydrological data gained from the fieldmonitoring (and extended by hydrological modelling)as well as water use data from dierent sectors, I undertook a combined assessment of societal wateruse and hydrological dynamics in these watersheds, and identified key vulnerabilities to drought (Chapter4). Domestic water extraction volumes in the Potrero and Caimital watersheds increased by 33%between 2005 and 2016. This is particularly of concern as the per capita water use was found tobe high in international comparison. Further, I found that domestic water demands were 18% higherduring the dry season compared to the wet season. In addition, agricultural water use via irrigation isconcentrated during the dry season, and accounted for almost 30% of total annual water extraction.This mismatch between water availability and water demand explains the primary dependence of thecommunities and agriculture on groundwater, which is essentially the only water supply during the dryseason. My analysis indicated that reliance on groundwater supplies is increasing in the Potrero and161Caimital watersheds from 77% groundwater use and 23% surface water use in 2005 to 86% groundwateruse in 2016. The high groundwater extraction rates and groundwater contributions to baseflow instreams led to seasonal fluctuation of groundwater levels (as indicated in observed groundwater levels).Groundwater levels typically increased with recharge during the wet season, and decreased towards theend of the dry season when community pumping wells often ran dry. Considering the high contributionsof rainfall to streamflow with mean annual runo coecients of 0.45 and 0.41 for the Caimital andPotrero river, respectively, this research highlighted a potential for increasing surface water use duringthe wet season. Yet, the high-frequency monitoring of streamflow also revealed the flashy nature ofthese streams with typically low baseflows even during the wet season. Most water runs o duringfast stormflow events which could make potential water extraction challenging. Evapotranspiration wasanother major water demand in the system, constituting on average 44% of total annual rainfall. In drieryears with reduced annual rainfall, however, total evapotranspiration could increase up to 54% of rainfall.Considering the high evapotranspiration rates, a socio-ecohydrological approach with inclusion of green(evapotranspirative) water fluxes for water resources management may be useful for the region.To improve water management and prepare for the future, better knowledge on the impacts of futureclimate variability on water resources is needed, especially on a watershed level that can be useful towater managers. However, translation of ENSO forecasts and climate change scenarios to impacts onwater resources was limited for Central America and often only existed at a regional level. In Chapter5, I provided one of the first quantification of potential El Niño/La Niña and climate change impacts onsurface water and groundwater resources on a watershed level for the region. I found that an extreme ElNiño conditions can reduce groundwater recharge and streamflow by ~60% relative to ENSO Neutral,which could lead to detrimental consequences for communities, agriculture and ecosystems. To assessa longer-term planning horizon towards the end of the 21st century and assess potential ‘worst-case’implications of climate change, I also analyzed a range of climate and population growth scenarios.Considering the complex climate of Guanacaste, I used a novel approach to model climate scenariosthat were specifically developed for Guanacaste through an expert elicitation protocol (Grossmann et al.2018). My results showed that climate change may lead to mean annual streamflow that is reduced by40-45% and groundwater recharge that is reduced by 26-28% in comparison to the historical baselines.Importantly, groundwater storage was found to decline under a drying climate with continued extraction.High population growth could further hasten potentially irreversible groundwater level declines.Groundwater recharge during the wet season is the key process for replenishing aquifers that communi-ties and agriculture rely upon during the dry season. However, in years of reduced rainfall during the wetseason, groundwater levels can fall below the level of many rural community pumping wells. Consideringthus the importance of groundwater recharge for water security, I further explored the relation betweengroundwater recharge and rainfall in more detail (Chapter 6). I found that total annual groundwaterrecharge can be predicted based on total cumulative rainfall to date. Based on this relationship, I162developed a novel ‘groundwater recharge indicator’ that can allow water managers to assess the likelyend-of-wet-season state of groundwater recharge early on during the wet season. This provides anopportunity for water managers and decision makers to react adaptively and prepare for a dry seasonwith reduced water availability before major impacts of drought hit, and importantly, can be used tocommunicate the need for adaptive responses to the public. The overall goal of adaptive responsesshould be to ‘bank’ groundwater during the remaining wet season for use during the upcoming dryseason when it is the only water supply. Adaptation strategies could also include supply management,such as the increased use of rain water and surface water, which is, as shown in Chapter 4, abundantduring the wet season. Considering the high per capita domestic water use and high agricultural irrigationvolumes (as also shown in Chapter 4), demand management could further lower the impacts on theaquifer. The eectiveness of specific adaptation strategies under dierent future change scenarios couldbe explored in future research.As is common with hydrological field and modelling studies, a number of limitations of this work shouldalso be noted. First of all, both hydrological field data and modelling are associated with uncertainties.Especially future change impacts are highly uncertain within the complex climate setting of Guanacaste,where a range of potential future trajectories of climate, population and potential socio-hydrologicalinteractions exist. Further, this research focused on climate and water extraction impacts, while landuse change may also play a role and could be investigated in future research. This research alsofocused on water quantity as a first step towards water security, while water quality is of importance aswell, especially in agriculturally-influenced watersheds, and future research could explore this aspect.Also, while this research explored social water use dynamics, there are, of course, a multitude ofother relationships between humans and water, some of which have been explored by previous watergovernance research in the region (e.g., Kuzdas et al. 2015b; Vignola et al. 2018; Kuzdas et al. 2015a).Future research could expand on socio-hydrological relations of, for instance, environmental awareness,emotional and cultural values regarding water, or psychological aspects of water use. I also discussedthe potential utility of combining ecohydrology with socio-hydrology towards a socio-ecohydrologicalapproach, and future research could expand on the ecohydrological dynamics in the region in moredetail. Further, while this research mentioned potential drought adaptation strategies through analysisof current and future vulnerabilities and impacts, and provided the groundwater recharge indicator tool,future research could expand on this and explore a range of specific adaptation strategies with thehydrological model and compare their eectiveness under dierent climate change scenarios.163Figure 7.1: Key concepts, research questions, methods and results of this research.In this research, I contributed to a number of scientific themes, ranging from development and deploy-ment of open-source data loggers, to a first holistic assessment of social and hydrological dynamicswith respect to water use and identification of emerging vulnerabilities to drought, to future climate andwater use impact assessments, to developing an adaptation tool that can support water managers inmaking their communities more resilient to seasonal drought. This research also provided one of thefirst combined analyses of surface water and groundwater for the region.Another important theme of this research was the socio-hydrological framing. Socio-hydrology is anemerging concept and this research provided a field-based example and exploration of social andhydrological drivers. Furthermore, considering high evapotranspirative water demands, and in line withrecent advances in ecohydrology that highlight the importance of including both blue (surface waterand groundwater) and green (evapotranspiration) water in water resources management, I also argued164for moving from socio-hydrology towards socio-ecohydrology to explicitly include these ecohydrologicalaspects into socio-hydrological analysis.A key component of this research was also its community-based approach, which was essential for itsfield application and current system assessment. The collaboration with stakeholders provided sharedlocal knowledge, and landowners of the monitoring stations developed stewardship towards the stations,thus ensuring the safety of the stations and continuity of data collection beyond the timeframe of thisthesis following transfer of the stations. The approach was also essential for developing the adaptationtool, which was informed by needs of the community. An important component of community-basedresearch is also communicating scientific results back to the community. As part of the Futuragua project,I participated in this communication through workshops held throughout the research period and at a finalworkshop, where research results were presented to a wide range of stakeholders, water managers, andthe general public, as well as followed by multiple discussion sessions and distribution of a handbookand information papers. Considering that most watersheds today are, on the one hand, influenced byhuman activities and, on the other hand, also increasingly exposed to hydrological extremes that impactpeople and their livelihoods, a community-based approach in combination with socio-ecohydrologicalsystem analysis may provide a way forward for scientific hydrological research.Using community-based monitoring and computer modelling, this thesis assessed the impacts of currentand future socio-hydrological drivers to inform adaptation to droughts and contribute to increasing thewater security of communities in two seasonally-dry watersheds in Costa Rica. This research can alsobe informative for many other similar seasonally-dry watersheds of the region. My research indicatedthat reduced rainfall under future climate change in combination with population growth and increasingwater demands may lead to groundwater storage declines. Thus, while communities currently primarilydepend on groundwater, a shift towards a more diverse set of water supplies, as for instance utilizing thehigh streamflows of the wet season or rainfall harvesting, may reduce pressure on groundwater. Highdomestic and agricultural water extraction rates also indicated potential for water conservation. Thegroundwater recharge indicator tool developed in this research can provide a means to trigger the adap-tive response to reduced rainfall to prepare for the oncoming dry season. Overall, this thesis provides aholistic socio-hydrological system assessment with respect to water use and novel approaches towardsmaking communities more resilient to drought and improving water security.165ReferencesACT. 