UBC Faculty Research and Publications

Temporal and Spatial Variability of Annual Rainfall Patterns in Guanacaste, Costa Rica Steyn, Douw; Moisseeva, Nadya; Harari, Ofir; Welch, William J. Nov 23, 2016

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

Item Metadata


52383-Steyn_D_et_al_Temporal_Spatial_Variability.pdf [ 5.79MB ]
JSON: 52383-1.0340318.json
JSON-LD: 52383-1.0340318-ld.json
RDF/XML (Pretty): 52383-1.0340318-rdf.xml
RDF/JSON: 52383-1.0340318-rdf.json
Turtle: 52383-1.0340318-turtle.txt
N-Triples: 52383-1.0340318-rdf-ntriples.txt
Original Record: 52383-1.0340318-source.json
Full Text

Full Text

Temporal and Spatial Variability of Annual Rainfall Patterns inGuanacaste, Costa RicaDouw Steyn and Nadya Moisseeva∗, Ofir Harari†, and William J. Welch‡November 23, 2016AbstractWe analyze a body of rainfall data covering 38 years from five meteorological stations inthe Nicoya Peninsula of the Guanacaste Province, Costa Rica. The purpose of the analysisis to uncover spatial and temporal variability of rainfall in order to support research intowater and sustainability under the FuturAgua project. We use a variety of statisticalanalysis and modelling techniques. The analysis uncovers a relatively suppressed spatialpattern of rainfall. Rainfall totals for periods shorter than two weeks are dominated bystrong stochastic variability, while longer totalizing periods reveal systematic variation.Monthly totals show the strong double peak, and associated midsummer drought thathas been previously reported. The annual cycle can be efficiently captured by a doubleGaussian model. A simple application of this model to individual years shows large inter-annual variability, and a strong dependence of the second rainfall peak on the Oceanic Nin˜oIndex (ONI). A Bayesian analysis confirms the appropriateness of the double Gaussianmodel, and quantifies the strength of the dependence on ONI. We discuss the implicationsof our statistical analyses for research under the FuturAgua project.1 IntroductionThe Pacific coast of Central America (including the Guanacaste province of Costa Rica)is an interesting example of a region experiencing a tropical wet-dry climate. The Nicoyapeninsula of Guanacaste (see Figure 1) experiences an extremely dry season from roughlySeptember to May, while the rainy season is punctuated by a mild mid-summer drought(MSD) locally called La Canicula, or Il Veranillo di San Juan. The Pacific coast ofCentral America is the only tropical region away from the equator which experiences thisdouble maximum of rainfall (Magan˜a et al., 1999). The marked annual cycle of rainfallhas profound influences on many aspects of the environment and human life, includingagriculture, electricity generation and tourism. The FuturAgua research project www.∗Department of Earth, Ocean and Atmospheric Sciences, The University of British Columbia, Vancouver,BC, Canada†Department of Statistics and Actuarial Science, Simon Fraser University, Burnaby, BC, Canada‡Department of Statistics, The University of British Columbia, Vancouver, BC, Canada1futuragua.ca is directed towards understanding water-availability, -use and -governancein the Nicoya peninsula.As rainfall variability is the primary driver of the water balance, a detailed character-ization of the temporal variability of rainfall on a variety of time scales over the Nicoyapeninsula is fundamental to FuturAgua. Long-term (multi-decadal) rainfall trends drivenby climate change will be important to water managers in the region, and inter annualvariability of wet- and dry-seasons as well as the MSD must be well-understood as muchof water management is driven by variability on a seasonal scale. Short-term rainfallvariability driving floods or droughts are important too, but will not be addressed in thiswork due to the nature of the available data. While the underlying physical mechanismsgoverning the annual wet/dry rainfall pattern, and the imbedded MSD are not directlyrelevant to FuturAgua, a general understanding of the mechanisms will serve the projectby placing the annual pattern in a climatological context, and might provide some under-standing of possible changes in rainfall accompanying climate change. The overall intentof this report is to provide a statistically defensible (but not necessarily fully robust) de-scription of rainfall variability in the FuturAgua study region as a starting