UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

The application of agent based modeling to simulate residential water use responses to urban growth,… Bepple, Jonathan 2016

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

Item Metadata

Download

Media
24-ubc_2016_september_bepple_jonathan.pdf [ 1.83MB ]
Metadata
JSON: 24-1.0300421.json
JSON-LD: 24-1.0300421-ld.json
RDF/XML (Pretty): 24-1.0300421-rdf.xml
RDF/JSON: 24-1.0300421-rdf.json
Turtle: 24-1.0300421-turtle.txt
N-Triples: 24-1.0300421-rdf-ntriples.txt
Original Record: 24-1.0300421-source.json
Full Text
24-1.0300421-fulltext.txt
Citation
24-1.0300421.ris

Full Text

The Application of Agent Based Modeling to Simulate Residential Water Use Responses to Urban Growth, Regulation, and Social Influence in Kelowna BC, Canada.  by Jonathan Bepple  B.Sc, The University of British Columbia, 2013  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF  THE REQUIREMENTS FOR THE DEGREE OF  MASTER OF SCIENCE in THE COLLEGE OF GRADUATE STUDIES (Environmental Science)  THE UNIVERSITY OF BRITISH COLUMBIA (Okanagan)  April 2016 ©Jonathan Bepple, 2016 ii   iii  Abstract  The complexity of human and environment systems breed problems with no clearly definable, single best solution. Water management along with other public goods management represent fields that require coping with these problems. Agent based models (ABMs) have been shown to be an effective method of modeling the complex cross-scale interactions and feedbacks that contribute to the difficulty of large scale management in complex social and ecological systems. Herein is presented a framework for an ABM that incorporates novel methods in a new application to residential water use. The ABM developed in this thesis is a contribution to the literature in five ways: (1) active agents who can adapt to their social setting and environment have been implemented, (2) heuristic-based decision making is used within the model to improve realism, (3) the decision making process is rooted in the Theory of Planned Behaviour framework, (4) agents’ multi-faceted personalities are modeled using quantitative social data, and (5) the model allows for the exploration of price and volume allocation structure changes, education, centralized irrigation programs, water restrictions, land-use changes, and social influences simultaneously.  The ABM is applied to the case study of Kelowna, BC to assess residential water users’ responses to urban growth, regulation, and social influences. The most conservative growth estimates (no new development) show no increase in outdoor water use, while urban sprawl could result in an increase of 1.8% annually. Watering restrictions (29%-49% decreases) and leasing irrigation rights to a centralized irrigator (11%-20% decreases) were found to be the most impactful single policies, while the combination of rationing, restrictions, and price increases were found to have the largest influence overall (38% - 55% decreases). The social composition of the study area is such that social influence leads the population towards more water wasting iv  behaviour. Although education programs were ineffective at reducing outdoor water use in the model, they were somewhat effective at mitigating the effect of water wasting influences. Altogether, the decision support tool provided by this ABM fills a gap in water modeling and can provide valuable insight into social dynamics and responses to system change. v  Table of Contents Abstract .......................................................................................................................................... iii Table of Contents ............................................................................................................................ v List of Tables ................................................................................................................................ vii List of Figures .............................................................................................................................. viii List of Abbreviations ..................................................................................................................... xi Acknowledgements ....................................................................................................................... xii Chapter 1 Introduction .................................................................................................................... 1 Chapter 2 Literature Review ........................................................................................................... 8 2.1 Water Management in the Okanagan .................................................................................... 8 2.1.1 Regional Context ............................................................................................................ 9 2.1.2 History .......................................................................................................................... 10 2.1.3 Present day .................................................................................................................... 11 2.1.4 The Okanagan Basin Water Board ............................................................................... 13 2.2 Agent Based Models as Governance Support Tools ........................................................... 18 2.2.1 Agent Based Models ..................................................................................................... 18 2.2.2 Applications in Governance ......................................................................................... 20 2.3 Humans as Agents in Residential Water Use Modeling ..................................................... 22 2.3.1 Creating heterogeneous agents through statistical clustering ....................................... 26 Chapter 3 Methods ........................................................................................................................ 31 vi  3.1 Study Area ........................................................................................................................... 31 3.2 Data ..................................................................................................................................... 33 3.2.1 Agent personality traits ................................................................................................. 33 3.3 Model description ................................................................................................................ 37 3.3.1 Scale and Series of Events ............................................................................................ 37 3.3.2 Agent Decision Framework .......................................................................................... 40 3.3.3 Scenario Descriptions ................................................................................................... 43 Chapter 4 Results & Discussion ................................................................................................... 47 4.1 Cluster Analysis of Water Demand Survey Data ................................................................ 47 4.2 Validation ............................................................................................................................ 51 4.3 Growth Scenarios ................................................................................................................ 54 4.4 Policy Scenarios .................................................................................................................. 58 4.5 Social Dynamics .................................................................................................................. 64 Chapter 5 Conclusion .................................................................................................................... 79 Bibliography ................................................................................................................................. 83 Appendices .................................................................................................................................. 103 Appendix A Simulation Resolution ........................................................................................ 103 Appendix B Supplementary Agent Information ..................................................................... 107 Appendix C ODD +D Technical Description ......................................................................... 113  vii  List of Tables Table 1: Datasets used in this thesis work, their function, and source ......................................... 33 Table 2: Component questions informing agent personality traits ............................................... 34 Table 3: Growth scenario descriptions ......................................................................................... 44 Table 4: Summary of the number of individuals and dwellings being represented                 in a 400x400 pixel simulation of Kelowna, BC for four growth scenarios. .................. 45 Table 5: Policy scenario description. All policy scenarios are run using the                “Restrictive Growth” growth scenario ............................................................................ 46 Table 6: Cluster groups and their attitude and general description .............................................. 50 Table 7: The mean proportion of irrigating agents that have upgraded                 their landscape in a “Restrictive Growth” growth scenario. .......................................... 70 Table 8: The relationship between the resolution of a simulated city of Kelowna                 and the area of various residential land use types represented .................................... 105 Table 9: New Ecological Paradigm (NEP) Scale questions ....................................................... 107 Table 10: Personality traits by cluster group .............................................................................. 108 Table 11: List of agent attributes and descriptions ..................................................................... 108 Table 12: List of patch attributes and descriptions ..................................................................... 110 Table 13: List of global attributes and descriptions .................................................................... 111   viii  List of Figures Figure 1: The collection of decision support tools developed by the Okanagan Basin                  Water Board, Okanagan Water Stewardship Council and their partners. ..................... 16 Figure 2: The Theory of Planned Behaviour. ............................................................................... 25 Figure 3: City extent of Kelowna, B.C. Canada. .......................................................................... 32 Figure 4: Bayesian information criterion (BIC) comparison for 14 cluster models in R. ............ 48 Figure 5: Personality types and their associated traits. ................................................................. 50 Figure 6: Residential water use for the City of Kelowna Water Utility in 2013. ......................... 52 Figure 7: Modeled water use for each of 12 months. ................................................................... 52 Figure 8: Wilcoxon each pair comparison of the z-scores of real data and                  twelve baseline model runs. .......................................................................................... 53 Figure 9: Flowchart of daily operations ........................................................................................ 39 Figure 10: The percent increase in water use for three modeled growth                   scenarios relative to modeled current use. .................................................................. 54 Figure 11: Modeled rates of current water use and three growth scenarios.. ............................... 56 Figure 12: Percent increase in outdoor water use per irrigator within a                    growth scenario relative to modeled current water use. ............................................. 58 Figure 13: Reduction in outdoor water use due to a 300% increase in 2015                   watering rates/prices for Kelowna, BC. ...................................................................... 59 Figure 14: Increase in indoor water use for a range of policy scenarios                    within the Restrictive Growth scenario....................................................................... 61 Figure 15: Change in outdoor water use for a range of policy scenarios                    within the Restrictive Growth scenario....................................................................... 62 ix  Figure 16: The amount of water used through the course of one year as a result of                    social influence for three policy scenarios within the Restrictive Growth scenario. .. 65 Figure 17: Mean amounts of change in the ENTITLEMENT trait of agents for three                   policy scenarios within the Restrictive Growth scenario. ............................................ 67 Figure 18: Mean amounts of change in the FORESIGHT trait of agents for three                    policy scenarios within the Restrictive Growth scenario. ........................................... 67 Figure 19: The difference between the amounts of water for scenarios with controlled                   proportions of water saver agent types (P3) within the Restrictive Growth                    scenario. ...................................................................................................................... 69 Figure 20: Total outdoor water use for scenarios with controlled proportions of water                    saver agent types within the Restrictive Growth scenario. ......................................... 72 Figure 21: Total indoor water use for scenarios with controlled proportions of water                   conscious agent types within the Restrictive Growth scenario. .................................. 73 Figure 22: Mean amounts of change in the ENTITLEMENT trait of agents for                    scenarios with controlled proportions of water conscious agent types within                    the Restrictive Growth scenario. ................................................................................. 74 Figure 23: Mean amounts of change in the FORESIGHT trait of agents for scenarios                   with controlled proportions of water conscious agent types within the                    Restrictive Growth scenario. ....................................................................................... 75 Figure 24: The final proportion of landscape area upgraded for policy scenarios with                    controlled initial proportions of landscape area within the Restrictive Growth                   scenario. ...................................................................................................................... 77 x  Figure 25: Total outdoor water use for policy scenarios with controlled initial                    proportions of landscape area within the Restrictive Growth scenario. ..................... 78 Figure 26: A comparison of a 100x100 resolution landscape and a 25x25                   resolution landscape .. ............................................................................................... 104 Figure 27: User interface of the model. ...................................................................................... 112xi  List of Abbreviations ABM – Agent based model BIC – Bayesian information criterion TPB – Theory of Planned Behaviour OBWB – Okanagan Basin Water Board WSC – (Okanagan) Water Stewardship Council ODD – Overview, Design concepts, and Details ODD +D – Overview, Design concepts, Details, and Decision GIS – Geographic Information Systems  xii  Acknowledgements  I sincerely appreciate the love, support, and encouragement from my parents, Keith and Dinah, my brothers Scott and Brian, and my sister Katrina who first inspired me to pursue my passion of science. I would also like to acknowledge the support from my friends for getting me through the long nights, my colleagues for giving me new perspectives, and my professors for teaching me how to get from A to B and so much more.    1  Chapter 1 Introduction Our world is made up of a multitude of physical, chemical, and biological processes that have been occurring and interacting across vastly different scales for billions of years. These interacting scales, processes, and entities make up a highly complex system in a state of a semi-fluid, adaptive flux. This dynamic state is subject to feedback loops and the cross-scale interactions associated with its complexity (Parrott & Meyer, 2012; Liu et al., 2007). Humans, more than any other species, have the ability to affect and shift this system and its dynamic processes. Our expanding population and prosperity require us to harvest and use natural resources and ecosystem services at consistently increasing rates (IPCC, 2014). If we seek to be a sustainable population, we must actively manage our finite resources to counteract our increasing resource use rates. New technology and methods – such as agent based modeling - give current day land and resource managers effective and valuable tools that have only recently become viable (Parrott, 2011; Parrott & Latombe, 2012; Parrott & Meyer, 2012; Polasky et al., 2008; Rammel et al., 2007; Schlüter & Pahl-wostl, 2007) Evidence of our power to change our environment can be seen perhaps most dramatically in our ability to change the massively macroscale process of climate. The Intergovernmental Panel on Climate Change (IPCC) states that “the period from 1983 to 2012 was likely the warmest 30-year period of the last 1400 years” (IPCC, 2014) and this trend does not appear to be abating.  Global temperatures in 2015 broke the past record for warmest year – set in 2014 – by the widest margin a record breaking year has ever had since record keeping began in 1880 (NOAA, 2016). This margin equated to a 0.16°C increase in the annual average from 2014, a warming that is contributing to warming oceans, receding glaciers and snow pack, and rising sea levels (IPCC, 2014). The Planetary Boundary concept proposes nine global limits, past which the 2  earth has an increased chance of changing from a “Holocene-like state” (Rockström et al., 2009; Steffen et al., 2015). Climate change is suggested as one of two key boundaries and is shown to be beyond the safe operating space. Beyond the safe operating space there is increased uncertainty and an increasing risk to endangering other boundaries such as freshwater use (Rockström et al., 2009; Steffen et al., 2015).  The impact of climate change on our water resources is perhaps one of the most significant consequences of this warming trend. For example, climate change may not have a strong influence on the overall precipitation volume in the Pacific Northwest of North America, but it is predicted to have a significant impact on the yearly timing and form of precipitation (Vano et al., 2010a; Vano et al., 2010b). Areas that previously received precipitation in the form of snow will begin receiving more precipitation as rain and less as snow (Vano et al., 2010a; Vano et al., 2010b). In regions characterized by seasonally snow-capped mountains, such as those in the interior of BC, a reduction in the amount of snow received will lead to a smaller snowpack, resulting in reduced water availability in the dry summer months. Further stress on the water supply will occur in the influence of climate change on the timing of streamflows. As landscapes transition to having more rain-dominated precipitation regimes, the timing of streamflows will change significantly. For example, between 1948 to 2002 there has been a significant advance in streamflow timings of one to four weeks across Western North America (Stewart et al., 2005). Climate change models indicate that this trend – shown to be independent of the Pacific Decadal Oscillation’s influence – has been modeled to result in a further 20 – 40 day advance of these streamflow timings across the Pacific Northwest by the end of the century (Stewart et al., 2004).  