2013. GIS Data for Nicoya/Hojancha. Area Conservación Tempisque, Nicoya. Data received in2013.Agudelo, Clara. 2006. Vulnerabilidad a la contaminación del acuífero Potrero-Caimital, Nicoya Gua-nacaste. Tech. rep. San José, Costa Rica: Servicio Nacional de Aguas subterraneas, riego y ave-namiento (SENARA).Aguilar, E., T. C. Peterson, P. Ramírez Obando, R. Frutos, J. A. Retana, M. Solera, J. Soley, I. GonzálezGarcía, R. M. Araujo, A. Rosa Santos, V. E. Valle, M. Brunet, L. Aguilar, L. Álvarez, M. Bautista,C. Castañón, L. Herrera, E. Ruano, J. J. Sinay, E. Sánchez, G. I. Hernández Oviedo, F. Obed, J. E.Salgado, J. L. Vázquez, M. Baca, M. Gutiérrez, C. Centella, J. Espinosa, D. Martínez, B. Olmedo,C. E. Ojeda Espinoza, R. Núñez, M. Haylock, H. Benavides, and R. Mayorga. 2005. Changes inprecipitation and temperature extremes in Central America and northern South America, 1961 to2003. Journal of Geophysical Research 110 (D23107): 1–15. doi:10.1029/2005JD006119.Allen, Richard G., L.S. Pereira, D. Raes, and M. Smith. 1998. Crop evapotranspiration - Guidelines forcomputing crop requirements. Irrigation and Drainage Paper No. 56, Food and Agriculture Organiza-tion (FAO) of the United Nations, no. 56: 300. doi:10.1016/j.eja.2010.12.001.AlMutairi, Bandar, Iris Grossmann, and Mitchell Small. 2018. Statistical Gaussian Mixture Approach toCharacterize Uncertainty in Bimodal Rainfall Cycle Projections. Under review: 1–47.Altman, Rachel M., Douw G Steyn, Nadya Moisseeva, Ofir Harari, andWilliam J. Welch. 2018. Statisticalmodelling of annual rainfall pattern in Guanacaste, Costa Rica. To be submitted.Amador, Jorge A. 1998. A Climatic Feature of the Tropical Americas: The Trade Wind Easterly Jet.Topicos Meteorologicos y Oceanograficos 5 (2): 91–102.Amato, Charlotte C., Daene C. Mckinney, Eusebio Ingol-Blanco, and Rebecca L. Teasley. 2006.WEAPHydrology Model Applied: The Rio Conchos Basin. Tech. rep. 1 - 69 p. The University of Texas atAustin: Center for Research in Water Resources, Bureau of Engineering Research.Arnold, J G, R Srinivasan, R S Muttiah, and J R Williams. 1998. Large area hydrologic modeling andassesment Part I: Model development. Journal of the American Water Resources Association 34 (1):73–89. doi:10.1111/j.1752-1688.1998.tb05961.x.166Asoka, Akarsh, Tom Gleeson, Yoshihide Wada, and Vimal Mishra. 2017. Relative contribution of mon-soon precipitation and pumping to changes in groundwater storage in India. Nature Geoscience 10(February): 109–119. doi:10.1038/NGEO2869.Ayantunde, A A, S Fernandez-Rivera, P H Hiernaux, H Van. Keulen, and H M J Udo. 2002. Day andnight grazing by cattle in the Sahel. Journal of Range Management 55 (2): 144–149. doi:10.2307/4003350.Babcock, Matthew, Gabrielle Wong-Parodi, Mitchell J. Small, and Iris Grossmann. 2016. Stakeholderperceptions of water systems and hydro-climate information in Guanacaste, Costa Rica. Earth Per-spectives 3 (3): 1–13. doi:10.1186/s40322-016-0035-x.Ballestero, Maureen, Virginia Reyes, and Yamileth Astorga. 2007. Groundwater in Central America: ItsImportance, Development and Use, with Particular Reference to its Role in Irrigated Agriculture. TCABInternational. The Agricultural Groundwater Revolution: Opportunities and Threats to Development:100–128.Beven, Keith. 2012. Rainfall-RunoModelling: The Primer. Second Ed. 457 p. Chichester, West Sussex; Hoboken, NJ: John Wiley & Sons Ltd. doi:10.1002/9781119951001.Birkel, Christian, Chris Soulsby, and Doerthe Tetzla. 2012. Modelling the impacts of land-cover changeon streamflow dynamics of a tropical rainforest headwater catchment. Hydrological Sciences Journal57 (8): 1543–1561. doi:10.1080/02626667.2012.728707.Biswas, Asit K. 1970. The History of Hydrology. 336 p. North-Holland Pub. Co., Amsterdam. AmericanElsevier Pub. Co., New York.Bitella, Giovanni, Roberta Rossi, Rocco Bochicchio, Michele Perniola, and Mariana Amato. 2014. ANovel Low-Cost Open-Hardware Platform for Monitoring Soil Water Content and Multiple Soil-Air-Vegetation Parameters. Sensors 14 (10): 19639–19659. doi:10.3390/s141019639.Blair, P., and W. Buytaert. 2016. Socio-hydrological modelling: A review asking "why, what and how?".Hydrology and Earth System Sciences 20 (1): 443–478. doi:10.5194/hess-20-443-2016.Blöschl, G, and M Sivapalan. 1995. Scale Issues in Hydrological Modelling: A Review. HydrologicalProcesses 9:251–290. doi:10.1002/hyp.3360090305.Blöschl, Günter. 2001. Scaling in hydrology. Hydrological Processes 15 (4): 709–711. doi:10.1002/hyp.432.Boiten, Wubbo. 2008. Hydrometry - A comprehensive introduction to the measurement of flow in openchannels. 3rd Ed. 247 p. CRCPress, Taylor / Francis Group, LLC. Leiden, Netherlands. : 04154676329780415467636.167Bojanowski, Jedrzej S. 2016. sirad: Functions for Calculating Daily Solar Radiation and Evapotranspira-tion. Visited on 10/01/2016. https://cran.r-project.org/web/packages/sirad/index.html.Brewer, Cynthia A. 2018. ColorBrewer. Visited on 01/05/2018. http://www.colorbrewer.org.Bronstert, A., J. Carrera, P. Kabat, and S. Lütkemeier. 2005. Coupled Models for the Hydrological Cycle- Integrating Atmosphere, Biosphere and Pedosphere. 345 p. Springer-Verlag Berlin Heidelberg. :3540223711.Buytaert, Wouter, Zed Zulkafli, Sam Grainger, Luis Acosta, Tilashwork C. Alemie, Johan Bastiaensen,Bert DeBievre, Jagat Bhusal, Julian Clark, Art Dewulf, Marc Foggin, David M. Hannah, Christian Her-garten, Aiganysh Isaeva, Timothy Karpouzoglou, Bhopal Pandeya, Deepak Paudel, Keshav Sharma,Tammo Steenhuis, Seifu Tilahun, Gert Van Hecken, and Munavar Zhumanova. 2014. Citizen sciencein hydrology and water resources: opportunities for knowledge generation, ecosystem service man-agement, and sustainable development. Frontiers in Earth Science 26 (October): 1–21. doi:10.3389/feart.2014.00026.Caballero, Luis A., Alon Rimmer, Zachary M. Easton, and Tammo S. Steenhuis. 2012. Rainfall RunoRelationships for a Cloud Forest Watershed in Central America: Implications for Water ResourceEngineering 1. Journal of the American Water Resources Association 48 (5): 1022–1031. doi:10.1111/j.1752-1688.2012.00668.x.Cadol, Daniel, Stephanie Kampf, and Ellen Wohl. 2012. Eects of evapotranspiration on baseflow ina tropical headwater catchment. Journal of Hydrology 462-463:4–14. doi:10.1016/j.jhydrol.2012.04.060.Carslaw, David C., and Karl Ropkins. 2012. openair — An R package for air quality data analysis.Environmental Modelling & Software 27-28:52–61. doi:10.1016/j.envsoft.2011.09.008.Castro, S M, G A Sanchez-Azofeifa, and H Sato. 2018. Eect of drought on productivity in a Costa Ricantropical dry forest. Environ. Res. Lett 13:1–14. 10.1088/1748-9326/aaacbc.Conrad, O, B Bechtel, M Bock, H Dietrich, E Fischer, L Gerlitz, J Wehberg, V Wichmann, and J Böhner.2015. System for Automated Geoscientific Analyses (SAGA) v. 2.1.4. Geoscientific Model Develop-ment 8:1991–2007. doi:10.5194/gmd-8-1991-2015.Crutzen, Paul J. 2006. The Anthropocene. In Earth system science in the anthropocene, ed. by EckartEhlers and Thomas Krat, 13–18. Springer Berlin Heidelberg New York. doi:10.1007/3-540-26590-2_3.CTI. 2017. Datos e indicadores claves para la gestion integrada del recurso hidrico (GIHR). Tech. rep.San José, Costa Rica: Comité Interinstitucional para las Estadísticas del Agua. Direccion de Agua,MINAE; Instituto de Estadistica y Censos (INEC).168Dee, D. P., M. Balmaseda, G. Balsamo, R. Engelen, A. J. Simmons, and J.-N.Thepaut. 2014. Towardsa consistent reanalysis of the climate system. Bulletim of American Meteorological Society 95:1235–1248. doi:10.1175/BAMS-D-13-00043.1.Denyer, P., T. Aguilar, and W. Montero. 2013a. Cartografia geologica de la peninsula de Nicoya, CostaRica - Mapa Geologico de la hoja Cerro Brujo. Direccion de Geologia y Minas, Costa Rica.– . 2013b. Cartografia geologica de la peninsula de Nicoya, Costa Rica - Mapa Geologico de la hojaMatambu. Direccion de Geologia y Minas, Costa Rica.Denyer, P, and P.O. Baumgartner. 2006. Emplacement of Jurassic-Lower Cretaceous radiolarites of theNicoya Complex (Costa Rica). Geologica Acta 4 (1): 203–218.Dermody, B. J., R. P H Van Beek, E. Meeks, K. Klein Goldewijk, W. Scheidel, Y. Van Der Velde, M. FP Bierkens, M. J. Wassen, and S. C. Dekker. 2014. A virtual water network of the Roman world.Hydrology and Earth System Sciences 18 (12): 5025–5040. doi:10.5194/hess-18-5025-2014.DHI. 2012. MIKE SHE User Manual. Tech. rep. Danish Hydrologic Institute, Hørsheim, Denmark.Di Baldassarre, G., A. Viglione, G. Carr, L. Kuil, J. L. Salinas, and G. Blöschl. 2013. Socio-hydrology:Conceptualising human-flood interactions. Hydrology and Earth System Sciences 17 (8): 3295–3303.doi:10.5194/hess-17-3295-2013.Dias, Lívia Cristina Pinto, Márcia N. Macedo, Marcos Heil Costa, Michael T. Coe, and Christopher Neill.2015. Eects of land cover change on evapotranspiration and streamflow of small catchments in theUpper Xingu River Basin, Central Brazil. Journal of Hydrology: Regional Studies 4 (PB): 108–122.doi:10.1016/j.ejrh.2015.05.010.Direccion de Agua. 2015. Reportes de concesiones. http://www.da.go.cr/reportes-d-a/.Doherty, John. 2015. Calibration and Uncertainty Analysis for Complex Environmental Models -PEST:complete theory and what it means for modelling the real world. 226 p. Watermark Numerical Com-puting, Brisbane, Australia.Dolman, A. Johannes, Diego G. Miralles, and Richard A.M. de Jeu. 2014. Fifty years since Monteith’s1965 seminal paper: The emergence of global ecohydrology. Ecohydrology 7 (3): 897–902. :19360592. doi:10.1002/eco.1505.Domenico, Patrick A., and Franklin W. Schwartz. 1997. Physical and chemical hydrogeology. 2nd Editio.528 p. John Wiley / Sons. : 978-0-471-59762-9.Doorenbos, J, and W.O. Pruitt. 1977.Guidelines for predicting crop water requirements. Tech. rep. Foodand Agriculture Organization of the United Nations, Rome.144 p.Ellison, David, Martyn N. Futter, and Kevin Bishop. 2012. On the forest cover-water yield debate: Fromdemand- to supply-side thinking. Global Change Biology 18 (3): 806–820. doi:10.1111/j.1365-2486.2011.02589.x.169Elshafei, Y., M. Sivapalan, M. Tonts, and M. R. Hipsey. 