point uponwhich the FuturAgua investigative team can base their research into water-balances, -use,-management and -decision making.Statistical climatological studies of mechanisms, forcing and variability underlying theannual cycle of rainfall in the wet-dry tropics of Central America have been undertakenusing station-, satellite- and reanalysis-data and model output. Waylen et al. (1996a)and Waylen et al. (1996b) analyzed rainfall in Costa Rica and confirmed earlier resultsof Portig (1976) that the annual total rainfall is related to El Nin˜o-Southern Oscillation(ENSO), with dry years generally corresponding to low Southern Oscillation Index (SOI)values, also called warm phases or El Nio. In a similar study, Enfield and Alfaro (1999)concluded that, in addition to ENSO effects, the relative sign of sea surface temperature(SST) anomalies in the Eastern Pacific and tropical North Atlantic Oceans have an effecton rainfall in Costa Rica. They do note the low levels of explained variance in theirstatistical analyses, and poorly understood aspects of the underlying mechanisms.Pena and Douglas (2002) noted a link between Central America rainfall, and cross-equatorial flow and trade winds. Karnauskas et al. (2013) point out that this region abutsthe east Pacific warm pool (EPWP), and is the rainiest place on earth in the Boreal sum-mer. Their analysis shows statistical links between mature phase El Nio conditions, EPWPwarm anomalies, enhanced eastern Pacific ITCZ and consequently enhanced rainfall overCentral America. These analyses show that there are multiple mechanisms underlyingrainfall in the wet/dry tropics, and not surprisingly that statistical associations are weak.The MSD, as a particular, local feature of the annual cycle of rainfall in the wet-drytropics of Central America has been studied in parallel with the cycle itself. Magan˜a et al.(1999) reject the idea that the MSD is associated with the double crossing of the intertropical convergence zone (ITCZ) over Central America. In agreement with Waylen et al.(1996b), they postulate that the MSD is driven by an intensification of the northeast tradewinds in July and August with subsequent subsidence downwind of the spine of CentralAmerica resulting in a suppression of rainfall. The noted trade wind intensification isrelated to fluctuations in strength and position of the eastern Pacific ITCZ. Waylen andQuesada (2002) further elaborate on a mechanism relating the MSD to SST contrastsbetween Pacific and tropical North Atlantic Oceans. Warmer Atlantic leading to less severeMSD. Curtis (2004) employed a satellite-derived rainfall data set and reanalysis winds to2study the diurnal cycle of rainfall in the MSD region of Central America. His analysisshows that the diurnal cycle of rainfall during the MSD has reduced evening rainfall,compared to that in non-MSD months. The local circulations driving this diurnal cycle areshown to be part of the trade wind system, thus linking the MSD to seasonal variability ofthe trades. Rauscher et al. (2008) investigated possible responses of the MSD to climatechange using model output from the Coupled Model Intercomparison Project phase 3.Their analysis suggests a future early onset and increased intensity of the MSD. Theyemphasize that mechanisms underlying the MSD are yet to be fully understood. Smallet al. (2007) examine a combination of local and remote forcings driving the MSD usingsatellite observations, reanalysis data and a linear baroclinic model. They conclude thatthe May-June early peak is part of the southern branch of the North American Monsoon.The MSD is then driven by northward movement of the ITCZ and an associated westwardextension of the Atlantic subtropical high which drives divergence and subsidence overCentral America. The August-September rainfall peak is coincident with the northwardlimit of the Eastern Pacific ITCZ and peak in Atlantic and Caribbean SST. Recently,Karnauskas et al. (2013) suggest a simple mechanism for MSD based on the biannualcrossing of solar declination in Central America. They posit that two peaks in convectiveinstability produce the two rainfall peaks, and thus reject the idea of a rainfall suppressionmechanism, in favour of two instances of a single precipitation enhancing mechanism. Theydo acknowledge the influence of the variety of remote processes invoked in earlier analyses.Most recently, Maldonado