In addition to the earlier timing of streamflows, an increased atmospheric temperature and reduced snowpack will result in a higher rate of flow 3  earlier in the year, and a reduced flow in the dry summer months (Hamlet et al., 2007). Increased variability and a reduced availability of water during summer months, when demand peaks, will likely create new difficulties for water planners and users in arid and semi-arid areas such as the Okanagan Valley.  In arid and semi-arid regions, water is often a limiting factor to the ecosystems present. Water related stresses can be exacerbated by three of the major human contributors to habitat and biodiversity loss: climate change, residential development, and agriculture (Austin et al., 2008). Climate change is expected to reduce the inflows to Okanagan Lake by upwards of 30% by 2080 (Cohen & Neale, 2006b; Cohen et al., 2004; Merritt et al., 2003). This reduction in supply will be coupled with an increase in demand due to a rapidly expanding human population. There have been variable rates of population growth the Okanagan Valley, ranging from 2.6% from 2006 – 2011 in the southern Penticton region, to 10.8% over the same period in Kelowna and its surrounding area (Statistics Canada, 2012a). Kelowna’s high rate of growth makes it the fourth fastest growing metropolitan area in the country (Statistics Canada, 2012b). Some studies suggest that residential growth in the Okanagan by 2040 will lead to overall demand increases of between 10 – 40% from 2006 levels (Polar Geoscience Ltd., 2012). Residential use is projected to increase by 2.5% per year indoors and 3 – 6% per year outdoors, while agricultural use will likely increase at a rate of 8% annually (Polar Geoscience Ltd., 2012). A reduction in water supply and increases in the water demand of city that is home to some of the highest water using residents in Canada, will likely have an effect on the surrounding landscape (Cohen & Neale, 2006b). These additional stresses may have an amplified negative effect in the semi-arid Okanagan which is home to some of the most endangered habitats in Canada, with upwards of 92% ecosystem loss in grassland ecosystems and 75% in 4  wetland ecosystems (Lea, 2008). Local governments must therefore carefully consider the impact of water management – which has previously been largely driven by development, tourism, and agricultural needs – on the surrounding landscape (Wagner & White, 2009)  Managing the priorities of agricultural demand, development and growth, ecosystem health and services, and industries – such as tourism – that depend on water, make equitable water governance in the Okanagan a “wicked problem”. Wicked problems arise in situations where there are multiple interacting entities that are trying to maximize local solutions for differing priorities or objectives (Churchman, 1967; Rittel & Webber, 1973). These problems persist because competing or conflicting priorities mean that there is often no single or best solution (Rittel & Webber, 1973). Through observation of governing bodies dealing with wicked problems, research into the implications of wicked problems on public policy and management suggests that there are three strategies to deal with these complex problems: (1) collaboration of those involved in the problem, (2) instituting adaptive and collaborative leadership, and (3) consideration of more holistic solutions and a broader scale influences to the problem (Chrislip et al., 1994; Head & Alford, 2015; Lasker & Weiss, 2003). Water management in the Okanagan is therefore well situated to deal with the wicked problem of effective governance by actively engaging in all three of these strategies. There is a long history of collaborative water management in the Okanagan (Wagner, 2008; Wagner & White, 2009). The Okanagan Basin Water Board (OBWB) is a regional-scale advisory body that represents three regional districts, First Nations people, the Water Supply Association of BC, and a technical advisory committee called the Water Stewardship Council (WSC) (Okanagan Basin Water Board, 2015b). The WSC itself is made of 28 members from academia, industry, and all levels of government. The collaboration and leadership provided by 5  these entities has led to the creation of valley-wide studies of water supply and demand and the implementation of a regional sustainability strategy to ensure continued growth despite limited water resources (Cohen & Kulkarni, 2001; Cohen & Neale, 2006a; Harma et al., 2012; Jatel, 2008; Wagner & White, 2009). Even with this collaboration and leadership, the complexity of the water management challenges in the region mean that there are gaps in the tools available for managers to assess many of the social aspects of these issues. Weber and Khademian (2008) suggest that wicked problems “place a critical emphasis on the tasks of knowledge transmission and integration. [I.e. tasks] that are grounded in social and political relationships involving heterogeneous actors with diverse interests and goals.” Traditional tools that were developed to give insight on water demand often took the form of mathematical, regression based models that used previous data to predict future trends (House-Peters & Chang, 2011). These types of mathematical models fail to integrate heterogeneous agents with variable goals and traits, or the social interaction of these agents, thus limiting our ability to understand the mechanism by which the patterns of interest emerge.   The shortfalls of traditional modeling methods can be addressed with the development of agent-based models (ABMs) (An & López-Carr, 2012; Batty, 2009; Bruch & Atwell, 2013; Filatova et al., 2013; Janssen & Ostrom, 2006; Railsback & Grimm, 2012; Rounsevell et al., 2012). Agent-based models simulate multiple interacting, unique decision makers in a spatial landscape, allowing exploration of cross scale interactions and the non-linear dynamics of a complex social system (Parrott, 2011; Parrott & Meyer, 2012). In the context of water demand, the agents can represent the individual users of water who can each have a different set of attributes or objectives. A modeller can therefore test the effects of social, environmental, or 6  regulatory changes in a way that was not possible with previous methods (House-Peters & Chang, 2011).  Although the use of ABMs in water demand management is a relatively new venture (Berglund, 2015; House-Peters & Chang, 2011), they have been shown to be uniquely effective for exploring the effects of regulatory changes (Athanasiadis et al., 2005; Galán et al., 2009; M. Giacomoni et al., 2013; M. H. Giacomoni & Berglund, 2015; Moss & Edmonds, 2005; Schlüter & Pahl-wostl, 2007; Schwarz & Ernst, 2009a; T. Chu et al., 2009), urban development patterns (Galan et al., 2008; Galán et al., 2009; Lopez-Paredes et al., 2005; M. H. Giacomoni & Berglund, 2015), and social influence (Moss & Edmonds, 2005) on population level water use patterns. Recent uses of ABMs have highlighted their utility in adaptive management scenario testing (Berglund, 2015). The possibility of integrating social and physical data in an agent-based model, and the possibility to simulate bottom up system change in a spatial and easily recognizable landscape, make ABMs particularly well-suited for knowledge communication, and a valuable tool to add to the water management arsenal (Parrott & Meyer, 2012). It is the goal of this thesis to present an ABM developed to assess the response of residential water users to the combination of social influences, heterogeneous personality types, and regulatory measures of water demand management. More explicitly, the objective of this thesis is to answer these three questions: 1. What policy or policies are most effective at reducing residential water use?  2. How will growth patterns of urban sprawl, densification, and restrictive growth impact residential water use? 3. What influence does social interaction have on residential water use?  7  This case study was developed for the city of Kelowna, BC, located at the centre of the Okanagan Valley. As the largest city in the Okanagan, and one that is experiencing the most significant growth, Kelowna water management issues provide an ideal starting point for the application of an ABM in the Okanagan. It is the hope of the author that this model will be a useful tool for assessing the social dynamics of water demand management in the region. This work also represents a unique contribution to the ABM and water management literature in five ways: 1. residential water use agents are active decision makers and can adapt to their social setting and environment; 2. the use of “human-like” heuristic based decision making is applied in a residential water use setting; 3. the decision making framework of the Theory of Planned Behaviour is applied in a residential water use setting;  4. multi-faceted agent personality traits are empirically derived using local data; 5. the model allows for the exploration of price and volume allocation structure changes, education, centralized irrigation programs, water restrictions, land-use changes, and social influences simultaneously to allow for the investigation of possible synergies between management strategies. 8  Chapter 2 Literature Review  2.1 Water Management in the Okanagan  In a national report on the state of water management in Canada, the Senate Committee on Energy, the Environment and Natural Resources stated that the current practices were “shocking” and that major knowledge gaps were “more than regrettable, [they were] unacceptable” (Senate of Canada, 2005). Although research on Canada’s water resources and management has been improving (Bakker & Cook, 2011), and new legislation such as BC’s Water Sustainability Act (BC Ministry of Environment, 2016) are steps in the right direction, our understanding of potential future water demands is incomplete since it is based on traditional methods that have the inability to integrate spatial and social components (House-Peters & Chang, 2011).     The Okanagan Valley has been an area focused on agriculture and lake-based tourism since it was established (Wagner, 2008; Wagner & White, 2009). Water is essential in this semi-arid desert, and is used by every organism and every industry. As such, the effective distribution of water is difficult because of the range of competing and conflicting uses. The complexity inherent in this human and environmental system creates wicked problems that necessitate the use of new methods for effective management (Head & Alford, 2015; Parrott & Meyer, 2012; Martin-Ortega et al., 2015; Weber & Khademian, 2008). Agent-based models represent a new method applied to the field of water management that is capable of providing better information to support improving current management practices (Berglund, 2015; House-Peters & Chang, 2011). 9  2.1.1 Regional Context  The Okanagan Basin is situated in south central British Columbia and encompasses approximately 8,000 km2 of mountainous terrain surrounding a large central valley and lake (Okanagan Water Stewardship Council, 2008). The region is semi-arid with high evaporative potential. Of the 30 cm of precipitation that the valley bottom sees in an average year, approximately 85% is lost to evaporation and evapotranspiration (Cohen & Kulkarni, 2001). With a low amount of water flowing through the landscape,  water has a residence time of 60 years within the lake and therefore the lake replenishes slowly (Wagner & White, 2009).  Despite the large surface area of roughly 350 km2, Okanagan Lake is a poor source of freshwater. Only the top 1.5 m of water is replenished naturally each year and so any additional water removed will drop lake levels (Okanagan Water Stewardship Council, 2008). Tributary streams are a major source of water for most municipalities in the region but most streams are either fully allocated or over allocated affording no buffers in drought years (Cohen & Kulkarni, 2001). Although most water licenses are rarely fully used in a year, the seniority basis of this system, new mandate for the consideration of environmental flows by the Water Sustainability Act, and the potential for drought could result in years where not all licenses could be honoured and therefore pose a threat to current lifestyles within the Okanagan.   The Okanagan is home to a population that is, in general, dangerously unaware of the water supply issues in the region. Despite having one of the lowest freshwater availabilities per person in Canada, citizens use an average of 675 L/person/day, greater than double the national average (Statistics Canada, 2013b; Summit Environmental Consulting Inc., 2010). Only 150 L/person/day is used indoors, and so the region’s high average is a result of outdoor water use which jumps to over 1,000 L/person/day for six months of the year to support non-essential 10  irrigation (Summit Environmental Consulting Inc., 2010). This discrepancy between available supply and high demand represent a danger to the surrounding semi-arid desert ecosystems and the sustained growth of regional economies that depend on readily available access to water. These challenges emphasize the present lack of awareness of the average water user but they also represent an opportunity for demand side management. The dangers of poor planning and over use were illustrated during a drought in 2003 that caused the town of Summerland, BC to disregard federal regulations and fully stop the flow of water from their reservoir to Trout Creek to instead preserve reservoir levels (Wagner & White, 2009). Trout Creek is just one of many habitats at risk due to the over use of water resources in the basin. Due to human development and high sensitivity to water availability, the semi-arid ecosystems within the basin are some of the most endangered in Canada (Austin et al., 2008). Climate change is expected to amplify these challenges for both natural ecosystems and the human environment at the centre of this study (Cohen & Kulkarni, 2001).  2.1.2 History  Canadian water governance falls largely within the jurisdiction of provinces and territories. While the federal government presides over fisheries, aquatic navigation, and international waters, provincial and territorial governments are responsible for water resources and supply(Bakker & Cook, 2011).  The provinces retain ownership of the waters within their borders and delegate water management down to the sub-provincial levels. In BC, the province issues water licenses to cities, regional districts, and water purveyors for prescribed amounts of usage and for prescribed purposes. These licenses set limits to three things: surface water withdrawal, infrastructure construction, and allowable uses (Wagner & White, 2009). In BC the 11  passage of the 1914 Water Act allowed four types of entities to apply for water licenses: “i) water user communities, ii) mutual water companies, iii) land and water companies, and iv) public irrigation corporations” (Wilson, 1989).   After the passage of the original Water Act, privately owned irrigation infrastructure and licenses in the Okanagan were purchased by public irrigation corporations that became the irrigation districts that now supply the Okanagan (Wilson, 1989). The Water Act notably omitted any minimum environmental flow requirements. Although this opened the way for rapid development, it left little consideration for sustainable development until this omission was amended in 1977  with the passing of the Fisheries Act which protected fish habitat in particular, and then amendments to the Water Act in 1978 that protected streams more generally (Mattison, 2016). This previous century of overdevelopment of irrigation infrastructure encouraged urban sprawl and agricultural development on semi-arid land that would otherwise be infertile. Today this agricultural development equates to 220 km2 of agricultural land, of which 190 km2 are irrigated (Okanagan Water Stewardship Council, 2008)  2.1.3 Present day  As the Okanagan population expanded, increasing demand was addressed by constructing reservoirs to increase the basin’s capacity to store water. Further capacity development is prohibitively expensive as all suitable and cost effective locations have already been developed (McNeill, 2006). Recent analyses found that in the Okanagan Basin, the water storage capacity will be unable to meet the regions needs by 2050 (Harma et al., 2012). In some Okanagan communities such as Summerland, demand could outpace supply by as early as 2020 without even considering the effects of population growth on the region (Cohen et al., 2004). The 12  Okanagan’s relative inability to expand supply is one of a number of pressures, along with population growth, that contribute to an almost certainly dry future for the area.  According to the 2011 Canadian census, Kelowna is the fourth fastest growing metropolitan area in Canada, growing 10.8% from 2006 - 2011 (Statistics Canada, 2012b). This growth is projected to expand the current population of roughly 117,000 individuals to around 161,000 by 2030 (City of Kelowna, 2013; Statistics Canada, 2012a).  In addition to this steady increase in the resident population, there is a large transient population in the form of an estimated 1.5 million tourists that visit the city annually (InterVISTAS Counsulting Inc., 2011). Okanagan beaches, outdoor activities and wine industries attract tourists, who are generally unaware of the water challenges faced in the Okanagan. The temporary expansion of water users is particularly draining in the summertime when water supply is at its lowest. The city’s demographic profile and trend of growth are the putting enormous strain on the Okanagan’s water management system, a system on which many in the region depend for their livelihood.   The Okanagan is still a strongly agricultural region. Although the basin represents less than 2% of the land in BC, it produces the majority of the province’s pears, 89% of wine-grapes, and 98% of the apples for the province’s agricultural industry (British Columbia Ministry of Agriculture, Food, and Fisheries, 2004; Cohen & Neale, 2006b). This industry is highly dependent on consistent access to water throughout the summer months, when residential demand is also at its peak.    There are 14 communities along the Okanagan Lake shoreline and each represent populations with multiple competing stakeholders and objectives for water use. The competing uses make effective water management in the Okanagan Basin a “wicked problem”. Wicked problems are often associated with public administration and management issues. These 13  problems arise when multiple interacting factors compete for individually optimised results and the problems persist because there is no single best, definitive, or permanent solution that maximizes the benefit of all stakeholders. Although originally described as being so complex as to be nearly unsolvable (Churchman, 1967; Rittel & Webber, 1973), recent literature provides some direction in dealing with wicked problems (Head & Alford, 2015; Martin-Ortega et al., 2015; Weber & Khademian, 2008).  Head and Alford (2015) maintain that wicked problems rarely have definitive solutions that are permanent, sustainable, and equally beneficial for all stakeholders involved, but that provisional solutions are possible. Head and Alford advocate broader, more holistic approaches to the problem, collaboration of those involved, along with adaptive and collaborative leadership to address wicked problems. Cross basin collaboration and adaptive and collaborative leadership are present in the Okanagan through the leadership of the Okanagan Basin Water Board and Water Stewardship Council and the high levels of collaboration they facilitate. An inclusion of this model into the suite of modeling approaches currently being undertaken (Figure 1) would complement the currently available water models present in the valley, and provide a more holistic solution to water management by supplementing the current hydrological and mathematical modeling with social modeling of residential water use.  2.1.4 The Okanagan Basin Water Board  The Okanagan Basin Water Board (OBWB) is unique in the province by way of its composition of collaborating non-governmental organisations, municipal governments, and First Nations government. It consists of 12 directors, three from each of the South, Central, and North Okanagan Regional Districts and one representative from each of the Okanagan Nation Alliance, 14  the Water Supply Association of BC, and the Okanagan Water Stewardship Council. This local government group  is comprised of a variety of stakeholders in the region and was created in 1965 by municipal governments throughout the basin as a way to address water management issues in a coordinated manner across the watershed and distribute grant money for associated water projects (Okanagan Basin Water Board, 2015b). Originally intended to tackle issues of water quality and milfoil spread,  the OBWB has now expanded to deal with issues relating to climate change and lobbying to protect the basin’s water supply (Okanagan Basin Water Board, 2015b).   