2014. A prototype framework for models of socio-hydrology: Identification of key feedback loops and parameterisation approach. Hydrology and EarthSystem Sciences 18 (6): 2141–2166. doi:10.5194/hess-18-2141-2014.Emerton, R., H. L. Cloke, E. M. Stephens, E. Zsoter, S. J. Woolnough, and F. Pappenberger. 2017.Complex picture for likelihood of ENSO-driven flood hazard. Nature Communications 8:1–9. doi:10.1038/ncomms14796.Esquivel-Hernández, Germain, Ricardo Sánchez-Murillo, Christian Birkel, and Jan Boll. 2017a. Climateand water conflicts coevolution from tropical development and hydro-climatic perspectives A casestudy of Costa Rica. Journal of the American Water Resources Association 54 (2): 451–470. doi:10.1111/1752-1688.12617.Esquivel-Hernández, Germain, Ricardo Sánchez-Murillo, Christian Birkel, Stephen P Good, and JanBoll. 2017b. Hydroclimatic and ecohydrological resistance/resilience conditions across tropical biomesof Costa Rica. Ecohydrology 10 (6): 1–12. doi:10.1002/eco.1860.Fabrega, J., T. Nakaegawa, R. Pinzón, K. Nakayama, O. Arakawa, and SOUSEI Theme-C ModelingGroup. 2013. Hydroclimate projections for Panama in the late 21st Century. Hydrological ResearchLetters 7 (2): 23–29. doi:10.3178/HRL.7.23.Falkenmark, M, and J Rockström. 2006. The New Blue and Green Water Paradigm Breaking NewGround for Water Resources Planning and Management. Journal of Water Resources Planning andManagement, no. June: 129–133.Falkenmark, Malin, and Carl Folke. 2002. The ethics of socio-ecohydrological catchment management:towards hydrosolidarity. Hydrology and Earth System Sciences Discussions 6 (1): 1–10. doi:10.5194/hess-6-1-2002.Famiglietti, J. S. 2014. The global groundwater crisis. Nature Climate Change 4 (11): 945–948. doi:10.1038/nclimate2425.FAO. 2016. El Niño and La Niña Preparedness and Response - Situation Report July 2016. Food,Agriculture Organization of the United Nations. Dominique Burgeon | Director, Emergency, and Re-habilitation Division |Strategic Programme Leader, Resilience, Rome, Italy.Feng, Xue, Amilcare Porporato, and Ignacio Rodriguez-Iturbe. 2013. Changes in rainfall seasonality inthe tropics. Nature Climate Change 3 (9): 811–815. doi:10.1038/nclimate1907.Fernandez-Sing, Mariana. 2009. Alternatives de abastecimiento de agua potable para la ciudad deNicoya. 247 p. Thesis Licenciatura, Facultad de Ingenieria, Universidad Fidelitas, San Jose, CostaRica.170Foglia, Laura, Iacopo Borsi, Steen Mehl, Giovanna De Filippis, Massimiliano Cannata, Enric Vasquez-Suñe, RotmanCriollo, andRudyRossetto. 2018. FREEWAT, a Free andOpen Source, GIS-Integrated,Hydrological Modeling Platform. Groundwater 56 (4): 521–523. doi:10.1111/gwat.12654.Folke, Carl. 2006. Resilience: The emergence of a perspective for social-ecological systems analyses.Global Environmental Change 16 (): 253–267. doi:10.1016/j.gloenvcha.2006.04.002.Folke, Carl, Reinette Biggs, Albert V. Norström, Belinda Reyers, and Johan Rockström. 2016. Social-ecological resilience and biosphere-based sustainability science.Ecology and Society 21 (3): 41 1–16.: 17083087. doi:10.5751/ES-08748-210341.Folke, Carl, ThomasHahn, Per Olsson, and JonNorberg. 2005. AdaptiveGovernance of Social-EcologicalSystems. Annual Review of Environment and Resources 30 (1): 441–473. doi:10.1146/annurev.energy.30.050504.144511.Freeze, R. Allen, and John A. Cherry. 1979. Groundwater. 604 p. Englewood Clis, N.J. : Prentice-Hall.García, Mario A., and Federico Arellano. 2012. Modelo Hidrogeologico conceptual del acuifero Nimboy-ores, Guanacaste, Costa Rica. Revista Geologica de America Central 47:143–161.Garcia, Pamela M., and Clara L. Agudelo. 2014. Modelo hidrogeologico conceptual del sistema acuiferoPotrero-Caimital, Nicoya, Guanacaste, Costa Rica. Ciencia y Tecnologia 30 (2): 35–56.Garcia-Serrano, Pamela. 2015. Determinacion de la disponibilidad del agua subterránea en el sistemaacuífero Potrero y Caimital, Nicoya, Guanacaste. 208 p. Thesis Maestria. Universidad de Costa Rica,Ciudad Universitaria Rodrigo Facio, Costa Rica.Gleick, Peter H. 2003. Water Use. Annual Review of Environment and Resources 28 (1): 275–314.doi:10.1146/annurev.energy.28.040202.122849.Gleick, Peter H, Newsha Ajami, Juliet Christian-Smith, Heather Cooley, Kristina Donnelly, Julian Fulton,Mai-Lan Ha, Matthew Heberger, Eli Moore, Jason Morrison, Stewart Orr, Peter Schulte, and VeenaSrinivasan. 2014. The World’s Water: The Biennial Report on Freshwater Resources. Data Table 2 -Freshwater Withdrawal by Country and Sector (2013 Update). 483 p. Pacific Institute for Studies inDevelopment, Environment, / Security. Island Press Washington D.C. : 9781610914819.Grolemund, Garrett, and Hadley Wickham. 2011. Dates and Times Made Easy with {lubridate}. JournalOf Statistical Software 40 (3): 1–25.Grossmann, Iris, Bandar AlMutairi, andMitchell J. Small. 2018. Scenarios of future precipitation changesin Northwest Costa Rica using expert assessment and model projections. In preparation.Guzman, Jose M. 2015. Informe sobre produccion de agua en Sistemas de la GAM (UEN de Aguapotable). Tech. rep. Instituto Nacional de Acueductos y Alcantarillados (AyA), San Jose, Costa Rica.GWP. 2000. Towards water security: Framework for Action. Tech. rep. Global Water Partnership, Stock-holm, Sweden.171Hall, Budd, Rajesh Tandon, and Crystal Tremblay. 2015. Strengthening Community University ResearchPartnerships: Global Perspectives. 331 p. University of Victoria, Victoria, Canada.United Nations Ed-ucational, Scientific / Cultural ORganization. : 978-1-55058-562-9.Hamel, Perrine, Diego Riveros-Iregui, Daniela Ballari, Trevor Browning, Rolando Célleri, David Chandler,Kwok Pan Chun, Georgia Destouni, Suzanne Jacobs, Scott Jasechko, Mark Johnson, Jagdish Krish-naswamy, María Poca, Patrícia Vieira Pompeu, and Humberto Rocha. 2017. Watershed services inthe humid tropics: Opportunities from recent advances in ecohydrology. Ecohydrology, no. July: 1–16.doi:10.1002/eco.1921.Hannah, Lee, Camila I. Donatti, Celia A. Harvey, Eric Alfaro, Daniel Andres Rodriguez, Claudia Bouron-cle, Edwin Castellanos, Freddy Diaz, Emily Fung, Hugo G. Hidalgo, Pablo Imbach, Peter Laederach,Jason P. Landrum, and Ana Lucía Solano. 2017. Regional modeling of climate change impacts onsmallholder agriculture and ecosystems in Central America. Climatic Change 141 (1): 29–45. doi:10.1007/s10584-016-1867-y.Harbaugh, ArlenW. 2005. MODFLOW-2005, The U.S. Geological SurveyModular Ground-Water Model-The Ground-Water Flow Process. U.S. Geological Survey Techniques and Methods.Healy, Richard W. 2010. Estimating groundwater recharge. 245 p. Cambridge University Press, Cam-bridge, United Kingdom / New York, NY, USA. doi:10.1017/CBO9780511780745.Hicks, S. D., A. K. Aufdenkampe, D. S. Montgomery, S. Damiano, and H. Brooks. 2015. A New ArduinoDatalogger Board for Simple, LowCost Environmental Monitoring and the EnviroDIYWebCommunity.Presented at 2015 Fall Meeting, American Geophysical Union, San Francisco, USA.Hidalgo, H. G., E. J. Alfaro, and B. Quesada-Montano. 2017. Observed (1970 - 1999) climate variabilityin Central America using a high-resolution meteorological dataset with implication to climate changestudies. Climatic Change 141 (1): 13–28. doi:10.1007/s10584-016-1786-y.Hidalgo, Hugo G., Jorge A. Amador, Eric J. Alfaro, and Beatriz Quesada. 2013. Hydrological climatechange projections for Central America. Journal of Hydrology 495, no. 1 (): 94–112. doi:10.1016/j.jhydrol.2013.05.004.Hidalgo, Hugo G., Ana M. Durán-Quesada, Jorge A. Amador, and Eric J. Alfaro. 2015a. The caribbeanlow-level jet, the inter-tropical convergence zone and precipitation patterns in the intra-Americas Sea:A proposed dynamical mechanism.Geografiska Annaler, Series A: Physical Geography 97 (1): 41–59.doi:10.1111/geoa.12085.Hidalgo, Hugo, Carolina Herrero, Eric J Alfaro, Ángel G. Muñoz, Natalie P Mora, Darner A Mora, andVíctor Chacón. 2015b. Urban Waters in Costa Rica. Urban Water Challenges in The Americas: 202–225 p.172Hoekstra, A, Joost Buurman, and Kees CH van Ginkel. 2018. Urban water security: A review. Environ-mental Research Letters 13 (053002): 1–14. doi:10.1088/1748-9326/aaba52.Holland, Margaret Buck, Sierra Zaid Shamer, Pablo Imbach, Juan Carlos Zamora, Claudia MedellinMoreno, Efraín J.Leguía Hidalgo, Camila I. Donatti, M. Ruth Martínez-Rodríguez, and Celia A. Har-vey. 2017. Mapping adaptive capacity and smallholder agriculture: applying expert knowledge at thelandscape scale. Climatic Change 141 (1): 139–153. doi:10.1007/s10584-016-1810-2.Hunter, Christian, Jorge Gironás, Diogo Bolster, and Christos A Karavitis. 2015. A Dynamic, Multivari-ate Sustainability Measure for Robust Analysis of Water Management under Climate and DemandUncertainty in an Arid Environment. Water 7 (1): 5928–5958. doi:10.3390/w7115928.Hut, R. 2013. New Observational Tools and Datasources for Hydrology. 104 p. Dissertation. TechnischeUniversiteit Delft, Netherlands.IIASA. 2018.SSPPublic Database (Shared Socioeconomic Pathways) - Version 1.1. Visited on 04/02/2018.https://tntcat.iiasa.ac.at/SspDb.Imbach, Pablo, Megan Beardsley, Claudia Bouroncle, Claudia Medellin, Peter Läderach, Hugo Hidalgo,Eric Alfaro, Jacob Van Etten, Robert Allan, Debbie Hemming, Roger Stone, Lee Hannah, and CamilaI. Donatti. 2017. Climate change, ecosystems and smallholder agriculture in Central America: anintroduction to the special issue.Climatic Change 141 (1): 1–12. doi:10.1007/s10584-017-1920-5.Imbach, Pablo, Sin Chan Chou, André Lyra, Daniela Rodrigues, Daniel Rodriguez, Dragan Latinovic,Gracielle Siqueira, Adan Silva, Lucas Garofolo, and Selena Georgiou. 2018. Future climate changescenarios in Central America at high spatial resolution.PloS one 13 (4): 1–21. doi:10.1371/journal.pone.0193570.Imbach, Pablo, LuisMolina, Bruno Locatelli, Olivier Roupsard, Gil Mahé, Ronald Neilson, Lenin Corrales,Marko Scholze, and Philippe Ciais. 2012. Modeling Potential Equilibrium States of Vegetation and Ter-restrial Water Cycle of Mesoamerica under Climate Change Scenarios. Journal of Hydrometeorology13 (2): 665–680. doi:10.1175/JHM-D-11-023.1.INEC. 2014. Estadísticas demográficas. 2011-2025. Proyecciones distritales. Población total por gruposde edades, según provincia, cantón, distrito y sexo. Tech. rep. Instituto Nacional de Estadistica yCencos. San Jose. Costa Rica.– . 2001. IX Censo Nacional de Población y V de Vivienda del 2000: Resultados Generales. Tech. rep.80 p. Instituto Nacional de Estadística y Censos. San Jose, Costa Rica.– . 2015. VI Censo Nacional Agropecuario: actividades pecuarias, prácticas y servicios agropecuarios.Tech. rep. 283 p. Instituto Nacional de Estadística y Censos. San Jose, Costa Rica.173– . 2011. X Censo Nacional de Población y VI de Vivienda: Resultados Generales. Tech. rep. San Jose,Costa Rica: 140 p. Instituto Nacional de Estadística y Censos.IPCC. 2014. Climate Change 2014: Synthesis Report. Contribution of Working Groups I, II and III to theFifth Assessment Report of the Intergovernmental Panel on Climate Change [Core Writing Team, R.K.Pachauri and L.A. Meyer (eds.)]