et al. (2016) quantified inter-annual variability of the MSD inthe entire Central American Region, and show that the MSD intensity and magnitudeshow a negative relationship with Nio 3.4 and a positive relationship with the Caribbeanlow-level jet. They also show that the MSD is dependent on sea surface temperatureanomalies in the Pacific, Tropical North Atlantic and Caribbean waters.These studies show that the MSD is a persistent climatological feature of the southwestcoast of Central America, and is well simulated by a variety of global climate models. Itis probably governed by a complex interplay of local and remote processes, includingENSO, ITCZ migration, the North American Monsoon, the North Atlantic subtropicalhigh, tropical low level jets, regional SST, local convection and topographic forcing. Notsurprisingly, statistical climatological analyses fail to uncover a simple dominant governingmechanism.The general objectives of this report are to provide a defensible statistical model ofannual rainfall cycle in the Guanacaste Peninsula of Costa Rica. The rainfall model will beused to drive hydrological models for use in the FuturAgua project, and to inform publicstakeholder engagement efforts. The model must include a statistical representation ofinter annual variability in the form of variances of appropriately time-averaged rainfall,long term trends over multiple decades, co-variability with known large-scale atmosphere-ocean forcing (such as ENSO), and should give some indication of dependence on changingclimates over a span of decades. Given the multiple mechanisms underlying MSD, it islikely that statistical modelling will have to employ particularly robust techniques. Themodel will be based on available station rainfall data.2 Guanacaste Annual Rainfall CycleThe purposes of this section are to describe data used in the analysis and to present apreliminary, simple statistical analysis designed to illustrate the annual cycle of rainfall in3Guanacaste, referring specifically to wet- and dry seasons and the MSD.2.1 Instituto Meteorolgico Nacional (IMN) Rainfall DataThe longest possible sequences of rainfall data as daily totals for as many stations aspossible were sought from IMN. After reviewing all possible stations, data from Nicoya,Paquera, Garza, La Guinea and Liberia were selected for analysis (see Figure 1 for stationlocations and Table 1 for station details). In order to facilitate inter station analysis,missing data were filled, where possible. Single missing points were filled using simplelinear interpolation. Missing blocks of data were identified and filled using time-averageddata from five surrounding years (˘ 2 years). Missing blocks in terminal segments of datafor each station were filled using time-averaged data from three preceding years. Figure 2and Figure 3 show raw (unfilled) and filled data sequences to illustrate the amount offilling needed. As can be seen, data gaps at Paquera, Garza and La Guinea were deemedtoo long to be filled. Filled data sequences will now be subjected to simple, exploratorystatistical analysis.Table 1: Nicoya Peninsula Meteorological StationsStation IMN Coordinates Elevation PeriodName Identifier N W m ASLNicoya 72101 10˝9102 85˝27102 120 1980-2013Paquera 72111 9˝491192 84˝561172 15 1980-2013Garza 72135 9˝57102 85˝36102 10 1977-2012La Guinea 74006 10˝251102 86˝281242 40 1977-2006Liberia 74020 10˝36102 85˝32102 80 1980-20132.2 Inter-station CorrelationsA simple view of spatial rainfall structure across the study domain can be derived frominter-station correlation analyses. In order to avoid the influence of data gaps in the filledsequences, inter-station correlation was tested in subgroups for limited overlapping timeperiods as follows:• Group 1: Nicoya, Paquera, La Guinea and Liberia were for years 1980-2004• Group 2: Nicoya, Garza and Libera for years 1997 - 2014Daily and weekly correlation coefficient matrices show that rainfall amount is onlyweakly correlated between the stations on these shorter time scales. This is not surpris-ing, given that most rainfall in this region is convective and hence highly intermittentand spatially discontinuous. Biweekly total rainfall amounts show stronger inter-stationcorrelation, and monthly totals show high correlation, as seen in Tables 2 and 3 for groups1 and 2 respectively.4Figure 1: Location map of the FuturAgua study domain showing