One of the OBWB’s most significant initiatives in water management in the Okanagan has been the 2006 creation of their technical advisory body, the Water Stewardship Council (Wagner & White, 2009). The WSC includes 28 members that are representatives from industry, academia, and all levels of government. The primary function of the Water Stewardship Council is to provide technical advice for the OBWB and support fact-based management, legislation, and communication (Okanagan Basin Water Board, 2015a). One of the most significant, tangible products of the council to date has been the water supply, demand, and sustainability reporting and research they have conducted. Their Okanagan Sustainable Water Strategy formalized a long term, cross jurisdictional direction for the valley’s water governance (Okanagan Water Stewardship Council, 2008). This report details a framework for addressing the water needs for all stakeholders from human use, to industry, and environmental flows in the face of the challenges facing the region (Okanagan Water Stewardship Council, 2008).  15  The OBWB with the Water Stewardship Council aim to inform the valley’s governance with evidence based water modeling and accounting tools. The groups have facilitated the creation of the six following tools:  1. the Okanagan Water Demand Model (OWDM) to assess demand needs in the valley, 2. the Okanagan Basin Hydrology Model (OBHM) to assess available supply, 3. the Okanagan Basin Water Accounting Model (OBWAM) to assess potential impacts from climate change, mountain pine beetle infestation, and other stressors,  4. the Okanagan Hydrologic Connectivity Model (OHCM) to assess water availability down the valley gradient, 5. the Groundwater Balance Analysis spreadsheet Tool (GWBAT) to assess flows between aquifers, 6. the Okanagan Water Allocation Tool (OWAT) is in progress and is meant to assist in water licensing decisions in the basin,  These models and tools are interconnected and build upon each other. The flow of data and influences from one model to the next can be seen in Figure 1.  16   Figure 1: The collection of decision support tools developed by the Okanagan Basin Water Board, Okanagan Water Stewardship Council and their partners. Arrows represent the flow of output from one model or tool into another. At time of publishing, the completion of the OWAT is still underway. This collection of models represents perhaps one of the most thorough suites of its kind in Canada. The work presented in this thesis is meant to supplement the modeling tools already developed by the OBWB and the Water Stewardship Council. Specifically, the agent-based model (ABM) herein is an attempt to improve upon the current state of water demand modeling in the Okanagan by integrating new methods to assess how present and future populations in Kelowna, BC might respond to demand management measures. 17   The Okanagan Sustainable Water Strategy highlights twelve guiding principles to support sustainable and equitable use of water in the basin. This agent-based model aims to support four of these guiding principles: 1) integrate land use planning and water resource management, 2) collect and disseminate scientific information on Okanagan water, 3) provide sufficient resources for local water management initiatives, and 4) practice adaptive water and land management. The model supports these principles in the following ways: 1. modeling growth scenarios of restricted growth, densification, and urban sprawl while simultaneously modeling the effects of policy implementation allow us to integrate land use planning and water resources management;  2. agent-based modeling of residential decision makers’ responses to policy implementation provides insight into the potential effects of policy implementation for demand side management in a way that is not yet available in the region; 3. production of this model on open software allows local groups to use the model free of charge;  4. a user interface that allows control over the variables within the simulation and the ability to load Geographic Information System files (that have been modified to include the required attribute values) allows users of this model to explore a range of scenarios that are adaptable to new initial conditions. The work within this thesis represents a new tool that can be useful to practitioners and planners in the region when assessing the potential impacts of introducing policy or planning development patterns on the water use of Kelowna residents. The ABM produced in this thesis is meant to be used as an evolving decision support tool. With continued support alongside the suite of other Okanagan Basin models (Figure 1) this ABM can provide a bottom-up assessment 18  of residential water use for a more detailed investigation of residential water use in response to policy implementation and a range of growth scenarios.    2.2 Agent Based Models as Governance Support Tools 2.2.1 Agent Based Models  Agent based models are a form of computer based modeling that depends on interactions between multiple variable decision makers to create the dynamics seen within the system. There are three primary components of any agent based model: the agent, the landscape, and the interactions (Railsback & Grimm, 2012). The agent  The agent within an ABM is the decision maker of the system that is being modeled. An agent can represent, for example, an organism, a household, or an institution. The agents are necessarily unique from one another in any of a number of ways. Often they differ in available resources, location, or behaviour type.  They make decisions based on a set of rules or objectives that can vary from one agent to the next. The landscape  The landscape of the ABM contributes to the dynamics of the agents. Commonly the landscape is represented by squares, pixels, or patches which are often heterogeneous. The patches can be assigned any number of relevant attributes such as slope, elevation, land use type, and resource availability or demand. Although abstract landscapes can be used, ABMs are often linked with Geographic Information Systems (GIS) data to create an accurate representation of a real-world landscape within the computer model.  19  The interactions  The interactions encompass the rules and relationships that structure the system. Agents are often linked with each other and/or their landscape and interact through these linkages. Information or resources can be shared through these links and influence how agents make decisions. It is these interactions that can generate feedback between agents and their social and physical environments, and that give rise to the dynamics of the system.  Traditional systems research has used top-down methods such as system dynamics modeling to assess whole system changes by characterizing the system components as a series of aggregated sources and sinks with flows of resources, information, etc. between them. In contrast, ABMs model a system from the bottom-up by simulating multiple autonomous entities and their local interactions.  The emergence of system-wide changes within an ABM provides a better understanding of the mechanisms that led to that change because it allows one to test the interactions and feedbacks that led to those system level dynamics (Batty, 2009; Bruch & Atwell, 2013; Filatova et al., 2013; Janssen & Ostrom, 2006; Railsback & Grimm, 2012; Rounsevell et al., 2012). This unique property of ABMs makes them particularly well-suited for complex social and ecological systems and the “wicked” problems they face. It is perhaps because of this utility, and recent improvements in computational power that there has been an increase in ABM research from a single paper being published on the subject in 1990, to over 110 papers being published annually - only twenty years later - in fields including ecology, engineering, psychology, economics, and resource management (Niazi & Hussain, 2011)  Although they offer a new and useful tool, they are not without their limitations. Agent based models have been criticized for the difficulty associated with validating them, or the subjective nature of the validation process (Edmonds & Chattoe, 2005). ABMs, largely used to 20  model social systems, can be difficult to validate because of human behaviour frameworks often include stochastic elements, making point predictions difficult (Berglund, 2015). Additionally, social systems are complex with large numbers of contributing factors that can be difficult to calibrate. Validation of ABMs can been done via comparing past observed patterns to those produced by the model, in an approach called pattern-oriented modeling (Chion et al., 2011; Railsback & Grimm, 2012). Finally, there is a developing formal system for communicating model components of an ABM. The Overview, Design concepts, and Details (ODD) protocol (Railsback & Grimm, 2012) is one such tool that has seen recent progress to improve the emphasis on the decision making components within the model (Müller et al., 2013).  2.2.2 Applications in Governance Managers of large landscapes deal with “wicked” problems due to the assortment of natural and anthropological processes occurring at and across the landscape scale. The ability of an agent based model to explore cross scale interactions of spatial agents, and non-linear dynamics of a social and ecological system, make it a useful decision support tool for management (Parrott, 2011; Parrott & Meyer, 2012). An agent based model can be a decision support tool that has qualities that make it uniquely powerful among other modeling methods in management. Although data intensive, ABMs often use data that is available through existing monitoring programs and thus facilitates collaboration and learning opportunities between relevant parties. Additionally, ABMs often model tangible, recognizable systems and as such can be easily communicated to non-modellers  (Parrott et al., 2011). Despite their utility, the use of ABMs in water demand management is relatively new (Berglund, 2015; House-Peters & Chang, 2011)  21  Water managers have created mathematical models of historical water usage that are used to project future water use for decades. These early models are often simple linear regressions that lack a spatial component and utilize aggregated usage (House-Peters & Chang, 2011). In their review of urban water demand modeling, House-Peters and Chang review how ABMs improve upon early methodologies (2011). In their comparison, ABMs are shown to be superior to simple econometric and time series regression models in their ability to include numerous social and physical variables (both qualitative and quantitative), model policy intervention, simulating individual decisions, and accounting for uncertainty (House-Peters & Chang, 2011). Since ABMs do not need to rely heavily on previous trends to model future ones, they can be particularly useful when there is limited time series data or when demand rates are changing due to technological or attitudinal shifts (J. Chu et al., 2009).  In a review of current methods for the application of ABMs to the assessment of water resource systems, Berglund finds that there is a focus on the use of ABMs in adaptive management scenario testing (2015). In particular Berglund highlights cases in which ABMs are used to predict the expansion of infrastructure – such as higher efficiency irrigation technology - and water demand. The current residential water use ABMs reviewed investigate few policy implementation options in each study, eliminating the ability for the researchers to test for possible synergies between different demand management measures. Some of the available models assess the effects of education on water use (Athanasiadis et al., 2005; Galán et al., 2009; Moss & Edmonds, 2005; Schwarz & Ernst, 2009b), investigate the influence of urban development patterns on water use (Galan et al., 2008; Galán et al., 2009; Lopez-Paredes et al., 2005; M. H. Giacomoni & Berglund, 2015), the effects of pricing on water use (Athanasiadis et al., 2005; J. Chu et al., 2009), the effects of water restrictions on water use (M. H. Giacomoni & 22  Berglund, 2015), the effect of social influence on water use (Moss & Edmonds, 2005), or even the effects of subsidies on efficiency technology upgrade uptake (Schwarz & Ernst, 2009b). None, however, investigate the potential effects of the combination of all of these in their models, negating the potential to assess possible synergistic effects.  Traditional modeling methods were limited in the number of simultaneous factors included because of the difficulty of modeling increased complexity (House-Peters & Chang, 2011). Agent based models allow for an increase in the complexity of the modeled system. Despite this, the ABMs of residential water demand described above have investigated limited sets of policy options. Although the focus on only one or two of these elements allows for a simplified model, the lack of policy options that can be potentially investigated reduces the flexibility of the tool to decision makers who may be interested in a range of policy options. In an effort to create an ABM decision support tool that is as flexible as possible and potentially better suited for addressing the wicked problem of residential water demand management, the ABM within this thesis integrates a wide range of influencing factors at a scale not yet available in the literature. This model investigates how water use is affected by: subsidies, price and volume allocation structure, education, centralized irrigation programs, water restrictions, land-use change, social influences, and empirically based, individual personalities.  2.3 Humans as Agents in Residential Water Use Modeling    One of the significant benefits of using agent based modeling in landscape scale management is that a population of human decision makers can be included. Simulating 23  individual decision makers allows for the emergence of system dynamics through bottom up processes and thus provides a clearer understanding of the mechanisms leading to change compared to top-down modeling approaches (Parrott, 2011; Parrott & Meyer, 2012; Müller et al., 2013). Additionally, ABMs can be fully stochastic, they are ideally suited for using GIS data, and lend themselves naturally to a very detailed approach. The power of ABMs lie in their ability to realistically represent a system but at a simplified scale to avoid becoming complicated and to remain computationally feasible (Le et al., 2008). The widespread proliferation and integration of GIS data provide an effective method of accurately representing a real environment within a model but there is no standardized framework to characterize  people within a model as faithfully as GIS characterizes the landscape (An, 2012; Valbuena et al., 2008) In agent based modeling an agent’s respond to a stimulus can be classified as either active, or reactive (Berglund, 2015; Woolridge, 2002). Reactive agents or passive agents are ones that react to stimuli either from the environment or from other agents in a consistent, logical way. Models with reactive agents are generally suited for simulating large populations of simple actors (Berglund, 2015). Active agents are ones that have goals and change the way they react to stimuli based on these goals. Models with active agents are more computationally intensive and generally well suited for modeling system dynamics of a smaller group of optimizing agents working towards independent goals (Berglund, 2015). Reactive agents generally do not learn or change their behaviour during a simulation, while active agents are capable of learning and behaviour change (Woolridge, 2002)  Residential water demand ABMs aim to simulate the large, population-scale, water use in a system. Previous models contain solely reactive agents (Altaweel et al., 2009; Athanasiadis et 24  al., 2005; Galán et al., 2009; J. Chu et al., 2009; Kanta & Zechman, 2014; Lopez-Paredes et al., 2005; M. H. Giacomoni & Berglund, 2015; Moss & Edmonds, 2005). One important limiting factor of reactive agents is that they follow predefined logical rules (Berglund, 2015). These logical rules often assume fully rational agents which may be inappropriate for modeling the bounded rationality of human decision makers (An, 2012; Camerer & Loewenstein, 2004; Manson & Evans, 2007) Early work in social sciences modeled decision making of a fully rational decision maker while recent efforts have highlighted the importance of considering the bounds that are placed on human rationality in the form of constrained information, time, resources, or memory (An, 2012). Humans were first assumed to make fully rational, perfectly informed decisions to maximize personal gain because this type of behaviour could be easily modeled with utility functions (Kahneman & Tversky, 1979; Manson & Evans, 2007). This method of representing human decision making is criticized for its oversimplification the human decision making process (Camerer & Loewenstein, 2004; Jones, 1999). In reality decisions are rarely made with access to all the relevant information needed to create an optimized action. Even with all available information, decision makers are influenced by biases from habituation, past experiences, and beliefs (Bennett & Tang, 2005; Brenner, 2006; Gelenbe et al., 2001; Hodgson & Knudsen, 2004), errors of prediction (Gintis, 2000; Kahneman & Tversky, 1979; Lux & Marchesi, 1999; Müller et al., 2013; Schlüter & Pahl-wostl, 2007), limitations of memory (Ariely et al., 2000; Müller et al., 2013; Schlüter & Pahl-wostl, 2007), social influences (Little & McDonald, 2007), and limited time or cognitive power (Simon, 2000). A founder of bounded rationality research and Nobel Prize winner Herbert Simon states that “[humans] must use approximate methods to handle most tasks” (Simon, 1990). Indeed, 25  most people make decisions using “rules of thumb” or heuristics (Arló-Costa & Pedersen, 2012; Hutchinson & Gigerenzer, 2005). Heuristics can be used to supplement utility function to create a more boundedly rational decision making within an ABM and therefore improve upon the realism of the decision making process (Haggith et al., 2003; Le et al., 2008). The conceptual framework by which agents make decisions is also of importance when attempting to accurately model the human decision making process. In perfectly rational agents there is a simple relationship between stimulus and an output: if this, then that (Kahneman & Tversky, 1979). Ajzen proposes a framework called the Theory of Planned Behaviour (TPB) for describing the human decision making process (Figure 2)  Figure 2: The Theory of Planned Behaviour. Adapted from (Ajzen, 1991)   In essence, the TPB states that the strength of a decision maker’s intention to engage in a behaviour is predictive of the likelihood he/she will engage in that behaviour (Ajzen, 1991). 