. IPCC, Geneva, Switzerland: 1–151.Islam, Asraful, Taoufikul Islam, Minhaz Ahmed Syrus, and Nova Ahmed. 2014. Implementation of flashflood monitoring system based on wireless sensor network in Bangladesh. In 3rd international confer-ence on informatics, electronics & vision.Jasechko, Scott, Jean S. Birks, Tom Gleeson, Yoshihide Wada, Peter J. Fawcett, Zachary D. Sharp,Jerey J Mcdonnell, and Jerey M Welker. 2014. The pronounced seasonality of global ground-water recharge. Water Resources Research 50 (11): 8845–8867. doi:10.1002/2014WR015809.Received.Jasechko, Scott, and Richard G Taylor. 2015. Intensive rainfall recharges tropical groundwaters. Envi-ronmental Research Letters 10 (124015): 1–8. doi:10.1088/1748-9326/10/12/124015.Jepson, Wendy, Jessica Budds, Laura Eichelberger, Leila Harris, Emma Norman, Kathleen O’Reilly,Amber Pearson, Sameer Shah, Jamie Shinn, Chad Staddon, Justin Stoler, Amber Wutich, and SeraYoung. 2017. Advancing human capabilities for water security: A relational approach. Water Security1:46–52. doi:10.1016/j.wasec.2017.07.001.Jiménez-Rodríguez, César D., Julio C. Calvo-Alvarado, and John K. Jackson. 2015. Performance ofTwo Hydrological Models in Predicting Daily Flow under a Climate Change Scenario for MountainousCatchments in Northwestern Costa Rica. Mountain Research and Development 35 (3): 240–253.doi:10.1659/MRD-JOURNAL-D-14-00109.1.Karmalkar, Ambarish V., Raymond S. Bradley, and Henry F. Diaz. 2011. Climate change in Central Amer-ica and Mexico: regional climate model validation and climate change projections. Climate Dynamics37 (): 605–629. doi:10.1007/s00382-011-1099-9.Karp, Daniel S, Luke O Frishko, Alejandra Echeverri, Jim Zook, Pedro Ju, and Kai M A Chan. 2017.Agriculture erases climate-driven beta-diversity in Neotropical bird communities. Global Change Biol-ogy: 1–22. doi:10.1111/gcb.13821.KC, Samir, and Wolfgang Lutz. 2017. The human core of the shared socioeconomic pathways: Popu-lation scenarios by age, sex and level of education for all countries to 2100. Global EnvironmentalChange 42:181–192. doi:10.1016/j.gloenvcha.2014.06.004.Kiem, Anthony S., and Danielle C. Verdon-Kidd. 2011. Steps toward useful hydroclimatic scenarios forwater resource management in the Murray-Darling Basin. Water Resources Research 47 (): 1–14.doi:10.1029/2010WR009803.174Kuhn, Max. 2008. Building Predictive Models in R Using the caret Package. Journal Of Statistical Soft-ware 28 (5): 1–26. doi:10.1053/j.sodo.2009.03.002. =.Kuil, Linda, GemmaCarr, Alberto Viglione, Alexia Prskawetz, andGünter Blöschl. 2016. Conceptualizingsocio-hydrological drought processes: The case of the Maya collapse. Water Resources Research52:6222–6242. doi:10.1002/2014WR015716.Kuzdas, Christopher Paul. 2014. Toward Sustainable Governance of Water Resources; The Case ofGuanacaste, Costa Rica. PhD thesis, 450 p. Arizona State University.Kuzdas, Christopher, Benjamin Warner, Arnim Wiek, Mariel Yglesias, Raaele Vignola, and AlonsoRamírez-Cover. 2015a. Identifying the potential of governance regimes to aggravate or mitigate localwater conflicts in regions threatened by climate change. Local Environment 21 (11): 1387–1408.doi:10.1080/13549839.2015.1129604.Kuzdas, Christopher, Arnim Wiek, Benjamin Warner, Raaele Vignola, and Ricardo Morataya. 2015b.Integrated and participatory analysis of regional water governance regimes - The Case of the CostaRican Dry Tropics. World Development 66:254–268. doi:10.1016/j.worlddev.2014.08.018.Lathuillière, Michael J., Michael T. Coe, and Mark S. Johnson. 2016. A review of green-and blue-water resources and their trade-os for future agricultural production in the Amazon Basin: Whatcould irrigated agriculture mean for Amazonia?Hydrology and Earth System Sciences 20:2179–2194.doi:10.5194/hess-20-2179-2016.Lehner, B., K. Verdin, and A. Jarvis. 2008. New global hydrography derived from spaceborne elevationdata. EOS, Transactions, American Geophysical Union 89 (10): 93–94. doi:https://doi.org/10.1029/2008EO100001.Levy, Morgan C., Margaret Garcia, Peter Blair, Xi Chen, Sharlene L. Gomes, Drew B. Gower, JohannaGrames, Linda Kuil, Ye Liu, Landon Marston, Paul F. McCord, Mahendran Roobavannan, and RuijieZeng. 2016. Wicked but worth it: Student perspectives on socio-hydrology. Hydrological Processes30 (9): 1467–1472. doi:10.1002/hyp.10791.Lindsay, J B. 2016. Whitebox GAT: A case study in geomorphometric analysis. Computers and Geo-sciences 95:75–84. doi:10.1016/j.cageo.2016.07.003.Linton, Jamie, and Jessica Budds. 2014. The hydrosocial cycle: Defining and mobilizing a relational-dialectical approach to water. Geoforum 57:170–180. doi:10.1016/j.geoforum.2013.10.008.Lopez, Jess Christopher B, and Harreez M Villaruz. 2015. Low-Cost Weather Monitoring System withOnline Logging and Data Visualizaiton. December. 8th IEEE (The Institute of Electrical / ElctronicsEngineers Inc). Philippine Section.9-12 Dec, 2015, Cebu, Philippines. : 9781509003600.175Losilla, M., and C. Agudelo. 2003. Rendimiento seguro y susceptibilidad a la contaminacion del acuiferocoluvio aluvial de las cuencas altas de los rios Potrero y Caimital, Nicoya, Guanacaste, C.R. Tech. rep.1 - 70 p. Servicio Nacional de Aguas SubterraÌneas, Riego y Avenamiento (SENARA), San Jose,Costa Rica.Madrigal-Ballestero, Róger, and María A. Naranjo. 2015. Adaptive capacity, drought and the perfor-mance of community-based drinking water organizations in Costa Rica. Journal of Water and ClimateChange 6 (4): 831–847. doi:10.2166/wcc.2015.154.Magaña, Víctor, Jorge A Amador, and Socorro Medina. 1999. The Midsummer Drought over Mexico andCentral America. Journal of Climate 12:1577–1588. doi:10.1175/1520-0442(1999)012<1577:TMDOMA>2.0.CO;2.Magrin, G.O., J.A. Marengo, J.-P. Boulanger, M.S. Buckeridge, E. Castellanos, G. Poveda, F.R. Scarano,and S. Vicuña. 2014. Central and South America. In Climate change 2014: impacts, adaptation, andvulnerability. part b: regional aspects. contribution of working group ii to the fifth assessment report ofthe intergovernmental panel on climate change [barros, v.r., c.b. field, d.j. dokken, m.d. mastrandre,1499–1566. Cambridge University Press, Cambridge, United Kingdom / New York, NY, USA. doi:10.1017/CBO9781107415386.007.Maher, N, DMatei, and SMilinski. 2018. ENSO change in climate projections: forced response or internalvariability? Geophysical Research Letters In print. doi:10.1029/2018GL079764.Mao, Feng, Julian Clark, Timothy Karpouzoglou, Art Dewulf, Wouter Buytaert, and David Hannah. 2017.HESS Opinions: A conceptual framework for assessing socio-hydrological resilience under change.Hydrology and Earth System Sciences 21:3655–3670. doi:10.5194/hess-21-3655-2017.Markstrom, Steven L., Richard G. Niswonger, R. Steven Regan, David E. Prudic, and Paul M. Barlow.2008. GSFLOW - Coupled Ground-Water and Surface-Water Flow Model Based on the Integrationof the Precipitation-Runo Modeling System (PRMS) and the Modular Ground-Water Flow Model(MODFLOW-2005).Massuel, S., J. Riaux, F. Molle, M. Kuper, A. Ogilvie, A. L. Collard, C. Leduc, and O. Barreteau. 2018.Inspiring a Broader Socio-Hydrological Negotiation Approach With Interdisciplinary Field-Based Ex-perience. Water Resources Research 54:2510–2522. doi:10.1002/2017WR021691.Maurer, E. P., J. C. Adam, and A. W. Wood. 2009. Climate model based consensus on the hydrologicimpacts of climate change to the Rio Lempa basin of Central America. Hydrology and Earth SystemSciences 13:183–194. doi:10.5194/hess-13-183-2009.McDaniels, Tim, Matt Babcock, Kai M A Chan, Iris Grossmann, Silja Hund, Mark S Johnson, MathildeLarghi, Grégoire Leclerc, Paige Olmsted, Jennifer Romero Valpreda, Mitchell Small, Douw G Steyn,Raaele Vignola, and Gabrielle Wong-Parodi. 2014. FuturAgua - Enhancing Adaptation and Re-176silience to Drought in Dry Tropical Socio-Ecological Systems: The Guanacaste, Costa Rica Example.In American geophysical union conference, dec 15-19, 2014. San Francisco, U.S.McDaniels, Tim, Douw Steyn, Mark S. Johnson, Kai M Chan, Iris Grossmann, Mitchell Small, GabrieleWong-Parodi, Raaele Vignola, and Gregoire LeClerc. 2013. FuturAgua Research Project Proposal:1–51.Mehan, Sushant, Tian Guo, Margaret Gitau, and Dennis C. Flanagan. 2017. Comparative Study ofDierent StochasticWeather Generators for Long-TermClimate Data Simulation.Climate 5 (26): 1–40.doi:10.3390/cli5020026.Mileham, Lucinda, Richard G. Taylor, Martin Todd, Callist Tindimugaya, and Julian Thompson. 2009.The impact of climate change on groundwater recharge and runo in a humid, equatorial catchment:sensitivity of projections to rainfall intensity. Hydrological Sciences Journal 54 (): 727–738. doi:10.1623/hysj.54.4.727.Miller, Fiona, Henny Osbahr, Emily Boyd, Frank Thomalla, Sukaina Bharwani, Gina Ziervogel, BrianWalker, Jörn Birkmann, Sander Van der Leeuw, Johan Rockström, Jochen Hinkel, Tom Downing,Carl Folke, and Donald Nelson. 2010. Resilience and vulnerability: Complementary or conflictingconcepts? Ecology and Society 15 (3): 1–25.Milly, P C D, Julio Betancourt, Malin Falkenmark, Robert M Hirsch, W Zbigniew, Dennis P Lettenmaier,and Ronald J Stouer. 2008. Stationarity Is Dead: Whither Water Management? Science 319 (5863):573–574. doi:10.1126/science.1151915.Montanari, A., G. Young, H. H. G. Savenije, D. Hughes, T. Wagener, L. L. Ren, D. Koutsoyiannis, C.Cudennec, E. Toth, S. Grimaldi, G. Blöschl, M. Sivapalan, K. Beven, H. Gupta, M. Hipsey, B. Schaefli,B. Arheimer, E. Boegh, S.J. Schymanski, G. Di Baldassarre, B. Yu, P. Hubert, Y. Huang, A. Schumann,D. Post, V. Srinivasan, C. Harman, S. Thompson, M. Rogger, A. Viglione, H. McMillan, G. Characklis,Z. Pang, and V. Belyaev. 2013. Panta Rhei Everything Flows: Change in hydrology and society TheIAHS Scientific Decade 2013 to 2022. Hydrological Sciences Journal 58 (6): 1256–1275. doi:10.1080/02626667.2013.809088.Moore, R DDan. 2005. Slug Injection Using Salt in Solution.StreamlineWatershedManagement Bulletin8 (2): 1–28.Moore, R.D. Dan. 2014. R script for fitting a power-law rating curve. http://ibis.geog.ubc.ca/%7B~%7Drdmoore/rcode/RatingCurve.r.Morataya Montenegro, Ricardo. 2004. Plan de Manejo Sub Cuenca del Rio Potrero. Tech. rep. 167p. Nicoya, Costa Rica. Comisión para el Manejo de la Cuenca del Río Potrero. Asocioción para eldesarrollo sostenible del area de conservación Tempisque.177Moriasi, D. N., J. G. Arnold, M. W. Van Liew, R. L. Bingner, R. D. Harmel, and T. L. Veith. 2007.Model Evaluation Guidelines for Systematic Quantification of Accuracy in Watershed Simulations.Transactions of the American Society of Agricultural and Biological Engineers 50 (3): 885–900. doi:10.13031/2013.23153.Morillas, Laura, Silja V. Hund, and Mark S. Johnson. 