topography and the five meteorologicalstations.51977 1981 1985 1989 1993 1997 2001 2005 2009 2013Nicoya1977 1981 1985 1989 1993 1997 2001 2005 2009 2013Paquera1977 1981 1985 1989 1993 1997 2001 2005 2009 2013Liberia1977 1981 1985 1989 1993 1997 2001 2005 2009 2013LaGuinea1977 1981 1985 1989 1993 1997 2001 2005 2009 2013GarzaFigure 2: Raw (unfilled) Guanacaste IMN rainfall data coverage available for this study by station.1977 1981 1985 1989 1993 1997 2001 2005 2009 2013Nicoya1977 1981 1985 1989 1993 1997 2001 2005 2009 2013Paquera1977 1981 1985 1989 1993 1997 2001 2005 2009 2013Liberia1977 1981 1985 1989 1993 1997 2001 2005 2009 2013LaGuinea1977 1981 1985 1989 1993 1997 2001 2005 2009 2013GarzaFigure 3: Filled Guanacaste IMN rainfall data coverage available for this study by station.6Jan Feb Mar Apr May Jun Jul Aug Sept Oct Nov Dec0100200300400500600700Rainfall [mm]Figure 4: Annual cycle of Nicoya Peninsula composite rainfall. Plotted data are composite rainfall from 1977to 2013. Boxes extend from the lower to upper quartile values of the data, with a line at the median. Thewhiskers extend from the box to show the range of the data. Outlier points are those outside 1.5 times theinterquartile range. The green line is Equation 1 fitted to monthly median rainfall with fitting parameters α= 0.0 mm, A1 = 295 mm, A2 = 403 mm, l1 = 5.27 months, l2 = 8.93 months, φ1 = 1.41 months, φ2 = 1.66months.Table 2: Group 1: Monthly correlation coefficients - 1980-2004Station Name Nicoya Paquera La Guinea LiberiaNicoya 1. 0.79 0.82 0.85Paquera 0.79 1. 0.77 0.78La Guinea 0.82 0.77 1. 0.92Liberia 0.85 0.78 0.92 1.Table 3: Group 2 : Monthly correlation coefficients - 1997-2014Station Name Garza Nicoya LiberiaGarza 1. 0.83 0.79Nicoya 0.83 1. 0.80Liberia 0.79 0.80 1.Given the strong spatial inter-station correlation for monthly total rainfall, and onlyweak correlation for shorter totals, we construct a regional rainfall time series by takingaverages of filled monthly totals over stations for which data are available. The annualcycle of rainfall in this composite data series is represented in Figure 4, which clearlyshows wet and dry seasons and the MSD. These composite data will be used for allfurther analyses of the annual rainfall pattern, and MSD.2.3 Modelling the Annual Rainfall Cycle with a Double-GaussianThe annual cycle of rainfall with double peaks shown in Figure 4, is highly suggestive oftwo overlapping Gaussian functions separated widely enough to leave a distinct minimumbetween the peaks. Accordingly, we modelled the cycle by fitting a double Gaussianfunction for a range of sampling frequencies including weekly, biweekly and monthly totals.Initial attempts to fit a double Gaussian to data from single years using various nonlinear7optimization procedures produced variable results, mostly because of the difficulty infitting 7 parameters to 12 data points. However, using varying totalization periods showsthat the double-peak structure is clearly seen for monthly or longer totals. In order toavoid the optimization problem, a Markov Chain-Monte Carlo approach was adopted.As this approach employs data from all years, inter-annual variability was addressed byincluding the Ocean Nin˜o Index (ONI) as a covariate.The double Gaussian model assumes the form:µit “ δ `A1 exp"´pt´ l1q2φ1*`A2 exp"´pt´ l2q2φ2*, (1)where µit is the station mean total rainfall for year i and month t “ 1, . . . , 12. A1 “α1βxit1 and A2 “ α2βxit2 are amplitudes of the two Gaussian shaped rainfall peaks, wherexit is ONI for year i and month t. The effect of ONI is allowed to be different for thetwo peaks through β1 and β2. A log-normal error distribution is assumed, where observedrainfall yit for year i and month t is modelled as generated bylogpYit ` 1q „ Nplogpµitq, σ2q,where log is base Prior distributionsIndependent priors are:1. αj „ Gammap0.08, 0.0002q (mean 400, vague)2. φj „ Gammap2, 1q (mean 2, standard deviation?2)3. l1 „ Uniformp4, 7.5q4. l2 „ Uniformp7.5, 11q5. βj „ Gammap0.001, 0.001q (mean 1, vague)The prior distributions are shown by the solid red lines in Figure Posterior distributionsDetails of the model fitting procedure are found in Harari et al. (2016). The fitted modelis based on 1980-2006 data, using average data from the three stations (Nicoya, Liberiaand La Guniea) that have the longest, near-continuous data record. The parameters δ,α1, α2, β1, β2, l1, l2, φ1, φ2, σ are estimated via Bayes’ Rule, and the resultant posteriordistributions shown by the dashed black lines in Figure 5. Summary statistics are inTable 4.