26  Furthermore, there are three contributing components that influence an individual’s intention towards a behaviour: perceived behavioural control, the individual’s attitude towards the behaviour, and the subjective norm of that individual. In addition to these three elements contributing to the intention of an individual to engage in a behaviour, if the individual does not perceive self- control over engagement in that behaviour, this lack of efficacy can override strongly positive subjective norms and personal attitudes (Ajzen, 1991). In simpler terms: if individuals do not know they can do X, they will not do X regardless of personal desire or social pressure. The Theory of Planned Behaviour has been utilized in ABMs (Kaufmann et al., 2009; Richetin et al., 2010) including one investigating water-saving innovation diffusion and its inferred effect on water savings (Schwarz & Ernst, 2009b). This implementation of an ABM to simulate residential water use represents an improvement upon the decision making process modeled in other similar ABMs in four significant ways: (1) the agents in the model are implemented as active agents that can adapt to their social setting and environment, (2) agent decisions are heuristic-based within the model, (3) the decision making process is rooted in the TPB framework, and (4) agents’ multi-faceted personalities are informed using empirically based attitude survey results. In consideration of the attitudes that act as contributors to actions in the TPB, agent personality traits are heterogeneous and are informed by empirical data to more accurately represent the residential water use behaviours in Kelowna, BC.  2.3.1 Creating heterogeneous agents through statistical clustering  In experiments where individuals within a similar environment are required to make a decision, it has been shown that vastly different outcomes can arise from different decision 27  makers (Evans et al., 2001; Manson & Evans, 2007). These findings highlight the importance of considering the heterogeneity inherent in human decision makers. Conducting surveys to elucidate the differences in attitudes and behaviours is perhaps the most common method used to create empirically based, variable decision makers in a model (Le et al., 2010, 2008; Schmitzberger et al., 2005; Smajgl et al., 2011; Valbuena et al., 2008). Survey results can then be used to formulate a finite set of typologies or archetypes to give modelled agents variable behaviour or preferences (Valbuena et al., 2008). This finite set of typologies can be created from survey results through popular methods such as cluster analysis (Gorton et al., 2008; Guillem et al., 2012; Karali et al., 2011; Smajgl et al., 2011; Valbuena et al., 2010). Statistical clustering is a method used to cluster multidimensional data into a finite number of subset groups. The general goal of cluster analysis is to minimize differences within the cluster while maximizing differences between clusters. This clustering is done using unsupervised machine learning classification. In supervised machine learning, the classification algorithm is given a training dataset which contains the class membership for each object (Han et al., 2012b). In unsupervised learning, training datasets with class membership information are not provided and instead the machine learning occurs through iterative observation of the relationships between points in multidimensional space.  The method by which the relationships between points are assessed and clusters are created depends on the method of clustering used. There are four basic clustering methods: partitioning methods, hierarchical methods, density-based methods, and grid-methods. Of the four methods, partitioning and hierarchical methods are the most commonly used in agent based modeling (Fernandez et al., 2005; Kaufmann et al., 2009; M. M. Bakker & van Doorn, 2009; Smajgl et al., 2011). An in depth review of these methods can be found in Han et al. (2012).  28  Partitioning methods, which include k-means and k-medoids separate n objects into k clusters where k ≤ n. The partitioning begins by selecting k random centroids within the multidimensional space (in the case of k-medoids these centroids are objects selected from the dataset). Each object is then classified as belonging to the cluster of the closest centroid. Next, the centroids are recalculated within the cluster and objects are reclassified into clusters based on their proximity to their new centroids. This process is iterated until the centroid locations are steady between iterations, resulting in the finalized clusters (Berkhin, 2006). There are three main limitations of this method: k must be supplied to the algorithm, clusters tend to be spherical, and the method lacks effectiveness at large scales (Han et al., 2012b) Hierarchical methods are either agglomerative or divisive in their process of cluster creation. In both cases a dendogram is created that organizes n objects into k clusters ranging from a single cluster containing all objects, to k = n clusters with only a single object in each cluster. Divisive hierarchical clustering is a “top-down” approach in which a single supercluster is cut into successively smaller subclusters. The criteria for splitting the cluster depend on the statistical package being used and can range from size-priority splitting, temporary objectives, average similarities, cluster cohesion, or various stopping criteria (Ding & He, 2002).  In agglomerative hierarchical clustering the algorithm starts by assuming each object to be its own cluster and then successively merging the two closest clusters until all objects are in a single cluster. The way that distance is measured between clusters depends again on the statistical package that is used. The three most common distance measurements include single linkage distance (minimum distance between the two closest objects across two clusters), complete linkage distance (maximum distance between two objects across two clusters), average linkage (the average distance between objects across two clusters), or Ward’s method which 29  minimizes intracluster variance (Ding & He, 2002). A limitation of hierarchical clustering is subjective dendogram cutting. Hierarchical clustering does not indicate the ideal number of clusters and so it is up to the researcher to determine the desired number of clusters. Additionally hierarchical clustering is susceptible to outliers and objects are bounded by splits or merges that occurred early in the process and cannot be reclassified later despite potentially being more similar to other clusters than their own (Balijepally et al., 2011). The limitations that the popular hierarchical and partitioning methods impose and the subjectivity involved in determining the ideal number of clusters necessitates the use of a more advanced method in this case. In the basic methods described above, each object is assigned to only a single cluster, whereas a probabilistic model-based clustering method assumes that clusters are potentially overlapping distributions and objects are assigned a probability of belonging to each probabilistic cluster (Han et al., 2012a). These probabilistic results are then fed through an expectation-maximization (EM) algorithm. The EM algorithm iterates through an expectation step and a maximization step. In the expectation step, the centroids for the probabilistic clusters are computed and each object is temporarily assigned to the cluster of the closest centroid. The maximization step then recalculates the centroid to maximize the expected log-likelihood. These steps are repeated until the algorithm converges (Fraley & Raftery, 2002; Han et al., 2012a). It has been shown that by applying probabilistic model-based clustering methods repeatedly for a range of potential counts of clusters, and by comparing the Bayesian information criterion (BIC) values for each  model, the model of best fit can be found by selecting the method with the lowest BIC (Berkhin, 2006; Fraley & Raftery, 2002, 1998). In a comparison of three statistical packages that combine model-based clustering, EM algorithms, and BIC analysis 30  the “Mclust” package in in the statistical package R was found to outperform the others with the added benefit of being open source (Haughton et al., 2009). Therefore, the Mclust package in R (Fraley et al., 2012) was used to perform cluster analysis of personality types obtained from self-reported data.   31  Chapter 3 Methods  3.1 Study Area  The area of study is the city of Kelowna in the province of British Columbia, Canada (Figure 3). Kelowna is the largest city in the interior of British Columbia, and it is known for its economy being largely based in agriculture and tourism. A high standard of living contributes to the city and surrounding area being the fourth fastest growing metropolitan area in Canada (Statistics Canada, 2012b). In addition to this quick expansion taxing local resources, the urban population of Kelowna has access to some of the least available freshwater in Canada, meanwhile residents are also amongst the heaviest users per capita in the nation (Summit Environmental Consulting Inc., 2010). Extreme use rates and low availability add stress to the water limited semi-arid desert ecosystems in the region, ecosystems that are amongst the rarest and most threatened in Canada (Austin et al., 2008).   Kelowna is located within the larger Okanagan Valley, on the shores of Okanagan Lake. The valley bottom has an elevation ranging from 250 – 500 metres above sea level (masl) while the high elevation plateaus are between 1200 – 1500 masl (Spence & Hedstrom, 2015). The 1981 – 2010 mean temperature minimum occurs in December at -2.6 °C and the maximum in July at 19.5 °C with the unadjusted annual precipitation for the same timespan being 390 mm, making the region a semi arid region  (Environment Canada, 2015). The amount of precipitation that remains in the landscape is considerably less due to evaporation and evapotranspiration, with rates of evaporation reaching 725 – 835 mm a year from the lake surface (Spence & Hedstrom, 2015). The large size of Okanagan lake (384 km2) likely contributes to residents overestimating the availability of water, but with high rates of evaporation and 14 municipalities on the lake 32  shore, the water flowing into the basin is nearly fully allocated to current water liscenses (Cohen et al., 2004; Jatel, 2008)  Figure 3: City extent of Kelowna, B.C. Canada. Source: Esri, DigitalGlobe, GeoEye, Earthstar Geographics, CNES/Airbus DS, USDA, USGS, AEX, Getmapping, Aerogrid, IGN, IGP, swisstopo, and the GIS User Community. 33  3.2 Data  One of the strengths of modeling is the ability to pull together a range of data types to produce simulations that are cross-disciplinary in nature. Although this requirement of good data can be a limitation if data is unavailable, water use and development data in Kelowna, BC is readily available to the public and researchers alike. A table detailing the types of data used in this study, their function, and their sources can be found below: Table 1: Datasets used in this thesis work, their function, and source Data Function Source Household level data on water perceptions, attitudes, and behaviour (486 respondents)  Inform heterogeneous personality types that contribute to water use behaviour within the model Kelowna Water Demand Survey conducted by Dr. John Janmaat. 2014 Future land use data + legal parcel data for Kelowna (GIS polygons at metre resolution) Create the spatial landscape of the model with land-use types and designate areas of future development City of Kelowna Open Data Catalogue. Accessed 2015. (http://www.kelowna.ca/CM/Page3936.aspx) Lot size requirements, maximum dwelling counts and minimum/maximum lot coverage for land use types  Inform the parcels with land-use type specific attributes  City of Kelowna Zoning Bylaw No. 8000 – Section 13. Revised August 10th 2015 version. (http://www.kelowna.ca/CM/page2561.aspx) Number of occupants per household/dwelling Inform the number of occupants per household or dwelling (Statistics Canada, 2013a) Household level water use data for residents serviced by Kelowna Water Utility Validate model results against real data City of Kelowna Water Utility. March 2012 – March 2014 data (restricted access)  3.2.1 Agent personality traits  The data used to create the personality traits and inform the personality types in this model come from a water use and attitudes survey conducted by Dr. John Janmaat from the summer of 2009 through to the fall of 2010 (Janmaat, 2015). The survey was developed by Dr. 34  Janmaat in collaboration with local water use experts and refined through preliminary testing groups. The survey is comprised of 104 questions ranging from ones regarding demographics, to household behaviour, to political preferences. In total 517 individuals of the 2273 that were contacted responded to the survey. Of those that did respond, 486 had fully completed the sections of the survey relevant to this study.  The questions described in Table 2 were isolated from the survey data to inform the personality traits within the model. In response to the statements in Table 2, respondents were asked to indicate their agreement with the statements on a Likert scale from 1 (strongly agree) to 7 (strongly disagree): Table 2: Component questions informing agent personality traits Model personality trait Questions informing trait NEP (New Ecological Paradigm)  15 standardized statements (see Table 9 in Appendix B) ENTITLEMENT “People should reduce their water use, even if it costs them money to do so”   “People who waste water should pay more for it” FORESIGHT “There is plenty of water in the Okanagan to meet the needs of all users”  “There is plenty of water in the Okanagan to meet the needs of all users to at least the middle of the century (2050)”  “Climate change is going to cause more water problems in the Okanagan” (Likert value reversed)   The New Ecological Paradigm (NEP) is a metric developed to quantify an individual’s pro-ecological worldview (Dunlap & Van Liere, 1978). At the time of the metric’s inception, the dominant attitudes of society were ones devoted to growth, privatized rights, and faith in 35  abundance. This general set of attitudes were termed the “Dominant Social Paradigm” or “DSP”. The NEP was a world view that was seen as the opposite of the DSP and so the NEP metric was created to measure an individual’s placement on the spectrum from NEP-like (ecologically oriented) to DSP-like (anthropologically oriented). Although several other measures to measure environmental attitudes have been developed, the NEP is by far the most widely used in the literature (Stern et al., 1995). The 15 statements of the survey, with which respondents can agree or disagree, are centred on the rights of humans over the environment, the present and future state of the planet, and the relationship between humans and our environment (Appendix B). More explicitly, Dunlap and colleagues see the NEP as containing five facets: balance of nature, ecocrisis, antiexemptionalism, limits of growth, and antianthropocentrism (Dunlap, 2008). Although the NEP is sometimes treated as having separate subscores for each facet, Dunlap and others have indicated that reducing dimensionality and treating the score as a single value is more powerful than treating them as separate (Amburgey & Thoman, 2012; Dunlap, 2008). As a result, the 15 questions – with responses on a scale from 1 to 7 – were averaged (once inverse question scores were reversed) and used as a single score. This NEP score used to influence an agent’s outdoor water use, response to precipitation, and decision to purchase water saving technology as described in Appendix C. The two remaining agent personality traits, ENTITLEMENT and FORESIGHT, were created to additionally differentiate heterogeneous agents’ attitudes and beliefs from one another. These two personality traits are informed by response data from questions asked in Dr. John Janmaat’s water attitudes survey. While the NEP captures the general ecological worldview of the respondents, these two variables were created to capture the opinions of respondents about 36  their relationship with water locally. Similar to the questions that contribute to the NEP score, the questions for the ENTITLEMENT and FORESIGHT scores were answered on a scale of 1 to 7 to indicate the respondent’s agreement or disagreement with a statement. Like the NEP score, the questions that contribute to these other two attributes were an average of the responses for each constituent question. The ENTITLEMENT trait is informed by two questions that elucidate the respondent’s opinion on the inherent right to water, in particular whether one should pay for its use or not. Quantification of this attitude was seen as necessary, as an individual’s belief in the inherent value of water is likely a contributing factor to the way that individuals decide to use water, and may play a part in Kelowna’s high use rate. Within the survey results, the ENTITLEMENT score of a survey respondent was strongly correlated (p-value > 0.0001) with the number of self-reported water saving behaviours the individual engaged in. Within the model this trait has a direct influence on indoor water use, as well as the rate at which an agent may choose to neglect watering restrictions, and plays a partial role in an agent’s decision to purchase water saving technology.   The FORESIGHT trait variable is informed by three questions related to the medium and long term availability of water in the region and the possibility that climate change will have a negative effect on water availability. Quantification of this personality trait was seen as necessary because an individual’s ability to understand unsustainable water practices is a likely contributor to their water use behaviour. Kelowna’s low availability of water, compounded by increasing demand from development, and the expected effects of climate change in the area, makes it clear that current trends will lead to the demand to supply ratio increasing significantly. 37  This trait influences outdoor water use, and the rate at which an agent will discount the future benefit of upgrading their water saving technology.  3.3 Model description  The agent-based model simulates the response of residential water users to the implementation of a range of demand side management strategies. The model also simulates the social network of the residential water users, and the effect that social interactions can have on agent personality and behaviour. A full description of the model in the ODD + D format (Müller et al., 2013) – an enhancement on the standardized ODD method (Railsback & Grimm, 2012) - can be found in the “Info” tab of the accompanying model or in Appendix C.  3.3.1 Scale and Series of Events For the purpose of this thesis, simulations were run over the course of one year in daily increments. The spatial extent of the model encompasses the Kelowna city boundary at a resolution of 400 x 400 pixels or patches. Each patch represents an area of 69 m2.  Each agent in the model represents the individual that irrigates the property on which the agent resides. The irrigating agents also represent the additional residents that live on the property. For example, a single family home would have one irrigator with up to five other indoor users represented, while an apartment building would have a single irrigator with significantly more indoor water users. Each agent accounts for a number of dwellings on their property and the associated number of residents within each dwelling. The number of dwellings on a property is dependent on the number of properties expected, based on the land-use type. 38  This method of aggregating all residents on a property into a single agent was used to reduce computational load. This aggregation therefore assumes that residents on a property all exhibit similar water use behaviour. The model is initiated by placing an irrigator agent on each habitable residential parcel. Each agent is then assigned a personality type based on the proportion of each personality type observed in the survey data. Each agent is then assigned trait values from the associated distributions of trait values for each separate personality type. The assignment of personality types is done in either a random or clustered spatial distribution dependant on which type was specified in the user interface (an example of the user interface can be seen in Figure 27 of Appendix B). While the model is running, the following events occur each simulated day (in the order given): 1.1 Precipitation occurs according to 2012 data 1.2 Agents use water indoors 1.3 Agents use water outdoors 1.4 The day is advanced one time step Each week the following events occur: 2.1 Agents assess their ability to upgrade their water use technology 2.2 Agents assess their ability to lease their irrigation rights to a centralized irrigator 2.3 Agents have the opportunity to influence their social network At the end of each month the following occurs: 3.1 Residents pay for their previous months’ water use 3.2 The water infrastructure is maintained 39  3.3 The total water use is plotted and visualized on each property  A more detailed sequence of events and actions is shown in Figure 4 and is described in greater detail in Appendix C.  