2018. Ecohydrology and water use in an inten-sively managed agricultural landscape - double cropping of rainfed upland rice and irrigated melonsproduced under drought-prone tropical conditions. Under review.Moss, Richard H., Jae A. Edmonds, Kathy A. Hibbard, Martin R. Manning, Steven K. Rose, Detlef P.Van Vuuren, Timothy R. Carter, Seita Emori, Mikiko Kainuma, Tom Kram, Gerald A. Meehl, JohnF.B. Mitchell, Nebojsa Nakicenovic, Keywan Riahi, Steven J. Smith, Ronald J. Stouer, Allison M.Thomson, John P. Weyant, and Thomas J. Wilbanks. 2010. The next generation of scenarios forclimate change research and assessment. Nature 463:747–756. doi:10.1038/nature08823.Mostert, Erik. 2018. An alternative approach for socio-hydrology: case study research. Hydrology andEarth System Sciences 22:317–329. doi:10.5194/hess-2017-299.Mount, N. J., H. R. Maier, E. Toth, A. Elshorbagy, D. Solomatine, F. J. Chang, and R. J. Abrahart. 2016.Data-driven modelling approaches for socio-hydrology: Opportunities and challenges within the PantaRhei Science Plan. Hydrological Sciences Journal 61 (7): 1192–1208. doi:10.1080/02626667.2016.1159683.Nash, J. E., and J. V. Sutclie. 1970. River flow forecasting through conceptual models, Part I-A discus-sion of principles. Journal of Hydrology 10 (3): 282–290.Neelin, J. D., M. Munnich, H. Su, J. E. Meyerson, and C. E. Holloway. 2006. Tropical drying trends inglobal warming models and observations. Proceedings of the National Academy of Sciences 103 (16):6110–6115. doi:10.1073/pnas.0601798103.NOAA. 2018. El Niño Southern Oscillation (ENSO). Climate Prediction Centre of the National WeatherService, National Oceanic and Atmospheric Administration (NOAA). http://www.cpc.ncep.noaa.gov/products/precip/CWlink/MJO/enso.shtml.Ogden, Fred L., Trey D. Crouch, Robert F. Stallard, and Jeerson S. Hall. 2013. Eect of land coverand use on dry season river runo, runo eciency, and peak storm runo in the seasonal tropics ofCentral Panama. Water Resources Research 49 (12): 8443–8462. doi:10.1002/2013WR013956.Olaya, Victor. 2004. A gentle introduction to SAGAGIS. The SAGA User Group eV, Gottingen, Germany:216 p.Oldenborgh, Geert Jan van. 2018. KNMI Climate Explorer. Visited on 03/01/2018. https://climexp.knmi.nl/.178ORNL DAAC. 2008a. MODIS Collection 5 Land Products Global Subsetting and Visualization Tool.MOD15A2 product. Tech. rep. Oak Ridge, Tennessee, USA. Accessed August 01, 2017. Subsetobtained for MOD15A2 product at 10.0829N,85.4693W, time period: 2000-02-18 to 2017-01-01, andsubset size: 43 x 43 km. http://dx.doi.org/10.3334/ORNLDAAC/1241.– . 2008b. MODIS Collection 5 Land Products Global Subsetting and Visualization Tool. MOD16A2.Tech. rep. ORNL DAAC, Oak Ridge, Tennessee, USA. Accessed August 09, 2016. Subset obtainedfor MOD16A2 product at 10.0829N,85.4694W, time period: 2000-01-01 to 2014-12-27, and subsetsize: 21 x 21 km. http://dx.doi.org/10.3334/ORNLDAAC/1241.Owor, M, RGTaylor, C Tindimugaya, andDMwesigwa. 2009. Rainfall intensity and groundwater recharge:empirical evidence from the Upper Nile Basin. Environmental Research Letters 4 (): 1–6. doi:10.1088/1748-9326/4/3/035009.Pataki, D. E., C. G. Boone, T. S. Hogue, G. D. Jenerette, J. P. McFadden, and S. Pincetl. 2011. Socio-ecohydrology and the urban water challenge. Ecohydrology 4:341–347. doi:10.1002/eco.209.Paul, JonathanD.,Wouter Buytaert, Simon Allen, Juan A. Ballesteros-Cánovas, Jagat Bhusal, KatarzynaCieslik, Julian Clark, Sumit Dugar, David M. Hannah, Markus Stoel, Art Dewulf, Megh R. Dhital,Wei Liu, Janak Lal Nayaval, Bhanu Neupane, Arnulf Schiller, Paul J. Smith, and Robert Supper. 2017.Citizen science for hydrological risk reduction and resilience building. WIREs Water e1262:1–15.doi:10.1002/wat2.1262.Power, Scott, François Delage, Christine Chung, Greg Kociuba, and Kevin Keay. 2013. Robust twenty-first-century projections of El Niño and related precipitation variability. Nature 502 (7472): 541–545.doi:10.1038/nature12580.QGIS Development Team. 2017.QGIS Geographic Information System. Open Source Geospatial Foun-dation.Ramírez-Cover, Alonso. 2008. Conflictos socioambientales y recursos hídricos en Guanacaste ; unadescripción desde el cambio en el estilo de desarrollo ( 1997-2006 ) Introducción Naturaleza y cap-italismo : un breve marco analítico. Anuario de Estudios Centroamericanos, Universidad de CostaRica: 359–385.Rauscher, Sara A., Filippo Giorgi, Noah S. Dienbaugh, and Anji Seth. 2008. Extension and Intensifica-tion of the Meso-American mid-summer drought in the twenty-first century. Climate Dynamics 31 ():551–571. doi:10.1007/s00382-007-0359-1.Recha, John W., Johannes Lehmann, M. Todd Walter, Alice Pell, Louis Verchot, and Mark Johnson.2012. Stream Discharge in Tropical Headwater Catchments as a Result of Forest Clearing and SoilDegradation. Earth Interactions 16, no. 13 (): 1–18. doi:10.1175/2012EI000439.1.179Riahi, Keywan, Detlef P. van Vuuren, Elmar Kriegler, Jae Edmonds, Brian C. O’Neill, Shinichiro Fujimori,Nico Bauer, Katherine Calvin, Rob Dellink, Oliver Fricko, Wolfgang Lutz, Alexander Popp, Jesus Cre-spo Cuaresma, Samir KC, Marian Leimbach, Leiwen Jiang, Tom Kram, Shilpa Rao, Johannes Emmer-ling, Kristie Ebi, Tomoko Hasegawa, Petr Havlik, Florian Humpenöder, Lara Aleluia Da Silva, SteveSmith, Elke Stehfest, Valentina Bosetti, Jiyong Eom, David Gernaat, Toshihiko Masui, Joeri Rogelj,Jessica Strefler, Laurent Drouet, Volker Krey, Gunnar Luderer, Mathijs Harmsen, Kiyoshi Takahashi,Lavinia Baumstark, Jonathan C. Doelman, Mikiko Kainuma, Zbigniew Klimont, Giacomo Marangoni,Hermann Lotze-Campen, Michael Obersteiner, Andrzej Tabeau, and Massimo Tavoni. 2017. TheShared Socioeconomic Pathways and their energy, land use, and greenhouse gas emissions im-plications: An overview. Global Environmental Change 42:153–168. doi:10.1016/j.gloenvcha.2016.05.009.Romero Valpreda, Jennifer, Alejandra Echeverri, Silja V. Hund, Tim McDaniels, Laura Morillas, PaigeOlmsted, Douw G Steyn, Bandar AlMutairi, Matthew Babcock, Iris Grossmann, Raaele Vignola,Jeroen Houdijk, Mathilde Larghi, and Gregoire LeClerc. 2017. The FuturAgua Project: An interdis-ciplinary study of Climate Adaptation in Guanacaste, Costa Rica - Summary of Research Findings.Ed. by Jennifer Valpreda-Romero. 91 p. Department of Earth, Ocean / Atmospheric Sciences, Facultyof Science, The University of British Columbia.Roobavannan, Mahendran, Tim H.M. Van Emmerik, Yasmina Elshafei, Jaya Kandasamy, Matthew R.Sanderson, Saravanamuthu Vigneswaran, Saket Pande, and Murugesu Sivapalan. 2018. Norms andvalues in sociohydrological models.Hydrology and Earth SystemSciences 22 (2): 1337–1349. doi:10.5194/hess-22-1337-2018.Rossetto, Rudy, GiovannaDe Filippis, Iacopo Borsi, Laura Foglia, MassimilianoCannata, RotmanCriollo,and Enric Vázquez-Suñé. 2018. Integrating free and open source tools and distributed modellingcodes in GIS environment for data-based groundwater management. Environmental Modelling &Software 107:210–230. doi:10.1016/j.envsoft.2018.06.007.Saha, Suranjana, Shrinivas Moorthi, Hua Lu Pan, Xingren Wu, Jiande Wang, Sudhir Nadiga, PatrickTripp, Robert Kistler, John Woollen, David Behringer, Haixia Liu, Diane Stokes, Robert Grumbine,George Gayno, Jun Wang, Yu Tai Hou, Hui Ya Chuang, Hann Ming H Juang, Joe Sela, Mark Iredell,Russ Treadon, Daryl Kleist, Paul Van Delst, Dennis Keyser, John Derber, Michael Ek, Jesse Meng,Helin Wei, Rongqian Yang, Stephen Lord, Huug Van Den Dool, Arun Kumar, Wanqiu Wang, CraigLong, Muthuvel Chelliah, Yan Xue, Boyin Huang, Jae Kyung Schemm, Wesley Ebisuzaki, Roger Lin,Pingping Xie, Mingyue Chen, Shuntai Zhou, Wayne Higgins, Cheng Zhi Zou, Quanhua Liu, YongChen, Yong Han, Lidia Cucurull, Richard W. Reynolds, Glenn Rutledge, and Mitch Goldberg. 2010.The NCEP climate forecast system reanalysis. Bulletin of the American Meteorological Society 91 (8):1015–1057. doi:10.1175/2010BAMS3001.1.180Said, Rushdi. 1993. The river Nile: geology, hydrology, and utilization. 320 p. Oxford, England; New York,Pergamon.Sánchez-Murillo, R., A. M. Durán-Quesada, C. Birkel, G. Esquivel-Hernández, and J. Boll. 2016. Trop-ical precipitation anomalies and d-excess evolution during El Niño 2014-16. Hydrological Processes31:956–967. doi:10.1002/hyp.11088.Sánchez-Murillo, Ricardo, and Christian Birkel. 2016. Groundwater recharge mechanisms inferred fromisoscapes in a complex tropical mountainous region. Geophysical Research Letters 43:5060–5069.doi:10.1002/2016GL068888.Schewe, Jacob, Jens Heinke, Dieter Gerten, Ingjerd Haddeland, Nigel W. Arnell, Douglas B. Clark,Rutger Dankers, Stephanie Eisner, Balázs M. Fekete, Felipe J. Colón-González, Simon N. Gosling,Hyungjun Kim, Xingcai Liu, Yoshimitsu Masaki, Felix T. Portmann, Yusuke Satoh, Tobias Stacke,Qiuhong Tang, Yoshihide Wada, Dominik Wisser, Torsten Albrecht, Katja Frieler, Franziska Piontek,Lila Warszawski, and Pavel Kabat. 2014. Multimodel assessment of water scarcity under climatechange. Proceedings of the National Academy of Sciences 111 (9): 3245–3250. doi:10.1073/pnas.1222460110.Semenov, Mikhail A, and Elaine M Barrow. 2002. A Stochastic Weather Generator for Use in ClimateImpact Studies - User Manual. Hertfordshire, UK.Shuttleworth, W. James. 1993. Evaporation. In Handbook of hydrology, ed. by David R. Maidment,4.1–4.53. McGraw-Hill, Inc.Sieber, Jack, and Davide Purkey. 2015. Water Evaluation and Planing system (WEAP) User Guide.Sinton, C, R A Duncan, and P Denyer. 1997. Nicoya Peninsula, Costa Rica: A single suite of Caribbeanoceanic plateau magmas. Journal of Geophysical Research 102 (B7): 15507–15520. doi:10.1029/97JB00681.Sivapalan, M., H. H. G. Savenije, and G. Blöschl. 2012. Socio-hydrology: A new science of people andwater. Hydrological Processes 26 (8): 1270–1276. doi:10.1002/hyp.8426.Srinivasan, V. 2015. Reimagining the past use of counterfactual trajectories in socio-hydrological mod-elling: the case of Chennai, India. Hydrology and Earth System Sciences 19 (2): 785–801. doi:10.5194/hess-19-785-2015.Srinivasan, V., M. Konar, and M. Sivapalan. 2017. A dynamic framework for water security. WaterSecurity 1:12–20. doi:10.1016/j.wasec.2017.03.001.Steyn, Douw, NadyaMoisseeva, Ofir Harari, andWilliam J.Welch. 2016. Temporal and Spatial Variabilityof Annual Rainfall Patterns in Guanacaste, Costa Rica. The University of British Columbia Library,Vancouver, Canada. doi:10.14288/1.0340318.181Strand, K, S Marullo, N Cutforth, R Stoecker, and P Donohue. 2003. Principles of best practice forcommunity-based research. Michigan Journal of Community Service Learning 9 (3): 5–15.Swyngedouw, Erik. 2009. The Political Economy and Political Ecology of the Hydro-Social Cycle. Journalof Contemporary Water Research & Education 142:56–60. doi:10.1111/j.1936-704X.2009.00054.x.Taylor, Richard G., Martin C. Todd, Lister Kongola, Louise Maurice, Emmanuel Nahozya, Hosea Sanga,and Alan M. MacDonald. 2012. Evidence of the dependence of groundwater resources on extremerainfall in East Africa. Nature Climate Change 3:374–378. doi:10.1038/nclimate1731.Thompson, Lisa C, Marisa I Escobar, Christopher M Mosser, David R Purkey, David Yates, and Peter BMoyle. 