α1 α2 β1 β2 l1 l2 φ1 φ2 δ σMean 285 411 0.79 0.80 5.95 9.23 1.52 1.89 2.56 0.40SD 39 50 0.13 0.08 0.13 0.12 0.27 0.22 0.34 0.022.5% limit 217 324 0.56 0.65 5.72 8.99 1.10 1.50 1.95 0.3697.5% limit 367 518 1.07 0.96 6.24 9.46 2.13 2.36 3.26 0.44Table 4: Mean, standard deviation (SD), and 95% credible intervals (2.5% limit and 97.5% limit) of theposterior distributions.8100 400 7000.0000.006α1Density100 400 7000.0000.0040.008α2Density0.2 0.6 1.0β1Density0.2 0.6 1.0 1.4012345β2Density5 6 7 8 6 7 8 1 2 3 4φ1Density0 1 2 3 4φ2Density1 2 3 4δDensity0.2 0.4 0.6 0.805101520σDensityFigure 5: Distributions for the model parameters: prior (solid red line) and posterior (dashed black line).9January April July October0100200300400500rainfall (mm)ENSO positive (18)ENSO negative (18)Figure 6: Monthly average rainfall for ENSO positive and negative phases The ENSO neutral category has notbeen included as it comprises only one year. ENSO positive and negative categories consist of 18 years each.These results allow us to draw a number of conclusions:1. The 95% credible interval for β2 (second peak) is r0.65, 0.96s; i.e., there is a largeprobability that β2 ă 1. Taking the mean value of 0.80 implies that the amplitudeof the second peak, and hence rainfall in all months related to the second-peak, ischanged by multiplicative factors of β´12 “ 125%, β02 “ 100%, and β12 “ 80% whenONI is -1 C, 0 C, and +1 C, respectively.2. The 95% credible interval for β1 (first peak) is r0.56, 1.07s. A large range of β1 valuesis consistent with the data, including β1 “ 1 (no ONI effect) and β1 ! 1 (large ONIeffect).3. The credible intervals for β1 and β2 are consistent in the sense that they substantiallyoverlap. In this sense they do not establish that the ONI effect is different for the twopeaks. On the other hand, the ONI effect is statistically significant for the secondpeak; the evidence is weaker for the first peak.2.4 Effects of ENSO and Temporal TrendsFitted double Gaussian for composite averages of monthly rainfall in ENSO positive andENSO negative years are shown in Figure 6, which reveals that the annual rainfall patternsare effectively the same for both ENSO phases, but that the total rainfall during ENSOnegative years is substantially greater then that in ENSO negative years, and that thedifference is particularly large for the fall rainfall maximum.Simple trend analysis reveals a statistically significant upward trend in the fall maxi-mum rainfall intensity, which corresponds to the peak magnitude of the second Gaussian10Figure 7: Temporal trend from 1977 to 2014 in fall maximum rainfall. Red, black and blue symbols correspondto ENSO positive, neutral and negative phases respectively.as shown in Figure 6. The preponderance of ENSO positive/negative points below/abovethe linear trend line reflect the ENSO dependence seen in Figure 6. An investigation oftrends within each ENSO category reveals a similar rising trend in the fall maximum forENSO negative years, but no corresponding trend for ENSO positive years.Contrary to the rising trend in fall rainfall, local wisdom in the Guanacaste peninsula,as expressed by farmers and water managers is that rainfall has been decreasing. In orderto further investigate this apparent contradiction, we accessed annual total rainfall datafrom the Nicoya meteorological record from the Global Historical Climatology Network-Monthly data set http://www.ncdc.noaa.gov/ghcnm/. The IMN data for the Nicoyaagricultural extension station were appended to these data to form a 50 year time seriesof annual total rainfall at Nicoya. Figure 7 shows that annual total rainfall at this station(and presumably over the entire Guanacaste peninsula) decreased from 1950 to about themid 1990s, and thereafter increased, albeit with huge scatter around the overall trend. Itmay well be that human memory is dominated by the low annual totals in the last fiveyears.111950 1960 1970 1980 1990 2000 2010year500100015002000250030003500Annual Rainfall Total (mm)201520142013201220112010quadratic fitannual rainfall totalFigure 8: Nicoya long-term annual average rainfall trend. The fitted curve is for illustrative purposes only.Totals for years 2010 to 2015 are shown to highlight recent rainfall amounts.2.5 