Figure 4: Flowchart of daily, weekly, and monthly events.  40  3.3.2 Agent Decision Framework  Decisions are made by the agents using the framework of the Theory of Planned Behaviour (TPB) (Figure 2). An agent’s decisions to engage in an action are therefore moderated by its attitudes towards the action, its subjective norms, and its perceived behavioural control. Agent attitudes   Agent attitudes are separated into three traits as seen in Table 2. In simulations where agent attitudes contribute to behaviour, agent traits influence behaviour in the following ways (only the first relationship was observed in the data, the remaining are assumed due to lack of relevant data):  Agents with higher ENTITLEMENT scores will use more water indoors (there was a strong correlation between survey respondent’s ENTITLEMENT score and the number of self-reported indoor water saving behaviours he/she engaged in [p < 0.0001]).  Agents with higher ENTITLEMENT scores are less likely to upgrade to water saving technologies for non-monetary reasons.  Agents with higher NEP are more likely to reduce or forego their irrigation when there is sufficient precipitation for the landscape’s demands.  Agents with either high NEP or high FORESIGHT scores will reduce their assessed outdoor irrigation needs by 25% each day.  Only agents with NEP scores that surpass a threshold will consider upgrading to water saving technologies for non-monetary reasons.  In their assessment of the monetary benefit of upgrading to water saving technologies, agents with higher FORESIGHT will discount the future benefit of the upgrade less. 41  Subjective Norms  The influences of subjective norms are incorporated into the model in three distinct ways: 1. An agent’s perceived cost of an irrigation technology upgrade is a function of the prevalence of this behaviour in the agent’s social network. If nobody in their network has upgraded their irrigation, the upgrade is seen as 10% more costly. For each person in their network that does have the upgrade, the agent will perceive the cost as 10% less than the cost of being the first in their social network to upgrade.  2. Agents have the potential to lease their irrigation rights to a centralized irrigator that then takes over the responsibility of irrigation and technology upgrades. The rate of conforming to a centralized irrigation program depends on the number of individuals in an agent’s social network who have done so. For each individual in an agent’s network that has leased its watering rights to a centralized irrigator, there is an additional 5% chance of conforming to this behaviour. Once an agent leases their right to irrigate to a centralized irrigator, they skip the outdoor watering phase.   3. Agents have the potential to influence the attribute values of those in their social network. If an agent has sufficiently similar trait scores to another in its social network, it has a 50% chance to slightly modify the other's score to be closer to its own.  Perceived Behavioural Control An agent's perceived behavioural control is represented by a variable called "learning" within the model. The higher the value of the “learning” variable, the more likely an individual is to learn of its ability to either lease its watering rights or upgrade its water saving technology. Even if the agent has an attitude that is strongly predictive of a behaviour or action, or it has a 42  social network that is supportive of that behaviour, it is unable to engage in that behaviour if there is no perceived behavioural control (e.g. the agent has not learned that it can do X action).  It is important to note that although there is a comparison of cost and benefit of agent action, there is not a budget constraint for any action, i.e. all agents can afford to engage in any behaviour. The initial value for the “learning” variable is set in the user interface, and there are two simulation events that can increase this value. If an agent is in proximity to an education source, then it will increase its learning score relative to distance from the source. Additionally, if an agent has its water saving technology upgraded on its parcel, the “learning” score of each agent, within the social network of the focal agent, will increase by 5%.43  3.3.3 Scenario Descriptions In general, the scenarios tested within this thesis fall into three categories, (1) growth scenarios, (2) regulatory scenarios, and (3) social scenarios. Growth scenarios of restrictive growth patterns, densification, and urban sprawl test the hypothesis that restricting urban growth will lead to a reduction in overall water demand given an expanded population. Regulatory scenarios determine which policy implementation will be most effective within the system, and whether there are synergistic effects from combining multiple regulatory measures. Finally, the social scenarios let us test the hypothesis that increasing the proportion of a minority opinion group can reduce the social influences of other groups on that minority. In this case, the minority group of interest is water saving individuals. Each scenario variant was modeled twelve times to allow for comparison of average model results. Twelve was chosen as the number of replicates because model runs consistently converged on similar results, and therefore twelve was a large enough sample size for statistical testing and calculations of means, standard deviations, and other descriptive statistics. Secondarily, twelve simulation runs were selected because of the restrained time available to run simulations, and the easy divisibility of twelve for computers that were capable of only running two, three, or four simultaneous simulations. The population increase simulated is proportional to that expected to be seen in the city of Kelowna by 2030 (City of Kelowna, 2013; Statistics Canada, 2012a). There is expected to be a population increase from 117,000 people in 2011 to 161,000 in 2030, which translates to a 38% increase in population size. The spatial mechanism by which these simulated new residents fill the landscape changes between growth scenarios as described in Table 3, essentially either filling existing landscape parcels or creating new landscape parcels in areas designated for future urban growth. 44  Table 3: Growth scenario descriptions Scenario Description Current Agents can only occupy land that is currently designated as habitable land. There is no simulated growth. This represents the baseline scenario of the model Urban Sprawl Areas designated for future urban development are partitioned into single/two family parcels. Agents can occupy the same area as the “Current” scenario in addition to this newly partitioned area. A population 38% larger than the “Current” scenario is simulated. Densification Areas designated for future urban development are partitioned into medium density residential parcels. Only 50% of the land is converted (this value is adjustable in the user interface). Agents can occupy the same area as the “Current” scenario in addition to this newly partitioned area. A population 38% larger than the “Current” scenario is simulated. Restrictive Growth Areas designated for future urban development are left undeveloped. One percent of single/two family home parcels are converted to medium density residential parcels. A population 38% larger than the “Current” scenario is simulated.   At the resolution of 400 x 400, developing 100% of the future urban reserve into single/two dwelling parcels to accommodate new residents (urban sprawl) results in an 11% increase in the number of habitable parcels and a 6% increase in the maximum number of dwellings on the landscape (Table 4). In contrast, “Densification” leads to only a 4% increase in habitable parcels, markedly less than the “Urban Sprawl” scenario due to only half of the future urban reserved landscape being developed and the fewer new dwellings being placed on larger, medium density residential lots. Due to a higher capacity of dwellings on medium density residential parcels as opposed to single/two house parcels, “Densification” leads to the highest potential maximum number of dwellings in the urban reserve, while the “Restrictive Growth” and “Urban Sprawl” have relatively the same capacity of dwellings as each other.   45  Table 4: Summary of the number of individuals and dwellings being represented in a 400x400 pixel simulation of Kelowna, BC for four growth scenarios. Values with an * or ** represent values that are variable between simulation runs and so these values are means of 12 simulations. Values marked with ** represents values that are not statistically different from others that share this symbol. Values in parentheses represent the percentage of increase compared to the “Current”” growth scenario.  Count of Individuals Represented Irrigators/Parcels  Occupants Dwellings Max Possible Dwellings Growth Scenario         Current 5724 32984* 13725 18918 Urban Sprawl 6329(+11%) 45522** (+38%) 18918(+38%) 20128(+6%) Densification 5967*(+4%) 45500** (+38%) 18918(+38%) 24744*(+31%) Restrictive Growth 5724(+0%) 45471** (+38%) 18918(+38%) 19952(+5%)  In the “Restrictive Growth” scenario, no new land is used for development and parcels that transition from one/two dwelling parcels to medium density residential parcels do not change the size of their irrigable land. In scenarios with newly developed residential land (densification and urban sprawl), new parcels are created to be a size that ranges from the minimum described in the Kelowna zoning bylaws up to the mean currently found in the real-world data. Limiting new parcels to be between the minimum and the currently observed mean was done because it is assumed that the high cost of future expansion will limit the development of overly large parcels, and instead favour smaller parcels. This pattern is observed in the data as well with small parcels being favoured, and indeed some parcels being smaller than the current minimum allowable lot size, likely an effect of lower past minimum allowable lot sizes (these lots were adjusted to the minimum in an effort to counteract some of the underrepresentation of these small lots).  The policy scenarios explored include subsidies, education, leasing (centralized irrigation), ration level multi-tiered pricing, watering restrictions, and combinations of these 46  scenarios, as described in Table 5. Each of the future policy scenarios include the most conservative form of population growth, Restrictive Growth, as described in Table 3. In addition to these policy scenarios, a control scenario was run that simulated population growth under the Restrictive Growth pattern, but with no additional policy implementation.  Table 5: Policy scenario description. All policy scenarios are run using the “Restrictive Growth” growth scenario  Scenario Description Control No water management policies in effect. Subsidies The cost of upgrading landscape is reduced by 90% Education/ Education+ Education sources are placed throughout the landscape in areas of high agent density. These education sources increase the likelihood that agents in the vicinity will learn about water conservation measures and improve the agent personality traits. “Education” has 12 sources while “Education+” has 24 sources LeaseGovt Agents have the opportunity to lease its outdoor irrigation rights to a centralized irrigator. The agent no longer irrigates on its property as this responsibility is overtaken by the centralized irrigator. The centralized irrigator pays the parcel owner a monthly fee for this right. LeaseHouse Agents have the opportunity to lease its outdoor irrigation rights to a centralized irrigator. The agent no longer irrigates on its property as this responsibility is overtaken by the centralized irrigator. The parcel owner pays the centralized irrigator a monthly fee for this service. Rations The volume of water allowable in each tier of the tiered pricing scheme is reduced to levels that occur in New South Wales, Australia. Agents can still choose to water, but the pricing tiers are much steeper than in other scenarios. Restrictions/ Restrictions+ Houses are restricted to watering their lawns only a specific number of days in the week. There is a 75% compliance rate for agents maintaining this frequency. “Restrictions” allow watering 4x/week while “Restrictions+” limit irrigation to 2x/week. Combo1RS Combines the policies of rations and subsidies. Combo2RE Combines the policies of rations and education. Combo3RES Combines the policies of rations, education, and subsidies. Combo4RR Combines the policies of rations and restrictions. Combo5RR+ Combines the policies of rations and increased restrictions. 47  Chapter 4 Results & Discussion  Data analysis in this thesis was completed with two statistical suites. The cluster analysis of personality types was conducted in R using the Mclust package (Fraley et al., 2012). Model results and validation were analyzed using JMP®, Version 12 (SAS Institute Inc. 1989-2007).  4.1 Cluster Analysis of Water Demand Survey Data Cluster analysis of the survey data was necessary to find common response patterns to the survey and therefore determine the number of generalized personality types that are represented in the population. First, NEP, ENTITLEMENT, and FORESIGHT scores were calculated for all 486 respondents. The “Mclust” package in R was then used to test numerous possible probabilistic model-based clustering methods to find the most suitable model to describe the clusters found in the data (Fraley et al., 2012). Bayesian information criterion (BIC) is a measure used for model selection that indicates the ability of a model to fit the data, with a score closer to 0 being favorable (Fraley & Raftery, 1998). The BIC scores for the clustering models tested are compared in Figure 5 48   Figure 5: Bayesian information criterion (BIC) comparison for 14 cluster models in R. Within the legend, each label describes the multidimensional shape of the model represented. EII" = spherical, equal volume "VII" = spherical, unequal volume "EEI" = diagonal, equal volume and shape "VEI" = diagonal, varying volume, equal shape "EVI" = diagonal, equal volume, varying shape "VVI" = diagonal, varying volume and shape "EEE" = ellipsoidal, equal volume, shape, and orientation "EVE" = ellipsoidal, equal volume and orientation  "VEE" = ellipsoidal, equal shape and orientation  "VVE" = ellipsoidal, equal orientation  "EEV" = ellipsoidal, equal volume and equal shape "VEV" = ellipsoidal, equal shape "EVV" = ellipsoidal, equal volume  "VVV" = ellipsoidal, varying volume, shape, and orientation.    49  The Mclust package found that the three best fit clustering models all fit the data best when there are three clusters (VEE: BIC = -3904.9, VEV: BIC = -3907.4, VVE: BIC = -3907.8). The abbreviations that represent each tested model indicate the shape of the clusters (see the caption of Figure 4 for details). The best fitting model, VEE, suggests that clusters that are ellipsoidal, equal shape, and equal orientation while VEV indicates the same without specifying orientation and VVE indicates the same shape and orientation without indicating volume. Therefore, there is agreement between the tested models on the shape, volume, orientation, and number of clusters present in the data.   Survey respondents were categorized as belonging to the cluster assigned by the VEE model in the Mclust package. With each respondent now having an associated ENTITLEMENT, NEP, FORESIGHT, and personality cluster score distributions were created of the personality trait values within each cluster to inform agent attributes within the ABM (Figure 6). A qualitative description of the three personality types and associated attitudes is found in Table 6.  50   Figure 6: Personality types and their associated traits. Dots represent the mean while the whiskers represent the 95% confidence interval. P1 = “water wasters” (N = 83), P2 = “average heavy users” (N = 359), P3 = “water savers” (N = 44).  Table 6: Cluster groups and their attitude and general description Cluster Attitudes General description P1 Perceives a low chance of water shortages in the future The “water wasters”. Individuals who do not think there will be a problem with water supply in the future. They are less concerned with ecological impacts of water use and are instead focused on human development. They are less likely to have an opinion on whether or not heavy users should pay more for water.  Neutral on whether heavy users should pay more The most oriented towards human development  P2 Perceives a moderate chance of water shortages in the future The “average heavy user”. Individuals who think there may be issues with increasing demand and reduced supply in the future. They are aware to some degree of the ecological impacts humans have on the environment and therefore they believe heavy users should pay. Believe somewhat that heavy users should pay more Slightly ecologically oriented P3 Perceives a high chance of water shortages in the future The “water savers”. Individuals who are strongly environmentally minded and understand the significant impact that our activities have on our surrounding environment. These individuals naturally believe heavy users should pay more.  Strongly believes that heavy users should pay more Very ecologically oriented  51  4.2 Validation  Before providing results for the scenario testing that was done with this model, it must be established that the model reliably recreates the patterns that are observed in the real world system. Water use data was therefore obtained from the city of Kelowna and used to validate the model.   The data that were used to validate the model were provided by the city of Kelowna, BC and represented 14,524 water users. Only users that had recorded water use for each month from April 2013 to March 2014 were used for comparison purposes since agents within the model used water in each of the 12 months represented in simulation. In total 12,935 observations were used for comparison. Each observation represents a household’s water use for a year, segregated into monthly use rates reported in m3 of water or 1000 L increments. Real data were compared to 12 highly consistent model runs that represented a year of water use in a baseline scenario with no population growth, regulatory introductions, or educational programs.  Visually, the distribution of water use through the year in the real data (Figure 7) and modeled data (Figure 8) are broadly similar. The months of May and October show the largest discrepancy between the real and modeled data. These discrepancies are a result of the on/off nature of irrigation within the model. Whereas there is a ramping up of outdoor use rates between May and June and a ramping down between September and October in the real data, the model has no progressive increases or decreases and instead runs in a binary fashion for the beginning and end of the outdoor watering season. The ratio of water use from peak months to winter months, however, is quite similar between real and modeled data. In the real data the ratio of September mean use to January mean use is 4.3:1 while in the model this same ratio is 4.2:1 52   Figure 7: Residential water use for the City of Kelowna Water Utility in 2013.     Figure 8: Modeled water use for Kelowna, BC for each of the 12 months.  Values are based on the averages of 12 model runs of the baseline, current-day scenario. 