2012. Water Management Adaptations to Prevent Loss of Spring-Run Chinook Salmon inCalifornia under Climate Change. Journal of Water Resources Planning and Management 138:465–478. doi:10.1061/(ASCE)WR.1943-5452.0000194.Thompson, S. E., M. Sivapalan, C. J. Harman, V. Srinivasan, M. R. Hipsey, P. Reed, A. Montanari, andG. Blöschl. 2013. Developing predictive insight into changing water systems: Use-inspired hydrologicscience for the anthropocene. Hydrology and Earth System Sciences 17 (12): 5013–5039. doi:10.5194/hess-17-5013-2013.Toohey, Ryan C., Jan Boll, Erin S. Brooks, and Jerey R. Jones. 2018. Eects of land use on soilproperties and hydrological processes at the point, plot, and catchment scale in volcanic soils nearTurrialba, Costa Rica. Geoderma 315:138–148. doi:10.1016/j.geoderma.2017.11.044.Troy, T. J., M. Konar, V. Srinivasan, and S. Thompson. 2015. Moving sociohydrology forward: A synthesisacross studies. Hydrology and Earth System Sciences 19 (8): 3667–3679. doi:10.5194/hess-19-3667-2015.UN/ISDR. 2009. Drought Risk Reduction Framework and Practices: contributing to the implementationof the Hyogo Framework for Action.United Nations secretariat of the International Strategy for DisasterReduction (UNISDR), Geneva, Switzerland: 213 pp.USDA-ARS and Texas A&M. 2017. Soil and Water Assessment Tool (SWAT).Van Emmerik, T. H.M., Z. Li, M. Sivapalan, S. Pande, J. Kandasamy, H. H.G. Savenije, A. Chanan,and S. Vigneswaran. 2014. Socio-hydrologic modeling to understand and mediate the competitionfor water between agriculture development and environmental health: Murrumbidgee River basin,Australia. Hydrology and Earth System Sciences 18:4239–4259. doi:10.5194/hess-18-4239-2014.182Van Loon, Anne F., Tom Gleeson, Julian Clark, Albert I.J.M. Van Dijk, Kerstin Stahl, Jamie Hannaford,Giuliano Di Baldassarre, Adriaan J. Teuling, Lena M. Tallaksen, Remko Uijlenhoet, David M. Hannah,Justin Sheeld, Mark Svoboda, Boud Verbeiren, ThorstenWagener, Sally Rangecroft, NikoWanders,and Henny A.J. Van Lanen. 2016a. Drought in the Anthropocene. Nature Geoscience 9 (2): 89–91.doi:10.1038/ngeo2646.Van Loon, Anne F., Kerstin Stahl, Giuliano Di Baldassarre, Julian Clark, Sally Rangecroft, NikoWanders,Tom Gleeson, Albert I J M Van Dijk, Lena M. Tallaksen, Jamie Hannaford, Remko Uijlenhoet, AdriaanJ. Teuling, David M. Hannah, Justin Sheeld, Mark Svoboda, Boud Verbeiren, Thorsten Wagener,and Henny A J Van Lanen. 2016b. Drought in a human-modified world: Reframing drought definitions,understanding, and analysis approaches. Hydrology and Earth System Sciences 20 (9): 3631–3650.doi:10.5194/hess-20-3631-2016.Van Meter, Kimberly J., Michael Stei, Daniel L. McLaughlin, and Nandita B. Basu. 2016. The socioeco-hydrology of rainwater harvesting in India: Understanding water storage and release dynamics acrossspatial scales. Hydrology and Earth System Sciences 20:2629–2647. doi:10.5194/hess- 20-2629-2016.Vecchi, Gabriel A, and Andrew T Wittenberg. 2010. El Nino and our future climate: where do we stand?Wiley Interdisc. Rev. Clim. Change 1:260–270. doi:10.1002/wcc.33.Vignola, Raaele, Chris Kuzdas, Isabel Bolaños, and Karina Poveda. 2018. Hybrid governance fordrought risk management: The case of the 2014/2015 El Niño in Costa Rica. International Journalof Disaster Risk Reduction 28:363–374. doi:10.1016/j.ijdrr.2018.03.011.Voisin, N., H. Li, D. Ward, M. Huang, M. Wigmosta, and L. R. Leung. 2013. On an improved sub-regionalwater resources management representation for integration into earth system models. Hydrology andEarth System Sciences 17:3605–3622. doi:10.5194/hess-17-3605-2013.Vörösmarty, Charles J, Pamela Green, Joseph Salisbury, Richard B Lammers, and Charles J Voros-marty. 2000. Global Water Resources: Vulnerability from Climate Change and Population Growth.Science 289 (5477): 284–288. doi:10.1126/science.289.5477.284.Wada, Y., L. P. H. van Beek, and M. F. P. Bierkens. 2011a. Modelling global water stress of the recentpast: on the relative importance of trends in water demand and climate variability.Hydrology and EarthSystem Sciences 15:3785–3808. doi:10.5194/hess-15-3785-2011.Wada, Yoshihide, and Marc F P Bierkens. 2014. Sustainability of global water use: past reconstructionand future projections. Environmental Research Letters 9:104003. doi:10.1088/1748-9326/9/10/104003.183Wada, Yoshihide, Marc F P Bierkens, Ad DeRoo, Paul A Dirmeyer, James S Famiglietti, Naota Hanasaki,Megan Konar, Junguo Liu, and Hannes Müller Schmied. 2017. Human-water interface in hydrologicalmodeling: Current status and future directions. Hydrology and Earth System Sciences 21:4169–4193.doi:10.5194/hess-2017-248.Wada, Yoshihide, Inge E. M. de Graaf, and Ludivicus P. H. van Beek. 2014. High-resolution modelingof human and climate impacts on global water resources. Journal of Advances in Modeling EarthSystems 8:735–763. doi:10.1002/2013MS000282.Received.Wada, Yoshihide, L. P.H. Van Beek, Daniel Viviroli, HansH. Drr, RolfWeingartner, andMarc F.P. Bierkens.2011b. Global monthly water stress: 2. Water demand and severity of water stress. Water ResourcesResearch 47 (W07518): 1–17. doi:10.1029/2010WR009792.Wagener, Thorsten, Murugesu Sivapalan, Peter a. Troch, Brian L. McGlynn, Ciaran J. Harman, Hoshin V.Gupta, Praveen Kumar, P. Suresh C Rao, Nandita B. Basu, and Jennifer S. Wilson. 2010. The futureof hydrology: An evolving science for a changing world. Water Resources Research 46 (W05301):1–10. doi:10.1029/2009WR008906.Warfe, D. M., N. E. Pettit, P. M. Davies, B. J. Pusey, S. K. Hamilton, M. J. Kennard, S. A. Townsend,P. Bayliss, D. P. Ward, M. M. Douglas, M. A. Burford, M. Finn, S. E. Bunn, and I. A. Halliday. 2011. The’wet-dry’ in the wet-dry tropics drives river ecosystem structure and processes in northern Australia.Freshwater Biology 56:2169–2195. doi:10.1111/j.1365-2427.2011.02660.x.Waylen, Peter R., andMarvin Quesada. 2002. The eect of Atlantic and Pacific sea surface temperatureson the mid-summer drought of Costa Rica. Environmental Change andWater Sustainability: 197–209.Waylen, Peter, César Caviedes, German Poveda Oscar Mesa, and Marvin Quesada. 1998. Rainfall Dis-tribution and Regime in Costa Rica and its Response to the El Niño-Southern Oscillation. Conferenceof Latin Americanist Geographers 24:75–84.Waylen, Peter, and M Sadi Laporte. 1999. Flooding and the El Niño phenomenon along the Pacific coastof Costa Rica. Hydrological Processes 13:2623–2638.Westerberg, I., J.-L. Guerrero, J. Seibert, K. J. Beven, and S. Halldin. 2011. Stage-discharge uncertaintyderived with a non-stationary rating curve in the Choluteca River, Honduras. Hydrological Processes25:603–613. doi:10.1002/hyp.7848.Westerberg, Ida. 2011. Observational uncertainties in water-resources modelling in Central America.PhD thesis, 76 p. Digital Comprehensive Summaries of Uppsala Dissertations from the Faculty ofScience and Technology 833. Uppsala Universitet, Sweden.Western, Andrew W., Rodger B. Grayson, and Günter Blöschl. 2002. Scaling of Soil Moisture: A Hy-drologic Perspective. Annual Review of Earth and Planetary Sciences 30:149–180. doi:10.1146/annurev.earth.30.091201.140434.184Wickert, A. D. 2014. The ALog: Inexpensive, Open-Source, Automated Data Collection in the Field.Bulletin of the Ecological Society of America. 95 (2): 166–176. doi:10.1890/0012-9623-95.2.68.Wickham, H. 2009. ggplot2: Elegant Graphics for Data Analysis. Springer-Verlag New York. : 978-0-387-98140-6. http://ggplot2.org.Wickham, Hadley. 2016. tidyr: Easily Tidy Data with ‘spread()‘ and ‘gather()‘ Functions.Winter, Thomas C, Judson W Harvey, O Lehn Franke, and William M Alley. 1998. Ground water andsurface water: A single resource. U.S. Geological Survey, U.S. Department of the Interior.Wohl, E., A. Barros, N. Brunsell, N. Chappell, M. Coe, T. Giambelluca, S. Goldsmith, R. Harmon, J. M. H.Hendrickx, J. Juvik, J. McDonnell, and F. Ogden. 2012. The hydrology of the humid tropics. NatureClimate Change 2 (9): 655–662. doi:10.1038/nclimate1556.Wolf, Sebastian, Werner Eugster, Sonja Majorek, and Nina Buchmann. 2011. Aorestation of tropicalpasture only marginally aects ecosystem-scale evapotranspiration. Ecosystems 14 (8): 1264–1275.: 14329840. doi:10.1007/s10021-011-9478-y.Wright, Cynthia, Aurora Kagawa-Viviani, Cynthia Gerlein-Safdi, Giovanny M. Mosquera, María Poca,Han Tseng, and Kwok Pan Chun. 2018. Advancing ecohydrology in the changing tropics: Perspectivesfrom early career scientists. Ecohydrology 11 (3): 1–18. doi:10.1002/eco.1918.Wutich, Amber, Jessica Budds, Laura Eichelberger, Jo Geere, Leila M. Harris, Jennifer A. Horney,Wendy Jepson, Emma Norman, Kathleen O’Reilly, Amber L. Pearson, Sameer H. Shah, Jamie Shinn,Karen Simpson, Chad Staddon, Justin Stoler, Manuel P. Teodoro, and Sera L. Young. 2017. Ad-vancing methods for research on household water insecurity: Studying entitlements and capabilities,socio-cultural dynamics, and political processes, institutions and governance.Water Security 2:1–10.doi:10.1016/j.wasec.2017.09.001.WWAP. 2012. Managing Water under Uncertainty and Risk. UN Water Reports. World Water Assess-ment Programme 1:909.Yates, David, David Purkey, Jack Sieber, Annette Huber-lee, Hector Galbraith, Jordan West, SusanHerrod-julius, Chuck Young, Brian Joyce, and Mohammad Rayej. 2009. Climate Driven Water Re-sources Model of the Sacramento Basin , California. Journal of Water Resources Planning and Man-agement 135 (5): 303–313. doi:10.1061/?ASCE?0733-9496?2009?135:5?303?CE.Yates, David, Jack Sieber, David Purkey, and Annette Huber-Lee. 2005. WEAP21âA Demand-, Priority-,and Preference-Driven Water Planning Model. Part 1: Model Characteristics. Water International 30(4): 487–500. doi:10.1080/02508060508691893.Zambrano-Bigiarini, Mauricio. 2017a. hydroGOF: Goodness-of-fit functions for comparison of simulatedand observed hydrological time series. Tech. rep. R package version 0.3-10. http://hzambran.github.io/hydroGOF/.doi:10.5281/zenodo.840087.185– . 2017b. hydroTSM: Time Series Management, Analysis and Interpolation for Hydrological Modelling.Zebiak, Stephen E., Ben Orlove, Ángel G. Muñoz, Catherine Vaughan, James Hansen, Tara Troy,Madeleine C. Thomson, Allyza Lustig, and Samantha Garvin. 