Annual and Monthly Total Rainfall in Recent YearsIn order to further investigate annual total rainfall in the past five years, we focus onthe IMN monthly data for the Nicoya agricultural extension station for the years 2010 to2015.Figure 8 and 9 show that in the years 2010 and 2011, Nicoya had more rainfall thatthe 40 year annual average, and that the excess occurred in both early and late rains.2012 had somewhat less than average mainly because the MSD was particularly dry, andthat 2013, 2014 and 2015 were substantially drier than average mainly because the earlyrains were very weak. It is most likely that this recent paucity of early rains has leftthe impression that rainfall has been decreasing. Only time will tell if this is part of alonger-term trend, or part of the ENSO effect in the particularly strong El Nin˜o year of2105. Note that the index results show ONI ă ´0.5 degrees C (La Nin˜a conditions)from mid 2010 to the first quarter of 2012; ´0.5 ă ONI ă 0.5 (neutral conditions) fromfirst quarter of 2012 to beginning of 2015, and ONI ą 0.5 (El Nin˜o conditions) untilmid 2016. http://www.cpc.ncep.noaa.gov/products/analysis_monitoring/lanina/enso_evolution-status-fcsts-web.pdf2.6 Spatial variability of rainfall in Nicoya PeninsulaAs can be seen from Figure 10, there are small, but distinct spatial variations in rainfallamong the five stations in the Nicola Peninsula. This variation has been previously noted,and is sometimes represented in a contour map of annual total rainfall. Notable alsoare the weak temporal trends in all rainfall totals. As noted earlier, these trends aresuperposed on large inter-annual variability masked in Figure 10 by the running mean.Also of interest are the consistent (from year to year) relative rainfall amounts betweenstations. While this consistency is probably driven by systematic spatial variation in122010 2011 2012 2013 2014 2015year0100200300400500600Total Monthly Rainfall (mm)monthly rainfall totalsGaussian fit to composite dataFigure 9: Monthly Nicoya rainfall for years 2010 to 2014. Blue line is actual, while red line is double Gaussianfitted to average monthly rainfall from 1977 to 2013.annual total rainfall, the irregular spacing of the stations makes it unwise to develop acontour map of annual total rainfall as the degree of interpolation needed for such a mapwould introduce spatial features that are likely due to the contouring process used, ratherthan real variations in rainfall. It is however possible to calculate rainfall gradients andgradient directions in a small number of sections within the study domain. Such sectionscan be used to investigate ecological and other gradients.3 Questions Posed by FuturAgua Research Team3.1 Annual Cycle of RainfallHydrologic modellers active in FuturAgua require detailed knowledge of the annual cycleof rainfall in the study domain in order to drive their hydrologic models. This cycle iscaptured by the data plotted in Figure 4, which gives a representative mean, as well asvariation around that mean, on a monthly basis. If finer temporal resolution is needed,the fitted double Gaussian can be used. As noted earlier, rainfall is highly variabilityon a diurnal scale, and very little guidance can be given for modelling directed towardsunderstanding rainfall-driven flash flooding.An important component of FuturAgua is driven by questions about rainfall in futureclimates. This work is being conducted by an analysis of future climates in a suite ofglobal climate models. The initial step is to determine which models adequately capturethe annual cycle of rainfall. It is being assumed that the ability to capture the doublemaximum and MSD indicates that a given model is properly representing major rainfallgenerating mechanisms in the study region. As for hydrologic modelling, Figure 4 providesa thorough picture of the annual cycle of rainfall, and can be used as a basis for modelevaluation.130 5 10 15 20 25window #0500100015002000rainfall total (mm)NicoyaPaqueraLiberiaLaGuineaGarzaFigure 10: Time series (1977 to 2013) of rainfall totals in spring (dashed lines) and fall (solid lines) rainfallpeaks at Nicoya (black), Paquera (red), Garza (purple), La Guinea (green) and Liberia (blue) stations. Trendshave been smoothed by a 15 year running mean.3.2 Spatial Variability of RainfallEcologists working within FuturAgua would like a map of spatial variability of