53     The raw data for both the modeled and real data was converted into z-scores for statistical comparison of distribution shapes. Converting the use data to Z-scores allows us to compare the distributions of real data to model data more easily because each distribution is transformed to a distribution that maintains the relationship between the points within the dataset but with a mean of 0 and a standard deviation of 1. The Wilcoxon each pair test was used to compare the real data to each of the twelve model runs. The Wilcoxon each pair test accounts for multiple comparisons. The results show that each of the baseline model runs data are statistically the same as the real data (Figure 9). These results indicate that this model can reliably reproduce the general water use patterns observed in real data.  Figure 9: Wilcoxon each pair comparison of the z-scores of real data and twelve independent baseline model runs. Simulations 1 – 12 represent a single run of the model each. No pairs of observations are different from one another. 54  4.3 Growth Scenarios  In this section, the effect of population increases and a range of growth scenarios on the water use behaviour of agents within the ABM are assessed. The comparison of water use across scenarios has been done in terms that are relative to the “Current” scenario, which represents the control scenario. In the absence of any policy implementation, the total water used within the simulation is 35% higher in the “Urban Sprawl” scenario, 26% higher in the “Densification” scenario, and 17% higher in the “Restrictive Growth” scenario (Figure 10).   Figure 10: The percent increase in water use for three modeled growth scenarios relative to modeled current use without any policy implementation. Each bar represents the percent increase in each end use relative to that end use of the modeled current use. Bars represent the mean of 12 simulations and the whiskers represent the standard deviation.  55  These results suggest that by 2030, there will be little increase in outdoor water use under restrictive growth, under densification a 17% increase in outdoor use (0.9% increase annually), and under urban sprawl a 35% increase in outdoor use (1.8% increase annually). Previous methods of estimating residential demand in the Okanagan suggest that restrictive growth and densification (previous studies considered these two to be a single group with no new development) will result in only a 0.1% increase annually, while urban sprawl will result in a 3.4% increase annually (Polar Geoscience Ltd., 2012). Previous results do not account for the social dynamics associated with future water use and therefore may overestimate urban sprawl demands in the region by not considering the self-governance and feedback loops that could influence increased demand response to a limited supply. Conversely, the model described in this thesis uses simplified and more conservative daily landscape water requirement estimates that do not account for climate change and could therefore produce more conservative estimates of increasing demand. Future work run at improved resolutions and with more realistic landscape water requirements would improve this model’s ability to estimate outdoor requirements more accurately. The total indoor water use rose uniformly by 37% across the growth scenarios, equivalent to a rate of 2% per year. These results are consistent with previous estimates of an increase of 0.9% to 2.3% annually for indoor water demand in the region (Polar Geoscience Ltd., 2012). This rate is consistent with the growth rate at 38%, and is also consistent with the price inelasticity of indoor water use observed in literature (Galán et al., 2009; Russell & Fielding, 2010). The total, indoor, and outdoor water use for each scenario is shown in Figure 11. The modelled ratio of current outdoor water use to current indoor water use fits well with the trends 56  seen in validation data (Figure 7). The validation data suggests that Kelowna users’ outdoor water use accounts for 52% of their yearly total uses and their indoor use accounts for 48% of their total use. Results in Figure 11 indicate outdoor water use accounts for 55% of total water use in the baseline “Current” scenario. The “Restrictive Growth” scenario indicates that outdoor use will make up 47% of future water use, while “Densification” will result in 51% of all water use occurring outdoors, and “Urban Sprawl” or the business as usual development resulting in 54% of total water use occurring outdoors. Outdoor water use makes up a larger portion of the total water use as the amount of new landscape developed increases.   Figure 11: Modeled rates of current water use and three growth scenarios. Bars represent a mean of 12 simulations while the whiskers represent the standard deviation. 57   These proportions of outdoor to indoor use are in contrast to estimated Okanagan Basin wide values which suggest that, currently, only roughly one quarter of total water use should be for indoor use (Summit Environmental Consulting Inc., 2010). This discrepancy between results may be due to two factors. The validation data that are used in this model are from a watering district that has fewer large lots than would be found in the Okanagan Basin at large, thus outdoor water use may be reduced compared to the Basin average. Additionally, due to the coarse spatial resolution of this model, single dwelling parcels are underrepresented (Table 8), resulting in a reduced proportion of irrigators to occupants and therefore conservative total outdoor use estimates.  Irrigators in the “Urban Sprawl” and “Densification” scenarios have significantly higher usage per irrigator compared to the baseline and “Restrictive Growth” scenarios (Figure 12). The observed difference in per irrigator usage between scenarios is a result of the development of new landscape areas. With the addition of new parcels to the landscape that either meet or exceed the minimum allowable lot size (as opposed to many legacy lots being below this minimum size requirement) the mean lot size of habitable parcels increases as irrigators are added to the landscape. “Current” and “Restrictive Growth” scenarios have an average lot size of 2444 m2, “Densification” scenarios have an average of 2792 m2, and “Urban Sprawl” scenarios have an average of 3050 m2.   58    Figure 12: Percent increase in outdoor water use per irrigator within a growth scenario relative to modeled current water use. Bars indicate the mean of 12 simulations while the whiskers indicate the standard deviation. Results in Figure 12 indicate the importance of urban growth management. With new property owners potentially having larger lawns to water than irrigators of many grandfathered parcels, the resulting increase in per irrigator use could contribute to a social norm of increasing usage rates outdoors.  4.4 Policy Scenarios  The policy scenarios tested within this model demonstrate the effectiveness of watering restrictions and centralized irrigation leasing, while highlighting the variable impact of increasing water pricing depending on other regulations simultaneously implemented with price increases. The effects of high vs. low pricing on outdoor water use for each policy scenario are 59  presented in Figure 13. As might be expected, increasing pricing leads to decreased water use for all regulatory scenarios. The greatest effects of increasing prices occur for the “Rations” scenario, the combination scenarios that include rations, and the “LeaseHouse” scenario.     Figure 13: Reduction in outdoor water use due to a 300% increase in 2015 watering rates/prices for Kelowna, BC. Results are shown for just price increase (control group) and price increases in combination with other regulation. Whiskers indicate the maximum and minimum values observed, the box itself represents the first and third quartiles, and the middle line represents the median. A black dot outside the box plot indicates an outlier. Groups that do not share a letter are significantly different. (N = 12, α = 0.05)  B B B AB AB AB A C C CD DE DE E E 60   The difference observed between low and high pricing in the “LeaseHouse” scenario can be explained by the way an agent assesses its leasing opportunity. In these scenarios an agent can pay to be relinquish its right to irrigate and shift the responsibility of lawn care to a centralized irrigator. When the price of water is low, the amount that the agents are required to pay is higher than their bill and so they do not lease their irrigation rights. On the other hand, when prices are high, it is cheaper to pay to lease the watering rights than to pay to keep their lawn properly watered and so they lease, resulting in the centralized irrigator upgrading the landscape, and therefore less water being used.  The difference between water usages for high and low pricing in scenarios that have rationing occurs through a different mechanism. In the rationing scenarios, the volume of water allotted to each tier of pricing is reduced significantly. This leads more agents to reach the highest tier of pricing more quickly and so scenarios with a high pricing structure lead agents to begin considering the cost of their water use earlier on and to a higher degree than in scenarios with a low pricing structure.   Changing pricing structure has no effect on the indoor water use in any of the scenarios tested. This suggests that indoor water demand is relatively inelastic to pricing changes which is consistent with the literature (Galán et al., 2009; Russell & Fielding, 2010). Indoor water use was found to be relatively unresponsive to other regulatory measures as well except for the education scenarios which resulted in a 1 – 2% reduction compared to the control scenario (Figure 14). 61    Figure 14: Increase in indoor water use for a range of policy scenarios within the Restrictive Growth scenario. The control represents Restrictive Growth scenario with no policy implementation. Rates are relative to modeled current use for the same area. Whiskers indicate the maximum and minimum values observed, the box itself represents the first and third quartiles, and the middle line represents the median. A black dot outside the box plot indicates an outlier. Groups that do not share a letter are significantly different. (N = 12, α = 0.05)  The educational scenarios were the only ones found to have an effect on indoor use because education is the only policy option that can directly affect agents’ personality traits and therefore influence indoor water use. Scenarios with subsidies were only conducted for subsidies related to improving outdoor irrigation. It is possible that scenarios which assessed the effects of subsidies on indoor appliance upgrades would show some improvements on indoor water use but these scenarios are beyond the scope of the current study. DE  CD  E  AB  A  A  AB  A  AB  ABC  B  AB  BCD  AB  A  62   The effects of educational efforts on indoor water use do not extend to outdoor use rates (Figure 15). Both education and subsidies scenarios were found to have no appreciable effect on outdoor water use. Leasing watering rights to a centralized irrigator resulted in an 11% – 20% decrease in outdoor water use depending on whether the property irrigator is paying the centralized irrigator, or vice versa, while watering restrictions resulted in a 29% - 49% decrease in outdoor use compared to the control depending on the severity of restrictions.  Figure 15: Change in outdoor water use for a range of policy scenarios within the Restrictive Growth scenario. Rates are relative to modeled current use for the same area. Box plots’ whiskers indicates the maximum and minimum, the box itself represents the first and third quartiles and the middle line represents the median. A black dot outside the box plot indicates an outlier. Groups that do not share a letter are significantly different. (N = 12, α = 0.05)  A A A A C B C C C D E G F H 63   The lack of effect of education and subsidies is surprising and is likely a result of (at least in part) the way that agents make decisions to upgrade their irrigable landscape. When deciding whether or not to upgrade its irrigation technology, an agent makes a decision based on economic factors or moral personality-based factors. The results show that in the “Subsidies” scenario the economic incentive to upgrade the agent’s landscaping was not sufficient for the agent to make the change based on purely economic criteria and the moral push from the “Education” scenarios was insufficient to result in change based on purely moral/personality grounds. If perhaps the decision to upgrade an agent’s irrigation technology was made through a combination of both economic and moral factors considered together, there may have been an effect from these policy scenarios. In such a case of making decisions based on economic and moral factors, there is likely a greater chance that the propagation of an action may radiate through the social network since both personality factors and information flow through the social network.   The largest single policy effects on outdoor water use are seen in the “Restrictions” and “Restrictions+” scenarios, even with a 25% non-compliance rate. This value may change if a mechanic of enforcement or penalties was incorporated into the compliance mechanism. Combining regulatory measures was found to have mixed results. In the first three combination scenarios (“Combo1RS”, “Combo2RE”, and “Combo3RES”) that assessed the effects of combining rations with subsidies and/or education have no additional effect beyond that seen in the “Ration” scenario. The final two combination scenarios (“Combo4RR” and “Combo5RR+”) that impose watering restrictions and reductions in the tiered rate system volume structure do see cumulative effects. Combining the “Rations” scenario with the “Restriction” scenario provides a further 8% reduction beyond just the “Restrictions” scenario, while adding rations to the 64  “Restriction+” scenario provides a further 5% reduction beyond just the “Restrictions+” scenario. These results demonstrate that the policies implemented in the “Rations” and “Restrictions” scenarios interact in a nonlinear way, leading to results that are not simply the sum of their individual effects.  The non-linear dynamics seen in these two final combination scenarios can be explained by the threshold that agents must reach before starting to consider the cost of watering in their irrigation decisions. In the “Restrictions+” scenario, with a 75% compliance rate, most agents are adhering to the restrictions. With watering limited to two days per week, the addition of water rations does not cause as many agents to surpass the threshold as in the scenario which allows watering four days per week, since the agents in the more restricted system are already using less water. In essence, the agent’s water saving behaviour is reaching a saturation point. This trend could be assumed to continue in permutations of other scenarios that are adding rationing. In permutations of a policy that lead to less outdoor water use, the addition of water rations will have less of an effect than in permutations of that policy that lead to greater outdoor water use.  4.5 Social Dynamics  One of the strengths of agent-based models is their ability to simulate social dynamics between heterogeneous decision makers. In this section, results are presented for the effects of within model social dynamics during control simulations and the effects that modifying some of the initial conditions can have on the social dynamics of the model. First the effect of agent heterogeneity on water use in control and educational policy scenarios is assessed (Figure 16).     65    Figure 16: The amount of water used through the course of one year as a result of social influence for three policy scenarios within the Restrictive Growth scenario. Bars represent the mean of 12 simulations while whiskers indicate the standard deviation. Groups that do not share a letter are significantly different. (N = 12, α = 0.05)   Results in Figure 16 indicate that the addition of education efforts is reducing the social influence that is occurring within the model. In the control scenario there is a higher rate of water use due to social influence than there is in the scenarios with education available. This result makes sense in view of the personality compositions described in Table 10 and included in the heterogeneous population. The water wasters group (P1) is double the size of the water savers group (P3). Additionally, the in-between personality type (P2), which is over eight times as large as the water savers group, is more similar to the water wasters than the savers. Since there is a higher proportion of individuals with P1 and P2 personality types, there is a higher probability that any given agent will have more connections with those two personality types in comparison to the P3, water saver personality type. As a result, heterogeneous agents that are engaging in social interaction lead agents to conform to the social norm that they perceive through their social network, leading to a worsening of personality traits.  B A A 66   These results indicate that in the absence of intervention, the system of Kelowna as modeled here will trend towards negative attitudes towards water. It is possibly this mechanism that has lead Kelowna to its current state of extreme water use patterns relative to the Canadian average, despite the low water availability. Model scenarios suggest that educational programs can lessen the effect of this negative influence but even in the most intensive educational scenario tested here, there is still heavy influence of the dominant P1 and P2 personality types. These results also highlight the importance of supporting and reinforcing the desirable, minority opinions in the social system. By supporting the few water champions, their positive influence can be longer sustained in their social networks.    The water wasting influence that is occurring in the model is reflected in the changes of the agents’ personality trait scores over time. Results in Figure 17 show that the ENTITLEMENT score for the agents within the control increase through social interaction. Similarly, Figure 18 indicates that the FORESIGHT score decreases (the same trend is observed in the NEP score). In both cases, the educational scenarios negate some of this negative effect on personality traits.   67   Figure 17: Mean amounts of change in the ENTITLEMENT trait of agents for three policy scenarios within the Restrictive Growth scenario. Whiskers indicate the maximum and minimum values, the box itself represents the first and third quartiles, and the middle line represents the median. All groups are significantly different. (N = 12, α = 0.05)   Figure 18: Mean amounts of change in the FORESIGHT trait of agents for three policy scenarios within the Restrictive Growth scenario. Whiskers indicate the maximum and minimum values, the box itself represents the first and third quartiles, and the middle line represents the median. All groups are significantly different. (N = 12, α = 0.05) 68    The water-savers – representing only 9% of the population – are not numerous enough to exert a measurable water saving influence on the other agents. There is potential for Kelowna residents to transition towards a water saver mind set as climate change increases people’s exposure to drought and shifting precipitation regimes. To test the potential effects of this shift in general perceptions and social norms, the proportion of P3 agents and their spatial distribution was varied to see if either resulted in an overall effect on the way that water is used in the simulation.   The results displayed in Figure 19 show that clustering the P3 agents versus dispersing them randomly does not appear to have a significant effect on the overall water use of the agents. Increasing the proportion of P3 agents to 50% of the total population (while maintaining the P1:P2 ratio) leads to total water use that is comparable to the modeled current water use despite the 38% increase in population size.   69    Figure 19: The difference between the amounts of water for scenarios with controlled proportions of water saver agent types (P3) within the Restrictive Growth scenario. In the “Clustered” scenarios, the distribution of water savers on the landscape is clustered. In the other scenarios these agents are distributed randomly. Box plots’ whiskers indicates the maximum and minimum, the box itself represents the first and third quartiles and the middle line represents the median. Groups that do no share a letter are significantly different. (N = 12, Tukey HSD, α = 0.05)   It should also be noted that the variance around the mean changes significantly for the model run with 100% P3 agents. This change is due to a significant increase in the number of agents upgrading their outdoor watering practices as described in Table 7. This increase in upgrading can lead to higher variation in the number of agent upgrading. In some instance, a tipping point is reached and the action of upgrading propagates through the social network leading to all agents upgrading, whereas other instances this tipping point is not reached and only a portion of the agents upgrade.  