2015. Investigating El Niño-SouthernOscillation and society relationships. WIREs Climate Change 6:17–34. doi:10.1002/wcc.294.186Appendix AChapter 2 supporting informationA.1 Field site photosFigure A.1: Downstream Potrero River site (SW1) at baseflow. The PVC tube containing the CTD sensoris on the left river bank.187Figure A.2: Downstream Potrero River site (SW1) during low stormflows. The PVC tube containing theCTD sensor is on the left river bank.Figure A.3: Downstream Potrero River site (SW1) during high stormflows when the channel was almostfilled (View from upstream bridge, monitoring site is on the right river bank).188Figure A.4: Tributary to Potrero River site (SW2). Monitoring site is at far right river bank (hidden byvegetation).Figure A.5: Data logger at Tributary to Potrero River site (SW2). Family home can be seen inbackground.189Figure A.6: Upstream Potrero River site (SW3). Monitoring stations is on the far left river bank.Figure A.7: Upstream Potrero River site (SW3) with PVC tube for monitoring sensor.190Figure A.8: Upstream Potrero River site (SW3) data logger.Figure A.9: Upstream Caimital River site (SW4).191Figure A.10: Upstream Caimital River site (SW4) solar panel and surrounding rice fields. Stream withsensor and data logger is within riparian vegetation to the right.Figure A.11: Upstream Caimital River site (SW4) data logger.192Figure A.12: Downstream Caimital River site (SW5) during baseflow. Monitoring station (PVC tube forsensor) is on the left river bank. During high stormflows, the river bank can fill up. Therefore, the datalogger is located on top of the left river bank.Figure A.13: Downstream Caimital River site (SW5) during low stormflow.193Figure A.14: Downstream Caimital River site (SW5) data logger, next to landowner’s house.Figure A.15: Varillal Groundwater Monitoring site (GW1). Pipe is connecting pump to nearby household.194Figure A.16: Dulce Nombre Groundwater Monitoring site (GW2). During monitoring, the well is covered.Figure A.17: Dulce Nombre Groundwater Monitoring site (GW2) with CTD sensor cable.195Figure A.18: Gamalotal Groundwater Monitoring site (GW3). The well is located within the orchard of asmallholder farmer.196Figure A.19: Irrigated melon fields during the dry season. Note the dry hillside with seasonally-dry forest.Figure A.20: Rainfed rice during the wet season.197A.2 Stream monitoring stationsTable A.1: Start and end dates of stream monitoring stations. Monitoring is continuing for most stations,operated by Universidad Nacional (UNA) in Liberia.SW1 SW2 SW3 SW4 SW52014-03-25 2014-03-28 2013-12-07 2014-08-14 2014-03-292016-12-31 2016-12-31 2016-12-31 2016-04-27 2016-12-31Table A.2: Start and end dates of groundwater monitoring stations. Monitoring is continuing for moststations, operated by Universidad Nacional (UNA) in Liberia.GW1 GW2 GW32013-12-07 2014-04-02 2013-03-312016-12-31 2016-12-31 2016-12-31Table A.3: Latitude and longitude of monitoring stations.ID Location Latitude LongitudeSW1 Casitas 10°07’27.6" N 85°25’43.4" WSW2 Varillal 10°05’47.5" N 85°27’33.6" WSW3 Costeña 10°05’25.7" N 85°27’56.8" WSW4 Costeña 10°04’38.6" N 85°28’51.5" WSW5 La Virginia 10°02’51.9" N 85°32’28.5" WGW1 Varillal 10°05’46.8" N 85°27’33.3" WGW2 Dulce Nombre 10°05’10.8" N 85°28’55.4" WGW3 Gamalotal 10°03’25.1" N 85°29’59.4" WEC station Varillal/Caimital 10°04’58.4" N 85°28’09.9" W198Figure A.21: Cross sections of the channel geometry at the locations of the streammonitoring sites. Thevertical line indicates location of CTD sensor in perforated PVC tube. SW2 monitoring was discardeddue to channel erosion, see section below.A.3 Recorded stage dataThe following section describes the data recording at the five stream monitoring sites, including issuesexperienced in the field. The time series of recorded stage data are provided.The monitoring station at the Downstream Potrero River Site (SW1) recorded relatively continuouslythroughout the study period (Figure A.22), but some field issues were experienced. In April 2014 duringa storm, falling branches from riparian trees interrupted the power supply line from solar panel to datalogger leading to data outages. Further, starting in November 2014, the first version of the Arduino datalogger exhibited problems leading to occasional data outages until the beginning of April 2015, whenthe data logger was replaced with the second version of the Arduino data logger (the Ecohydro Logger).Between the monthly field visits of July and August 2016, a sensor connection cable became loose aftera storm and data outages occurred. During the dry season of 2015 and 2016, the house owners nearthe monitoring station operated a surface water pump. This pump drew water from the Potrero River inclose vicinity of the monitoring station and aected water depth measurements (as can be seen in the199stage data presented in Figure A.22). The periods of pumping had thus to be removed before furtherprocessing of the stage data to discharge.Figure A.22: 30-minute averages of observed stage (water depth) at the Downstream Potrero River site(SW1).The Tributary to Potrero River Station (SW2) in Varillal experienced repeated erosion of the river bankduring storm events, and consequent dislocation of the sensor tube. The erosion changed the crosssectional area continuously, changing the stage – discharge relationship that is essential for monitoring.It also changed the height of the sensor above the thalweg in the stream. Due to the data discontinuitiesand changes in the stage – discharge relation, the data from this station had to be discarded and werenot further processed.The Upstream Potrero River Site (SW3; Figure A.23) was one of the earliest monitoring stationsinstalled in December 2013. Power supply issues occurred with the first version of the Arduino datalogger from fall 2014 until spring 2015, when the newly developed Ecohydro Logger was installed toreplace the first logger version. During a storm in the fall of 2015, the cable connecting sensor and datalogger was damaged by tree fall in the riparian vegetation, and the sensor was replaced in early 2016.200Figure A.23: 30-minute averages of observed stage (water depth) at the Upstream Potrero River site(SW3).At the Upstream Caimital River Site (SW4; Figure A.24), similar to the other stations, power issueswere experienced in fall 2014 with the first Arduino data logger version, which was replaced in spring2015 with the new Ecohydro Logger. Water levels in the stream dropped during the dry season of 2016and the stream was dry for some time in March 2016, as was confirmed during field visits (the drop ofstage is visible in Figure A.24). In April 2016, with the start of the wet season, a major storm caused thetree to which the sensor tube was axed to fall into the river. The riverbank was eroded and the crosssection to which the rating curve had been developed was completely changed. Thus, monitoring at thisstation was discontinued afterwards.201Figure A.24: 30-minute averages of observed stage (water depth) at the Upstream Caimital River site(SW4). Note that the stream ran dry in spring 2016. Stage started rising again with first rainfall, until astorm in April 2016 damaged the station.The Downstream Caimital River Site (SW5; Figure A.25) experienced issues in 2014, due to a brokensensor after a major storm and power failure, followed by issues with the version 1 of the data logger.In spring 2015, the new Ecohydro Logger was installed and the station recorded continuously until July2016, when the sensor broke again during a storm, and was replaced as soon as access allowed.202Figure A.25: 30-minute averages of observed stage (water depth) at the Downstream Caimital Riversite (SW5).A.4 Rating curve developmentTo develop location-specific relationships between measurements of stream stage and stream dis-charge, I conducted a series of discrete discharge measurements at each monitoring site with thesalt solution slug injection method (Moore 2005). In this method, a known mass of salt is dissolvedin solution and injected into the stream where it mixes with stream water. Electrical conductivity isrecorded downstream of the injection site. Injection locations and downstream electrical conductivitymeasurement locations are chosen with a sucient separation to ensure that the tracer has mixedcompletely across the stream width. High-frequency recordings of electrical conductivity show a peakwith a leading and trailing edge of lower concentration due to the stream-wide dispersion (for example,Figure A.26). The electrical conductivity breakthrough curve can be used to compute discharge using theprinciple of conservation of mass. A calibration is conducted to relate the change in electrical conductivityin stream water to the known addition of the salt solution. Incremental volumes of the salt solution areadded to a known volume of stream water, and from this, a calibration constant is calculated.203Figure A.26: Example of discharge measurements at SW1 on 2015-10-23. Top row shows thethree calibrations, with the relative concentration of the tracer in the stream and measured electricalconductivity in stream water of known volume. The calibration coecient k represents the slope ofthe line between relative concentration and measured electrical conductivity. The bottom row showsthe electrical conductivity peaks of the three salt solution slug injections, measured downstream of theinjection site in five second intervals. Conductivity units are given in micro-Siemens per centimeter(mS/cm). Electrical conductivities were measured using conductivity sensors GS3 by METER GroupInc. (formerly Decagon Devices Inc.) .I conducted all discharge measurements during the wet season (between July and November) of 2014and 2015 using Arduino data loggers and the GS3 electrical conductivity sensors to measure the elec-trical conductivity downstream of the injection site. For each discrete discharge determination, threereplicate measurements and calibrations were made. The mean of the three discharge measurementswas used as input for the rating curve for conditions when water level had stayed constant throughoutthe three measurements. During high streamflows, often only one discharge measurement per waterdepth was possible as water depth and discharge was changing rapidly.The biggest challenge for the discharge measurements and rating curve development was measure-ments at high streamflows. With the convective rainfall regime in the region, most of the rainfall eventsoccur in the early evening and throughout the night. The response of streamflow to the rainfall eventsis flashy and often happens within hours of the rainfall event. This meant that streamflows were alreadyfalling and often back to baseflow conditions by the time it was possible to go to the field in the earlymorning. Thus, opportunities to measure peak streamflows were limited. Furthermore, the streams canhave high flows with the entire river cross section filling up and water flowing fast, when both securing aGS3 sensor in the stream as well as adding the salt solution to the stream can be dicult.Considering the limited data on peak streamflows, I also estimated discharge using the Manning’sEquation (Slope-Area method; Equation A.1) (Boiten 2008):204Q =1n R23  S12  A (A.1)where Q⇥m3s1⇤ is stream discharge, n⇥m3s1⇤ is the Manning’s roughness coecient, R [m] isthe hydraulic radius A/P with A⇥m2⇤as wetted area of cross section at stage and P [m] as wettedperimeter of the cross section at stage, and S[] is the energy gradient (slope) of river reach.Estimating streamflow using the slope-area method is approximate and less accurate than actual mea-surements. Yet, it can help to constrain