rainfall. Asindicated in section 2.6, the distribution of rainfall measuring stations is such that a rainfallmap cannot be reliably generated form data alone. Dependent on station location, it maybe possible to determine rainfall gradients along chosen ecological transects. Anotherpossibility would be to use proxy rainfall data to provide an overall rainfall pattern, and tocalibrate this pattern by fitting to data at measurement stations. Rather than using proxydata, rainfall fields from satellite missions could be employed. The most important ofthese are Tropical Rainfall measurement Mission (TRMM, https://pmm.nasa.gov/TRMM)and Global Precipitation Measurement (GPM, https://pmm.nasa.gov/GPM). Since thesefields are rather coarse, station data could be used to provide more spatial detail.3.3 Rainfall Trends and Recent Dry Years in the Nicoya Penin-sulaFarming communities, water management agencies, planning and environmental agenciesand a number of other sectors have a direct interest in rainfall in the Nicola Peninsula. Ofparticular concern is the occurrence of extremely dry years. As noted earlier, many resi-dents are of the impression that rainfall has been decreasing, and that drought frequencyand intensity is increasing.This region is an extreme case of tropical wet-dry climates, with the months December,January and February often experiencing no measurable rainfall, and March and Apriloften being very dry. This is in strong contrast to the remaining months, which can bevery rainy. The region is thus always subjected to seasonal droughts, and much of life (andparticularly rain-fed agriculture) in the Nicoya Peninsula is conditioned by this feature ofthe regional climate. As indicated in Figures 8 and Figures 9, the years 2010 and 201114had substantially more rain that the long-term average annual total, and the years 2012to 2015 have been substantially drier than the long-term average. As noted in section 2.4,the strong inter-annual variability makes it impossible to detect a statistically significanttrend over the long-term.3.4 Rainfall Outlook for Water Availability and Drought Man-agement StrategiesLocal water managers and farmers would welcome information that will allow them tomake judgements about the potential viability of drought management activities. Ourfindings do allow a probabilistic approach that may prove useful. There exist internationalefforts to predict the ENSO phenomenon with a lead time of months to a year. For exam-ple, the Climate Prediction Centre of the USA National Weather Service provides com-prehensive information through http://www.cpc.ncep.noaa.gov/products/analysis_monitoring/lanina/enso_evolution-status-fcsts-web.pdf. Forecast sign and mag-nitude of ONI, together with our modelling of the strength of the ONI effect on fallmaximum rainfall peak from Section 2.3 could be used to estimate the likely fall rainfallpeak some months ahead. We must emphasize that this estimate is a statistical one thatignores the influence of known additional factors on rainfall inter-annual variability. Atbest, these estimates could be the basis for some (uncertain) choices about planting orgrazing decisions by farmers, and water allocation by water management agencies.We are aware that many water stakeholders in the study region, especially membersof water management and distribution agencies and consortia struggle to understand therelationship between rainfall and water availability. This relationship is conditioned bydetails of the hydrologic balance, which involves rainfall as an input and evapotranspira-tion, storage and runoff as outputs. The hydrologic balance is not addressed in this work,but is the subject of research being undertaken by other groups working in the FuturAguaproject. Our rainfall modelling will form an essential input into hydrological modelling,and can provide defensible rainfall scenarios for that work, as noted above.While we have made a suggestion on a possible approach to estimation of fall rainfall,given predicted ENSO intensity, we do want to emphasize that the estimation of future fallrainfall should be done only as an adjunct to the promotion of general resilience-relevantwater management strategies.4 ConclusionsIn this analysis of spatial and temporal variability of rainfall in the Guanacaste Peninsula,we have found that the annual rainfall cycle can be effectively and simply modelled