A B B C C D D E 70   Table 7: The mean proportion of irrigating agents that have upgraded their landscape in a “Restrictive Growth” growth scenario. * indicates a scenario is not significantly different (N = 12, Tukey HSD p-value <0.0001 for significant differences) Scenario Mean (Standard Deviation) Control* 0.0298 (0.0049) 25% P3 Agents – Dispersed* / Clustered* 0.0157(0.0021) / 0.0026(0.0006) 50% P3 Agents – Dispersed* / Clustered* 0.0005(0.0002) / 0.0003(0.0003) 75% P3 Agents – Dispersed* / Clustered* 0.0021(0.0009) / 0.0011(0.0009) 100% P3 Agents  0.4918(0.2474)   In the scenario with 100% P3 agents there is significant variation between simulations. Across 12 runs of the model, the percentage of agents upgrading their landscape varies from a minimum of 12.6% to up to a maximum of 93%. It is likely that this variation is a result of spatial effects caused by upgrading landscape and direction of social influence. In each of the scenarios two requirements must be met for an agent to upgrade its landscape: 1) the agent must have a favourable attitude and 2) the agent must learn about the opportunity to upgrade. When an agent upgrades its landscape it improves the chance that members in its social network will learn about their ability to upgrade their own landscapes. It is possible that in simulations with 100% P3 agents, the two above considerations are being met more easily. In all of the simulations with 100% P3 agents, many agents would have a favourable attitude towards upgrading their irrigation technology. In some of the simulations with 100% P3 agents, by chance multiple agents within a social network may upgrade their property, which could lead to a chain reaction of upgrades and a tipping point to near complete uptake of irrigation upgrades.   71  These results indicate that there is a potential tipping point present that warrants further investigation. If the model were to be recalibrated, it is possible that this tipping point is significantly below the 100% water saver agent level.  Clustering of agents appears to have a minor influence on outdoor water use when there is only a 25% proportion of P3 agents, and no effect at higher proportions of P3 agents (Figure 20). Again, there is a large difference in the simulations with 100% water saver agents which is related to the elevated rates of landscape upgrading. The indoor water use on the other hand shows significant differences between clustered and non-clustered scenarios, except at the 75% P3 level (Figure 21). It is possible that these differences seen between outdoor and indoor water use with regards to agent clustering are a result of an effect occurring at the agent level (seen in indoor use) and then that effect is being lost in the variation of lot sizes.  72   Figure 20: Total outdoor water use for scenarios with controlled proportions of water saver agent types within the Restrictive Growth scenario. Rates are relative to modeled current use for the same area. In the “Clustered” scenarios, the distribution of water savers on the landscape is clustered. In the other scenarios these agents are distributed randomly. Whiskers indicate the maximum and minimum values, the box itself represents the first and third quartiles, and the middle line represents the median. A black dot represents an outlier. Groups that do not share a letter are significantly different. (N = 12, α = 0.05)   A B AB C C C C D 73   Figure 21: Total indoor water use for scenarios with controlled proportions of water conscious agent types within the Restrictive Growth scenario. Rates are relative to modeled current use for the same area. In the “Clustered” scenarios, the distribution of water savers on the landscape is clustered. In the other scenarios these agents are distributed randomly. Whiskers indicate the maximum and minimum values, the box itself represents the first and third quartiles, and the middle line represents the median. A black dot represents an outlier. All groups are significantly different except those that share an * above them.  (N = 12, α = 0.05)   Clustering of agents has a significant effect on the ENTITLEMENT and FORESIGHT traits (Figure 22 and Figure 23). In general, for both, high water consumers’ influences increase as the proportion of P3 agents is raised to 25% and then decrease as the proportion is increased to 50% and beyond. This effect at the 25% proportion is likely due to the fact more P3 agents are available to be negatively influenced while still not being numerous enough for their own water saving influence to overcome the other agent types’ water wasting influences. Surprisingly, clustering the P3 agents actually had a negative impact on their ability to positively influence the * * 74  other agents. Whereas the original assumption was that clustering the P3 agents together would provide protection from water wasting influence, what occurs instead is that clustering these agents stifles their ability to positively influence others, negating any positive influence from their “protection”.   Figure 22: Mean amounts of change in the ENTITLEMENT trait of agents for scenarios with controlled proportions of water conscious agent types within the Restrictive Growth scenario. In the “Clustered” scenarios, the distribution of water savers on the landscape is clustered. In the other scenarios these agents are distributed randomly. Whiskers indicates the maximum and minimum values, the box itself represents the first and third quartiles, and the middle line represents the median. All groups are significantly different. (N = 12, α = 0.05) 75   Figure 23: Mean amounts of change in the FORESIGHT trait of agents for scenarios with controlled proportions of water conscious agent types within the Restrictive Growth scenario. In the “Clustered” scenarios, the distribution of water savers on the landscape is clustered. In the other scenarios these agents are distributed randomly. Whiskers indicates the maximum and minimum values, the box itself represents the first and third quartiles, and the middle line represents the median. All groups are significantly different except those that share an * above them. (N = 12, α = 0.05)    It is perhaps wishful thinking that the personalities represented in the population that this model represents would change to the degrees assessed above. Increasing the numbers of individuals with certain personality types is not feasible in the real world as we cannot discriminate who decides to move to or live in Kelowna. Education can, however, play a role in * * 76  reducing the influences of water wasters (Figure 16) and as such can be explored as a way of improving social norms. Another actionable analogue for increasing the proportion of water conscious individuals could be systematically increasing the proportion of landscape that is upgraded in the form of xeriscaping, improved irrigation systems, etc. Considering the importance of social norms to decision making and the role that our landscaping and irrigation behaviours play in creating our social norms for water use, the effect of increasing the prevalence of upgraded parcels at the initialization stage of the model was tested.  Results in Figure 24 and Figure 25 show that increasing the proportion of landscape originally upgraded leads to reduced outdoor water use – as expected – but spatial clustering does not appear to have a significant influence on the number of agents upgrading their landscape. These results were somewhat expected given previous results showing the complete lack of any agents upgrading their landscape for purely economic reasons. Therefore, the final proportion of landscape upgrade is merely the initial amount of landscape converted plus the amount expected from conversions done for moral reasons. These results could likely change if the model was recalibrated to allow for agents to make upgrading decisions based on a combination of moral persuasion and economic factors combined, as opposed to only one or the other.   77   Figure 24: The final proportion of landscape area upgraded for policy scenarios with controlled initial proportions of landscape area within the Restrictive Growth scenario. In the “Clustered” scenarios the area that is initially upgraded in the landscape is clustered. In the other scenarios these upgraded parcels are distributed randomly. Whiskers indicates the maximum and minimum values, the box itself represents the first and third quartiles, and the middle line represents the median. A black dot represents an outlier. Groups that do not share a letter are significantly different. (N = 12, α = 0.05)  A A B B C C D E D 78   Figure 25: Total outdoor water use for policy scenarios with controlled initial proportions of landscape area within the Restrictive Growth scenario. Rates are relative to modeled current use for the same area. In the “Clustered” scenarios the area that is initially upgraded in the landscape is clustered. In the other scenarios these upgraded parcels are distributed randomly. Whiskers indicates the maximum and minimum values, the box itself represents the first and third quartiles, and the middle line represents the median. A black dot represents an outlier. Groups that do not share a letter are significantly different. (N = 12, α = 0.05) F E D D C C B A B 79  Chapter 5 Conclusion  This thesis presents the framework for an empirically based ABM that has been applied to the city of Kelowna, BC as a case study. The model developed herein represents an improvement on currently available methods for estimating the response of residential water users to a suite of regulatory measures, population growth and urban expansion scenarios, and social networks. It is an improvement in the sense that the water use results from this model are in moderate agreement with previously attempted modelling of residential water use in Kelowna, but it is implemented using new methodologies. These methodologies incorporate a greater level of detail regarding household water use decisions as well as landscaping configuration, allowing for a more accurate representation of the heterogeneity of residential water use at the parcel level. The model presented here also provides more flexibility in scenario testing compared to previous models.  Furthermore, this work supports efforts in four of the twelve guiding principles set out in the Okanagan Sustainable Water Strategy (Okanagan Water Stewardship Council, 2008): i) it integrates land use planning and water resource management, ii) it provides a collection and dissemination of scientific information on Okanagan water, iii) it provides a model and therefore improved resources for local water management initiatives, and iv) it can help managers in practising adaptive water and land management. This model fills a gap in integrating social data and provides a unique tool for water managers to assess the potential impact of regulatory measures on the water use patterns of the population. The model was used to address the first thesis objective and determine which policy or policies would be most effective at reducing residential water use. The results indicate that 80  increasing the price of water in the tiered rate system results in a 4% decrease in outdoor demand in control scenarios but that when paired with reduction in the tiered rate system’s volume structure, increasing prices can result in a 10% decrease in outdoor water demand (Figure 13). These results highlight the potential synergistic effect of combining regulatory measures. Similarly, it was found that reducing the tiered rate system’s structure volumes resulted in a 14% reduction in outdoor demand while watering restrictions resulted in a 29% decrease in outdoor demand and the combination of the two resulted in a 36% decrease. That ABMs can model such non-linear dynamics, using readily available GIS data, with a great level of detail, in a manner that is easily understandable, makes them uniquely powerful. These results indicate that there is great room for improvement in the sector of demand side management of water resources through land development planning, and policy implementation.  Testing growth scenarios using this ABM addressed the second objective of determining the effects of three future urban growth patterns on residential water use. The results indicate that transitioning toward a more conservative growth pattern compared to the currently observed urban sprawl could result in a 50% or greater reduction in the annual water demand increase. Although regulatory change can be a difficult decision for governments that thrive on public opinion and work on an election cycle basis, these results on the social dynamics of Kelowna highlight a need for regulatory intervention.  The social dynamics incorporated into this model allowed for the exploration of the third objective to determine the influence of social interaction on residential water use. The model indicates that the personality composition of the individuals in Kelowna is one that is not conducive to positive change (Figure 16). Although there is a small group of water saving individuals, the majority of the population is relatively unaware of potential future challenges in 81  the form of reduced supply and increased demand, feels that individuals have an inherent right to deregulated water use, or is not ecologically focused enough to induce positive change without outside influence (Figure 6). Education was found to precipitate some improvement in social norms but intervention in the form of regulation is still necessary to overcome increased water use due to pervasive negative social norms and influences. These results highlight the important role that can be played by regulation. Incentivizing or supporting desirable, minority opinions or actions could insulate the individuals with those behaviours from the negative influence of others.  The impact of regulations that attempt to induce change voluntarily, with the exception of the centralized irrigation leasing (i.e. the education and subsidies scenarios), were found to be generally less effective in comparison to scenarios that delegated change to irrigators, such as the rationing scenario. Upon first read these results may indicate that it is unnecessary for the public to become aware of these issues and instead that it is more effective to have governments impose change from the top-down. One must keep in mind that these results do not account for the impact that improved awareness might have on residents’ likelihood of supporting sustainability oriented regulations, initiatives, and political candidates. As such, awareness based, and voluntary measures must not be discounted outright.  The conclusions presented here are valuable in providing better information on the possible responses of residential water users to education, policy change, and urban growth patterns. The model was able to produce such results with a relatively simplistic representation of human decision making but one that is an improvement on the methods previously used by other researchers. There are areas within that model that could benefit from future refinement such as improving upon the way personality based and economically based decisions are 82  weighted by an agent, and by including additional sources of personality shifting influences such as awareness of droughts and climate change. Although the inclusion of these elements could change some outcomes, they are unlikely to dramatically affect model predictions. The creation of this ABM will creates new opportunities to investigate. Further work could explore drought years and wet years, including landscape water requirement model input data could yield more specific lot by lot results, running the model at a higher resolution could provide a more detailed assessment of specific areas or land-use types, the model could be expanded to the rest of the Okanagan Valley, and factors could be created to weight each land-use type differently to compensate for disproportionate representation when resolution is a limiting factor such as in a basin-wide analysis  Overall, the work presented within this thesis represent a first iteration of an agent-based model applied to residential water use and demand side management in Kelowna, BC. It represents a new type of decision support tool upon which policy makers and researchers can build. The open nature of the software that this framework was developed, and the technical description of the model provided within the model and in this thesis provide a basis from which improvements can be made.   