the higher values of the exponential rating curve. Inputs forthe Manning Equation include river cross section and slope over river reach, which were measured inthe field, and the Manning coecient that represents roughness on the streambed. I used a Manningcoecient of n = 0.04⇥m3s1⇤ for natural minor streams of winding nature, which best representedthe natural conditions at the monitored streams (Boiten 2008).Based on measured and estimated discharge and related stage values, the stage-discharge relation(rating curve) was then modeled as a power function (Equation A.2; Boiten 2008):Q =   (h + c)b (A.2)whereQ⇥m3s1⇤ is stream discharge, h [m] is stage,  ⇥m2s1⇤ and b [] are constants, and c [m]is a datum correction.Observed discharge and stage data were log-transformed, and then fitted to the log-transformed powerlaw, using nonlinear least-square estimates of the parameters following Moore (2014). Using the ratingcurve developed for each station, the 30-minute mean stage measurements were transformed into30-minute mean discharge, and then aggregated to daily mean discharge. Details on rating curvedevelopments for each monitoring station follow in the sections below.Downstream Potrero River site (SW1)For the SW1 monitoring station at the Potrero River, ten discharge measurements were available acrossa range from low to higher discharges (Table A.4). For the two highest flows, only one dischargemeasurement per water depth could be conducted, as water levels were falling. For these two highflow measurements, it was possible to add the salt solution to the middle of the stream as there wasa bridge upstream of the monitoring site (despite not being able to enter the river bed due to the deepand fast flowing waters). Considering that high streamflow measurements were available, no Manningestimates needed to be used.205Table A.4: SW1 discharge in liters per second and stage measurements in millimeters. Discharge 1 to3 indicate the three separate discharge measurements. The mean discharge and stage were used asinput for rating curve development.Date Stage DischargemeanDischarge 1 Discharge 2 Discharge 3[m]⇥m3s1⇤ ⇥m3s1⇤ ⇥m3s1⇤ ⇥m3s1⇤1 2014-07-14 0.084 0.011 0.012 0.011 0.0112 2014-08-12 0.181 0.046 0.046 0.045 0.0473 2014-09-22 0.193 0.261 0.256 0.268 0.2584 2015-10-23 0.247 0.392 0.397 0.382 0.3985 2015-10-21 0.261 0.410 0.378 0.467 0.3856 2014-09-24 0.266 0.453 0.428 0.485 0.4477 2014-09-17 0.278 0.567 0.464 0.6698 2014-09-29 0.472 1.397 1.470 1.3259 2015-10-24 0.548 2.946 2.94610 2015-10-24 0.712 6.466 6.466A nonlinear regression model was fitted to the ten, log-transformed, stage-discharge measurements anda rating curve was developed (Figure A.27). Parameters for the SW1 rating curve are provided in TableA.5. Discharge is estimated using these parameters and observed stage values in Equation A.2,Q = 10( + b og10 (h  c)) (A.3)where Q⇥m3s1⇤ is stream discharge,  ⇥sm3⇤, b ⇥sm4⇤ and c [m] are constants, and h [m] isstage.The overall coecient of determination between observed discharge and discharge values estimatedthrough the rating curve was R2 = 0.99.Table A.5: Downstream Potrero River site (SW1) rating curve parameter estimates for the terms inEquation A.3, standard error, t-Statistic and p-value.Parameter Estimate Standard error t-statistic p-valuea⇥sm3⇤1.21 0.19 6.28 0.0004b⇥sm4⇤2.65 0.59 4.47 0.0029c [m] 0.02 0.04 0.57 0.5868206Figure A.27: Downstream Potrero River site (SW1) rating curve (black line) and stage-discharge fieldmeasurements (red diamonds). The overall coecient of determination between observed dischargeand rating curve estimated discharge was R2 = 0.99.Upstream Potrero River site (SW3)For the Upstream Potrero River site (SW3), I had seven field discharge measurements available (TableA.6). However, most of the discharge measurements were relatively low, and therefore I also estimateddischarge for three higher stages using the Manning equation (Equation A.1). For the Manning equation,the slope over the river reach of 32 m at the location of the monitoring station was measured in the fieldas 0.3%. The station is located in the relatively flat alluvial valley and therefore the slope is low.When looking at the cross section of this monitoring station (Figure A.28a), a clear break in cross sectionprofile occurs at about 0.73 m of stage, and for higher stages, near-linear change in cross section profileled to a linear increase in Manning estimated discharge. Therefore, I decided to use a composite ratingcurve for this monitoring site, where stage below the stage threshold is transformed to discharge via arating curve based on observed discharge values, and stage above the threshold is calculated using aManning-based transformation.207Table A.6: Upstream Potrero River site (SW3) discharge measurements in liters per second andrelated stage measurements in millimeters. Discharge 1 to 3 indicate the three separate dischargemeasurements. Discharge numbers 1 to 7 indicate field measurements, and numbers 8, 9 and 10indicate Manning estimates.# Date Stage DischargemeanDischarge 1 Discharge 2 Discharge 3[m]⇥m3s1⇤ ⇥m3s1⇤ ⇥m3s1⇤ ⇥m3s1⇤1 2014-06-27 0.396 0.008 0.008 0.008 0.0082 2014-09-16 0.415 0.023 0.021 0.024 0.0233 2014-08-21 0.427 0.026 0.025 0.025 0.0274 2014-10-10 0.428 0.027 0.029 0.0245 2014-09-26 0.48 0.029 0.027 0.0326 2015-10-26 0.689 0.065 0.060 0.071 0.0497 2015-11-05 0.726 0.156 0.173 0.121 0.1768 Manning estimate 1 9.9269 Manning estimate 1.25 20.31710 Manning estimate 1.5 33.629The rating curve for the lower stage values was based on the seven discharge measurements (TableA.6), and the power law (Equation A.6) was used. Parameter estimates are given in Table A.7; noc-coecient was used for development of this rating curve as the sensor was installed against thestreambed and therefore no datum correction was necessary. The coecient of determination betweenobserved and estimated discharge was R2 = 0.82.Table A.7: UpstreamPotrero River Site (SW3) rating curve parameter estimates for the terms in EquationA.3, standard error, t-Statistic and p-value. This curve is fitted to the observed discharge measurementsonly, and is used to transform measured stage below 0.73 meters.Parameter Estimate Standard error t-statistic p-valuea⇥sm3⇤-0.47 0.23 -2.06 0.094b⇥sm4⇤3.36 0.70 4.80 0.005Considering the near-linear relation between stage and theManning estimated discharge values, a linearmodel (Equation A.4) was fitted to the Manning estimates and the two highest measured discharge val-ues (#6 to 10 in Table A.6). Parameter values are given in Table 10, and the coecient of determinationfor the linear model is R2 = 0.99. The linear model is given below,Q =  + b  h (A.4)whereQ⇥m3s1⇤ is stream discharge, ⇥m3s1⇤ and b⇥m2s1⇤ are parameters, and h [m] is stage.208Table A.8: Upstream Potrero River Site (SW3) linear model parameter estimates for the terms inEquation A.4, standard error, t-Statistic and p-value. This function is used to transform measured stageabove 0.73 meters.Parameter Estimate Standard error t-statistic p-valuea⇥m3s1⇤-29.7 2.7 -11.1 0.0016b⇥m2s1⇤41.2 2.5 16.5 0.0005Figure A.28: Upstream Potrero River site (SW3) rating curve (black line), stage-discharge field mea-surements (red diamonds) and Manning estimates (blue circle). (a) SW3 Cross section with indicationof stage threshold. (b) Rating curve based on discharge measurements. The overall coecient ofdetermination between observed discharge and rating curve estimated discharge was R2 = 0.82.(c)Composite rating curve, for stage < threshold of 0.73 based on rating curve from (b), and for stage >threshold based on linear rating model based on Manning Estimates. The coecient of determinationfor linear Manning model was R2 = 0.99.Upstream Caimital River site (SW4)For the Upstream Caimital River Site (SW4), only four, relatively low discharge measurements wereavailable (Table A.9) due to the dicult access to this site after intense storm events when water wouldbe standing high in the rice fields that needed to be crossed to reach the site. Therefore, similarly to theSW3 site, three Manning estimates were used to supplement the measured discharge values for highstreamflows. No slope measurements were available for SW4, and the measured slope of the SW3 site209was assumed, considering that they are in close proximity and both are small creeks in the centre ofthe flat alluvial valley. Also, similarly to SW3 rating curve development, I developed a composite ratingcurve, as Manning estimates increased nearly linearly for higher values (see also the cross sectionprofile, Figure A.29a). In this rating curve, stage values below a threshold of 0.53 m are estimatedusing a rating curve based on the four observed discharge values (for low flows), and higher flows arecalculated from a rating curve developed with the Manning estimates (A.29b and c).Table A.9: Upstream Caimital River site (SW4) discharge in liters per second and stage measurementsin millimeters. Discharge 1 to 3 indicate the three separate discharge measurements. The dischargemean and stage were used as input for rating curve development. Discharge values 5, 6 and 7 indicatethe Manning estimates.# Date Stage DischargemeanDischarge 1 Discharge 2 Discharge 3[m]⇥m3s1⇤ ⇥m3s1⇤ ⇥m3s1⇤ ⇥m3s1⇤1 2014-06-26 0.373 0.005 0.006 0.005 0.0052 2014-09-25 0.41 0.012 0.012 0.012 0.0123 2015-10-25 0.456 0.015 0.015 0.016 0.0164 2015-11-01 0.459 0.026 0.026 0.026 0.0255 Manning Estimate 1 1.6526 Manning Estimate 1.5 4.6357 Manning Estimate 2 8.159The rating curve for the lower stage values was based on the four discharge measurements, usingthe power law equation (Equation A.3, parameter estimates in Table A.10); no c-coecient was usedfor development of this rating curve as sensor was installed against the streambed and so no datumcorrection was necessary. The coecient of determination between observed and estimated dischargewas R2 = 0.78.Table A.10: Upstream Caimital River site (SW4) power law rating curve parameter estimates for theterms in Equation A.3, standard error, t-Statistic and p-value. This curve is fitted to the observeddischarge measurements only, and is used to transform measured stage below 0.53 meters.Parameter Estimate Standard error t-statistic p-valuea⇥sm3⇤-0.47 0.23 -2.06 0.094b⇥sm4⇤3.36 0.70 4.80 0.005Considering the near-linear relation between stage and theManning estimated discharge values, a linearmodel (Equation A.4) was fitted to the Manning estimates and the highest measured discharge value(#4 to 7 in Table A.9). Parameter values for the linear Model are given in Table A.11, and the coecientof determination is R2 = 0.97 for the linear model.210Table A.11: Upstream Caimital River site (SW4) linear model parameter estimates for the terms inEquation A.4, standard error, t-Statistic and p-value. This function is used to transform measured stageabove 0.53 meters.Parameter Estimate Standard error t-statistic p-valuea⇥m3s1⇤-29.7