bya double Gaussian function using monthly totals of rainfall. Strong temporal variabilityat time scales less than two weeks make it very difficult to say anything coherent aboutshorter term rainfall totals. There is strong inter-annual variability evident in the data,presumably driven by variations in rainfall forcing mechanisms inherent to the tropicalatmosphere in the study region. There is evidence for modest spatial variability, butspatial resolution of the monitoring network provides a rather rudimentary picture of thestructure. Remote sensing data may be useful in resolving the underlying spatial structure.Our simple analysis of single year data shows that there is a statistically significant trend15only in the magnitude of the September and October rainfall peak, and that the magnitudeis strongly affected by ENSO state, as captured by the ONI.A more sophisticated and robust Bayesian analysis confirms the appropriateness of thedouble Gaussian model applied to all 38 years of data, but shows no evidence of a timetrend over the nearly four decades. The Bayesian analysis does show strong evidence for anegative and highly significant annual ONI effect on rainfall. The magnitude of the effectis: if ONI in a given year is one degree C greater than ONI in an adjacent year, then theestimated average rainfall in the given year is expected to differ by a factor of 0.79 fromthat in the reference year.This report provided answers to a variety of applied hydrology and water managementquestions of importance to the FuturAgua project which it serves.5 AcknowledgementsThe FuturAgua project was funded by the Belmont Forum - G8 International Opportuni-ties Fund for Freshwater Security. The majority of funding for this research was providedby the Natural Science and Engineering Research Council of Canada. OH was supportedby funds from CANSI. We are grateful to scientists and colleagues at Instituto Meteo-rolgico Nacional, IMN and Area de Conservac´ion Tempisque, ACT of Costa Rica forfacilitating access to rainfall data. Silja Hund prepared the map.ReferencesCurtis, S., 2004. Diurnal cycle of rainfall and surface winds and the mid-summer droughtof mexico/central america. Climate Research 27 (1), 1–8.Enfield, D. B., Alfaro, E. J., 1999. The dependence of caribbean rainfall on the interactionof the tropical atlantic and pacific oceans. Journal of Climate 12 (7), 2093–2103.Harari, O., Altman, R., Moisseeva, N., Steyn, D., Welch, W., 2016. Statistical modellingof annual rainfall pattern in guanacaste, costa rica. in preparation Journal of Climate34, 35–50.Karnauskas, K. B., Seager, R., Giannini, A., Busalacchi, A., 2013. A simple mechanismfor the climatological midsummer drought along the pacific coast of central america.Atmosfera 26 (2), 261–281.Magan˜a, V., Amador, J. A., Medina, S., 1999. The midsummer drought over mexico andcentral america. Journal of Climate 12, 1577–1588.Maldonado, T., Rutgersson, A., Alfaro, E., Amador, J., Claremar, B., 2016. Interannualvariability of the midsummer drought in central america and the connection with seasurface temperatures. Advances in Geosciences 42, 35–50.Pena, M., Douglas, M. W., 2002. Characteristics of wet and dry spells over the pacific sideof central america during the rainy season. Monthly Weather Review 130, 3054–3073.Portig, W., 1976. The climate of central america. In: Schwerdtfeger, W. (Ed.), WorldSurvey of Climates. Elsevier, Amsterdam, pp. 405–478.16Rauscher, S. A., Giorgi, F., Diffenbaugh, N. S., Seth, A., 2008. Extension and intensifi-cation of the meso-american mid-summer drought in the twenty-first century. ClimateDynamics 31 (5), 551–571.Small, R. J. O., De Szoeke, S. P., Xie, S.-P., 2007. The central american midsummerdrought : Regional aspects and large-scale forcing. Journal of Climate 20, 4853–4873.Waylen, P. R., Caviedes, C., Quesada, M., 1996a. Interannual variability of monthlyprecipitation in costa rica. Journal of Climate 9, 2606–2613.Waylen, P. R., Quesada, M., 2002. The effect of atlantic and pacific sea surface temper-atures on the mid-summer drought of costa rica. Environmental Change and WaterSustainability, 193–205.Waylen, P. R., Quesada, M., Caviedes, C., 1996b. Temporal and spatial variability ofannual precipitation in costa rica and the southern oscillation. International Journal ofClimatology 16, 197–193.17


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items