It is the hope of the author that this work will assist Kelowna and eventually the Okanagan Valley in developing and testing adaptive management strategies to deal with the wicked problems present here, in an empirically-based, more holistic way.  83  Bibliography Ajzen, I. (1991). The theory of planned behavior. Organizational Behavior and Human Decision Processes, 50(2), 179–211. http://doi.org/10.1016/0749-5978(91)90020-T Altaweel, M., Altaweel, M. R., Alessa, L. N., & Kliskey, A. D. (2009). Forecasting Resilience in Arctic Societies: Creating Tools for Assessing Social-Hydrological Systems. Journal of the American Water Resources Association, 45(6), 1379. Amburgey, J. W., & Thoman, D. B. (2012). Dimensionality of the New Ecological Paradigm: Issues of Factor Structure and Measurement. Environment and Behavior. http://doi.org/10.1177/0013916511402064 An, L. (2012). Modeling human decisions in coupled human and natural systems: Review of agent-based models. Ecological Modelling, 229, 25–36. http://doi.org/10.1016/j.ecolmodel.2011.07.010 An, L., & López-Carr, D. (2012). Understanding human decisions in coupled natural and human systems. Ecological Modelling, 229, 1–4. http://doi.org/10.1016/j.ecolmodel.2011.10.023 Ariely, M., Ariely, D., Kahneman, D., & Loewenstein, G. (2000). Joint Comment on " When Does Duration Matter in Judgment and Decision, 129(4), 524–529. http://doi.org/10.1037/AW96-3445.129.4.5 Arló-Costa, H., & Pedersen, A. P. (2012). Fast and frugal heuristics: rationality and the limits of naturalism. Synthese, 190(5), 831–850. http://doi.org/10.1007/s11229-012-0188-6 Athanasiadis, I. N., Mentes, A. K., Mitkas, P. A., & Mylopoulos, Y. A. (2005). A Hybrid 84  Agent-Based Model for Estimating Residential Water Demand. Simulation (San Diego, Calif.), 81(3), 175–187. http://doi.org/10.1177/0037549705053172 Austin, M. A., Buffet, D. A., & Nicolson, D. J. (2008). Taking Nature’s Pulse: The Status of Biodiversity in British Columbia. (G. G. . Scudder & V. Stevens, Eds.). Victoria, BC. Retrieved from http://www.biodiversitybc.org/assets/pressReleases/BBC_StatusReport_Web_final.pdf Bakker, K., & Cook, C. (2011). Water Governance in Canada: Innovation and Fragmentation. International Journal of Water Resources Development. http://doi.org/10.1080/07900627.2011.564969 Bakker, M. M., & van Doorn, A. M. (2009). Farmer-specific relationships between land use change and landscape factors: Introducing agents in empirical land use modelling. Land Use Policy, 26, 809–817. http://doi.org/10.1016/j.landusepol.2008.10.010 Balijepally, V., Mangalaraj, G., & Iyengar, K. (2011). Are We Wielding this Hammer Correctly? A Reflective Review of the Application of Cluster Analysis in Information Systems Research. Journal of the Association for Information Systems, 12(5), 375–413. Retrieved from http://aisel.aisnet.org/jais/vol12/iss5/1 Batty, M. (2009). Urban Modelling. International Encyclopedia of Human Geography, 19(6), 1–18. Retrieved from http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.145.5025&amp;rep=rep1&amp;type=pdf BC Ministry of Environment. Bill 18 - 2014 Water Sustainability Act (2016). Retrieved from https://www.leg.bc.ca/Pages/BCLASS-85  Legacy.aspx#%2Fcontent%2Flegacy%2Fweb%2F40th2nd%2F3rd_read%2Fgov18-3.htm Bennett, D. A., & Tang, W. (2005). GAIA-RM : A Geographically Aware Intelligent Agents Framework for Rangeland Management, (Hardin 1968), 1–5. Berglund, E. Z. (2015). Using Agent-Based Modeling for Water Resources Planning and Management. Journal of Water Resources Planning and Management, 4015025. http://doi.org/10.1061/(ASCE)WR.1943-5452.0000544 Berkhin, P. (2006). A survey of clustering data mining techniques. In Grouping multidimensional data: Recent advances in clustering (pp. 25–71). http://doi.org/10.1007/3-540-28349-8_2 Brenner, T. (2006). Chapter 18 Agent Learning Representation: Advice on Modelling Economic Learning. Handbook of Computational Economics. http://doi.org/10.1016/S1574-0021(05)02018-6 British Columbia Ministry of Agriculture, Food, and Fisheries. (2004). Industry Overview: Apples and Pears in British Columbia. Retrieved from http://www.al.gov.bc.ca/treefrt/profile/pome.pdf Bruch, E., & Atwell, J. (2013). Agent-Based Models in Empirical Social Research. Sociological Methods & Research, 00(0), 1–36. http://doi.org/10.1177/0049124113506405 Camerer, C. F., & Loewenstein, G. (2004). Behavioral Economics: Past, Present, and Future. In C. F. Camerer, G. Loewenstein, & M. Rabin (Eds.), Advances in Behavioural 86  Economics. Princeton: Princeton University Press. Chion, C., Lamontagne, P., Turgeon, S., Parrott, L., Landry, J. -a., Marceau, D. J., … Dionne, S. (2011). Eliciting cognitive processes underlying patterns of human–wildlife interactions for agent-based modelling. Ecological Modelling, 222(14), 2213–2226. http://doi.org/10.1016/j.ecolmodel.2011.02.014 Chrislip, D. D., Larson, C. E., & American Leadership Forum. (1994). Collaborative leadership: How citizens and civic leaders can make a difference. An American Leadership Forum book. Retrieved from http://www.loc.gov/catdir/bios/wiley044/94010751.html\nhttp://www.loc.gov/catdir/description/wiley037/94010751.html\nhttp://www.loc.gov/catdir/toc/onix07/94010751.html Chu, J., Wang, C., Chen, J., & Wang, H. (2009). Agent-Based Residential Water Use Behavior Simulation and Policy Implications: A Case-Study in Beijing City. Water Resources Management, 23(15), 3267–3295. http://doi.org/10.1007/s11269-009-9433-2 Chu, T., Drogoul, A., & Boucher, A. (2009). Interactive Learning of Independent Experts ’ Criteria for Rescue Simulations, 15(13), 2701–2725. Churchman, C. W. (1967). Free For All. Management Science, 14(4), B141–B142. City of Kelowna. (2013). Chapter 3: Growth Projections. In 2030 Official Community Plan. City of Kelowna. Retrieved from http://apps.kelowna.ca/CityPage/Docs/PDFs/Bylaws/Official Community Plan 2030 Bylaw No. 10500/Chapter 03 - Growth Projections.pdf City of Kelowna. (2015). City of Kelowna Zoning Bylaw No. 8000 - Section 13 Urban 87  Residential Zones. Retrieved from http://www.kelowna.ca/CM/page2561.aspx Cohen, S., & Kulkarni, T. (2001). Water Management & Climate Change in the Okanagan Basin. Retrieved from http://cses.washington.edu/cig/outreach/workshopfiles/Apr03_Scenarios_water/Water_Management_and_Climate_Change_in_the_Okanagan_Basin.pdf Cohen, S., & Neale, T. (2006a). Participatory Integrated Assessment of Water Management and Climate Change in the Okanagan Basin , British Columbia. Integrated Assessment, 14. Retrieved from http://adaptation.nrcan.gc.ca/projdb/pdf/a846_summary_e.pdf Cohen, S., & Neale, T. (2006b). Participatory integrated assessment of water management and climate change in the Okanagan Basin, British Columbia. Final report. Integrated Assessment, 14. Retrieved from http://adaptation.nrcan.gc.ca/projdb/pdf/a846_summary_e.pdf Cohen, S., Neilsen, D., & Smith, S. (2004). Expanding the Dialogue on Climate Change and Water Management in the Okanagan Basin, British Columbia. Retrieved from http://www.wsabc.ca/wp-content/uploads/2011/04/Climate-Change-2004.pdf Ding, C., & He, X. H. X. (2002). Cluster merging and splitting in hierarchical clustering algorithms. 2002 IEEE International Conference on Data Mining, 2002. Proceedings. http://doi.org/10.1109/ICDM.2002.1183896 Dunlap, R. E. (2008). The New Environmental Paradigm Scale: From Marginality to Worldwide Use. The Journal of Environmental Education, 40(1), 3–18. http://doi.org/10.3200/JOEE.40.1.3-18 88  Dunlap, R. E., & Van Liere, K. D. (1978). The “New Environmental Paradigm”: a proposed measuring instrument and preliminary results. The Journal of Environmental Education, 9(4), 10–19. http://doi.org/10.1080/00958964.1978.10801875 Edmonds, B., & Chattoe, E. (2005). When Simple Measures Fail: Characterising Social Networks Using Simulation. In Social network analysis: Advances and empirical applications forum. Manchester, U.K.: Manchester Metropolitan University Business School. Environment Canada. (2015). Canadian Climate Normals. Retrieved December 1, 2015, from http://climate.weather.gc.ca/climate_normals/results_1981_2010_e.html?stnID=1700&lang=e&amp\nStationName=tuktoyaktuk&amp\nSearchType=Contains&amp\nstnNameSubmit=go&dCode=5&dispBack=1\nfiles/83/Canada - 2013 - Canadian Climate Normals - Environment Canada. Evans, T. P., Green, G. M., & Carlson, L. A. (2001). Multi-Scale Analysis of Landcover Composition and Landscape Management of Public and Private Lands in Indiana BT  - GIS and Remote Sensing Applications in Biogeography and Ecology. In A. C. Millington, S. J. Walsh, & P. E. Osborne (Eds.), (pp. 271–287). Boston, MA: Springer US. http://doi.org/10.1007/978-1-4615-1523-4_16 Fernandez, L. E., Brown, D. G., Marans, R. W., & Nassauer, J. I. (2005). Characterizing location preferences in an exurban population: implications for agent-based modeling. Environment and Planning B: Planning and Design, 32(6), 799–820. http://doi.org/10.1068/b3071 Filatova, T., Verburg, P. H., Parker, D. C., & Stannard, C. A. (2013). Spatial agent-based 89  models for socio-ecological systems: Challenges and prospects. Environmental Modelling and Software, 45, 1–7. http://doi.org/10.1016/j.envsoft.2013.03.017 Fraley, C., & Raftery, A. (2002). Model-Based Clustering, Discriminant Analysis, and Density Estimation. Journal of the American Statistical Association, 97(458), 611–631. http://doi.org/10.2307/3085676 Fraley, C., & Raftery, A. E. (1998). How Many Clusters? Which Clustering Method? Answers Via Model-Based Cluster Analysis. The Computer Journal, 41(8), 578–588. http://doi.org/10.1093/comjnl/41.8.578 Fraley, C., Raftery, A. E., Murphy, T. B., & Scrucca, L. (2012). mclust Version 4 for R: Normal Mixture Modeling for Model-Based Clustering, Classification, and Density Estimation. Galan, J. M., del Olmo, R., & Lopez-Paredes, A. (2008). Diffusion of Domestic Water Conservation Technologies in an ABM-GIS Integrated Model. In Hybrid artificial intelligence systems : third international workshop (pp. 567–574). SPRINGER-VERLAG BERLIN. Galán, J. M., López-Paredes, A., & del Olmo, R. (2009). An agent-based model for domestic water management in Valladolid metropolitan area. Water Resources Research, 45(5), n/a–n/a. http://doi.org/10.1029/2007WR006536 Gelenbe, E., Ieee, F., & Esin, S. (2001). Simulation with Learning Agents, 1–16. Giacomoni, M. H., & Berglund, E. Z. (2015). Complex Adaptive Modeling Framework for Evaluating Adaptive Demand Management for Urban Water Resources Sustainability. 90  Journal of Water Resources Planning and Management, 141(11). http://doi.org/10.1061/(ASCE)WR.1943-5452.0000543 Giacomoni, M., Kanta, L., & Zechman, E. (2013). Complex Adaptive Systems Approach to Simulate the Sustainability of Water Resources and Urbanization. Journal of Water Resources Planning and Management, 139(5), 554–564. http://doi.org/10.1061/(ASCE)WR.1943-5452.0000302 Gintis, H. (2000). THE HUMAN ACTOR IN ECOLOGICAL-ECONOMIC MODELS Beyond Homo economicus : evidence from experimental economics, 35, 311–322. Gorton, M., Douarin, E., Davidova, S., & Latruffe, L. (2008). Attitudes to agricultural policy and farming futures in the context of the 2003 CAP reform: A comparison of farmers in selected established and new Member States. Journal of Rural Studies, 24(3), 322–336. http://doi.org/10.1016/j.jrurstud.2007.10.001 Guillem, E. E., Barnes,  a P., Rounsevell, M. D. a, & Renwick,  a. (2012). Refining perception-based farmer typologies with the analysis of past census data. Journal of Environmental Management, 110, 226–35. http://doi.org/10.1016/j.jenvman.2012.06.020 Haggith, M., Iv, S., Muetzelfeldt, R. I., & Taylor, J. (2003). Modelling Decision-making in Rural Communities at the Forest Margin, 2(2), 241–258. Hamlet, A. F., Hamlet, A. F., & Lettenmaier, D. P. (2007). Effects of 20th century warming and climate variability on flood risk in the western U.S. Water Resources Research, 43(6). http://doi.org/10.1029/2006WR005099 91  Han, J., Kamber, M., & Pei, J. (2012a). Advanced Cluster Analysis. In Data Mining: Concepts and Techniques (Vol. 54, p. 703). Han, J., Kamber, M., & Pei, J. (2012b). Cluster Analysis—Basic Concepts and Methods. In Data Mining: Concepts and Techniques. Retrieved from http://library.books24x7.com.ezproxy.library.ubc.ca/assetviewer.aspx?bookid=44712&chunkid=140557590&rowid=1224 Harma, K. J., Johnson, M. S., & Cohen, S. J. (2012). Future Water Supply and Demand in the Okanagan Basin, British Columbia: A Scenario-Based Analysis of Multiple, Interacting Stressors. Water Resources Management, 26(3), 667–689. http://doi.org/10.1007/s11269-011-9938-3 Haughton, D., Legrand, P., & Woolford, S. (2009). Review of Three Latent Class Cluster Analysis Packages: Latent Gold, poLCA, and MCLUST. The American Statistician. http://doi.org/10.1198/tast.2009.0016 Head, B. W., & Alford, J. (2015). Wicked Problems: Implications for Public Policy and Management. Administration & Society, 47(6), 711–739. http://doi.org/10.1177/0095399713481601 Hodgson, G. M., & Knudsen, T. (2004). The complex evolution of a simple traffic convention: the functions and implications of habit. Journal of Economic Behavior & Organization, 54(1), 19–47. http://doi.org/10.1016/j.jebo.2003.04.001 House-Peters, L. a., & Chang, H. (2011). Urban water demand modeling: Review of concepts, methods, and organizing principles. Water Resources Research, 47(5), W05401. http://doi.org/10.1029/2010WR009624 92  Hutchinson, J. M. C., & Gigerenzer, G. (2005). Simple heuristics and rules of thumb: where psychologists and behavioural biologists might meet. Behavioural Processes, 69(2), 97–124. http://doi.org/10.1016/j.beproc.2005.02.019 InterVISTAS Counsulting Inc. (2011). The Economic Impact of Tourism in Kelowna. Retrieved from http://www.tourismkelowna.com/includes/content/docs/media/EconomicImpactReport.pdf IPCC. (2014). Climate Change 2014: Synthesis Report. Contribution of Working Groups I, II and III to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change. (Core Writing Team, R. K. Pachauri, & L. A. Meyer, Eds.). Geneva, Switzerland: IPCC. Janmaat, J. (2015). Household Water Conservation Choices in Kelowna: Hunting for a Price Response. Janssen, M. a, & Ostrom, E. (2006). Empirically Based , Agent-based models. Ecology And Society, 11(2), 37. http://doi.org/yes Jatel, N. (2008). Using social network analysis to make invisible human actor water governance networks visible - the case of the Okanagan valley. The University of British Columbia. Retrieved from https://circle.ubc.ca/bitstream/handle/2429/45523/ubc_2014_spring_Jatel_Nelson.pdf?sequence=1 Jones, B. D. (1999). Bounded Rationality. Annual Review of Political Science, 2(1), 297–321. http://doi.org/10.1146/annurev.polisci.2.1.297 93  Kahneman, D., & Tversky, A. (1979). Prospect Theory: An Analysis of Decision under Risk. Econometrica, 47(2), 263–291. Kanta, L., & Zechman, E. (2014). Complex Adaptive Systems Framework to Assess Supply-Side and Demand-Side Management for Urban Water Resources. Journal of Water Resources Planning and Management, 140(1), 75–85. http://doi.org/10.1061/(ASCE)WR.1943-5452.0000301 Karali, E., Rounsevell, M. D. a., & Doherty, R. (2011). Integrating the diversity of farmers’ decisions into studies of rural land-use change. Procedia Environmental Sciences, 6, 136–145. http://doi.org/10.1016/j.proenv.2011.05.014 Kaufmann, P., Stagl, S., & Franks, D. W. (2009). Simulating the diffusion of organic farming practices in two New EU Member States. Ecological Economics, 68(10), 2580–2593. http://doi.org/10.1016/j.ecolecon.2009.04.001 Lasker, R. D., & Weiss, E. S. (2003). Broadening participation in community problem solving: a multidisciplinary model to support collaborative practice and research. J Urban Health, 80(1), 14–60. http://doi.org/10.1093/jurban/jtg014 Le, Q. B., Park, S. J., & Vlek, P. L. G. (2010). Land Use Dynamic Simulator (LUDAS): A multi-agent system model for simulating spatio-temporal dynamics of coupled human–landscape system. Ecological Informatics, 5(3), 203–221. http://doi.org/10.1016/j.ecoinf.2010.02.001 Le, Q. B., Park, S. J., Vlek, P. L. G., & Cremers, A. B. (2008). Land-Use Dynamic Simulator (LUDAS): A multi-agent system model for simulating spatio-temporal dynamics of coupled human–landscape system. I. Structure and theoretical specification. Ecological 94  Informatics, 3(2), 135–153. http://doi.org/10.1016/j.ecoinf.2008.04.003 Lea, T. (2008). Historical (pre-settlement) ecosystems of the Okanagan Valley and Lower Similkameen Valley of British Columbia - pre-European contact to the present. Davidsonia, 19(1), 3–36. Retrieved from http://scholar.google.com/scholar?hl=en&btnG=Search&q=intitle:Historical+(pre-settlement)+ecosystems+of+the+Okanagan+Valley+and+lower+Similkameen+Valley+of+British+Columbia+?+pre-European+contact+to+the+present#0 Little, L. R., & McDonald,  a. D. (2007). Simulations of agents in social networks harvesting a resource. Ecological Modelling, 204(3-4), 379–386. http://doi.org/10.1016/j.ecolmodel.2007.01.013 Liu, J., Dietz, T., Carpenter, S. R., Alberti, M., Folke, C., Moran, E., … Taylor, W. W. (2007). Complexity of coupled human and natural systems. Science (New York, N.Y.), 317, 1513–1516. http://doi.org/10.1126/science.1144004 Lopez-Paredes, A., Sauri, D., & Galan, J. M. (2005). Urban Water Management with Artificial Societies of Agents: The FIRMABAR Simulator. Simulation (San Diego, Calif.), 81(3), 189–199. http://doi.org/10.1177/0037549705053167 Lux, T., & Marchesi, M. (1999). Scaling and criticality in a stochastic multi-agent model of a financial market. Nature, 397(February), 498–500. http://doi.org/10.1038/17290 Manson, S. M., & Evans, T. (2007). Agent-based modeling of deforestation in southern Yucatan, Mexico, and reforestation in the Midwest United States. Proceedings of the National Academy of Sciences of the United States of America, 104(52), 20678–83. http://doi.org/10.1073/pnas.0705802104 95  Martin-Ortega, J., Perni, A., Jackson-Blake, L., Balana, B. B., & Mckee, A. (2015). A transdisciplinary approach to the economic analysis of the European Water Framework Directive. Ecological Economics, 116, 34–45. http://doi.org/10.1016/j.ecolecon.2015.03.026 Mattison, J. (2016). Environmental Flow Needs in British Columbia. McNeill, R. (2006). Chapter 11: Costs of adaptation options. In S. Cohen & T. Neale (Eds.), Participatory integrated assessment of water management and climate change in the Okanagan Basin, British Columbia. Final report (p. 188). Vancouver BC: Environment Canada and University of British Columbia. Merritt, W. S., Alila, Y., & Barton, M. (2003). Exploring Impacts of Climate Change on the Hydrology of the Okanagan Basin. In Water Stewardship: How are we managing? Proceedings of the Canadian Water Resources Association Conference, June 11-13, 2003, Vancouver, BC (pp. 1–11). Retrieved from http://www.cses.uw.edu/cig/outreach/workshopfiles/Apr03_Scenarios_water/BC_Merritt_etal_cwra2003.pdf Moss, S., & Edmonds, B. (2005). Sociology and Simulation: Statistical and Qualitative Cross‐Validation1. The American Journal of Sociology, 110(4), 1095–1131. http://doi.org/10.1086/427320 Müller, B., Bohn, F., Dreßler, G., Groeneveld, J., Klassert, C., Martin, R., … Schwarz, N. (2013). Describing human decisions in agent-based models – ODD + D, an extension of the ODD protocol. Environmental Modelling & Software, 48, 37–48. http://doi.org/10.1016/j.envsoft.2013.06.003 96  Niazi, M., & Hussain, A. (2011). Agent-based computing from multi-agent systems to agent-based models: a visual survey. Scientometrics, 89(2), 479–499. http://doi.org/10.1007/s11192-011-0468-9 NOAA. (2016). State of the Climate: Global Analysis for December 2014. Published online. Retrieved from http://www.ncdc.noaa.gov/sotc/global/201512 Okanagan Basin Water Board. (2015a). About the Council. Retrieved October 1, 2015, from http://www.obwb.ca/about-the-council/ Okanagan Basin Water Board. (2015b). History of the Okanagan Basin Water Board. Retrieved October 1, 2015, from http://www.obwb.ca/overview/history/ Okanagan Water Stewardship Council. (2008). Okanagan Sustainable Water Strategy - Action Plan 1.0. Retrieved from http://www.obwb.ca/fileadmin/docs/osws_action_plan.pdf Parrott, L. (2011). Hybrid modelling of complex ecological systems for decision support: Recent successes and future perspectives. Ecological Informatics, 6(1), 44–49. http://doi.org/10.1016/j.ecoinf.2010.07.001 Parrott, L., Chion, C., Martins, C. C. a., Lamontagne, P., Turgeon, S., Landry, J. a., … Dionne, S. (2011). A decision support system to assist the sustainable management of navigation activities in the St. Lawrence River Estuary, Canada. Environmental Modelling & Software, 26(12), 1403–1418. http://doi.org/10.1016/j.envsoft.2011.08.009 Parrott, L., & Latombe, G. (2012). Agents , Individuals , and Networks : Modeling Methods to Inform Natural Resource Management in Regional Landscapes. Ecology and Society, 97  17(3). Parrott, L., & Meyer, W. S. (2012). Future landscapes: managing within complexity. Frontiers in Ecology and the Environment, 10(7), 382–389. http://doi.org/10.1890/110082 Polar Geoscience Ltd. (2012). Phase 3 Okanagan Basin Water Supply and Demand Project: Projected Water Supply and Use in the Okanagan Basin (2011 - 2040) - Okanagan Basin Water Accounting Model Results. Polasky, S., Nelson, E., Camm, J., Csuti, B., Fackler, P., Lonsdorf, E., … Tobalske, C. (2008). Where to put things? Spatial land management to sustain biodiversity and economic returns. Biological Conservation, 141(6), 1505–1524. http://doi.org/10.1016/j.biocon.2008.03.022 Railsback, S. F., & Grimm, V. (2012). Agent-based and individual-based modeling: a practical introduction. Princeton: Princeton University Press. Rammel, C., Stagl, S., & Wilfing, H. (2007). Managing complex adaptive systems - A co-evolutionary perspective on natural resource management. Ecological Economics, 63, 9–21. http://doi.org/10.1016/j.ecolecon.2006.12.014 Richetin, J., Sengupta, A., Perugini, M., Adjali, I., Hurling, R., Greetham, D., & Spence, M. (2010). A micro-level simulation for the prediction of intention and behavior. Cognitive Systems Research, 11(2), 181–193. http://doi.org/10.1016/j.cogsys.2009.08.001 Rittel, H. W. J., & Webber, M. M. (1973). Dilemmas in a General Theory of Planning. Policy Sciences, 4(2), 155–169. Retrieved from 98  http://www.jstor.org.ezproxy.library.ubc.ca/stable/4531523 Rockström, J., Steffen, W., Noone, K., Persson, Å., Chapin, F. S., Lambin, E., … Foley, J. (2009). Planetary boundaries: Exploring the safe operating space for humanity. Ecology and Society, 14(2). http://doi.org/10.1038/461472a Rounsevell, M. D. a, Robinson, D. T., & Murray-Rust, D. (2012). From actors to agents in socio-ecological systems models. Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences, 367(1586), 259–69. http://doi.org/10.1098/rstb.2011.0187 Russell, S., & Fielding, K. (2010). Water demand management research: A psychological perspective. Water Resources Research. http://doi.org/10.1029/2009WR008408 Schlüter, M., & Pahl-wostl, C. (2007). Mechanisms of Resilience in Common-pool Resource Management Systems : an Agent-based Model of Water Use in a River Basin, 12(2). Schmitzberger, I., Wrbka, T., Steurer, B., Aschenbrenner, G., Peterseil, J., & Zechmeister, H. G. (2005). How farming styles influence biodiversity maintenance in Austrian agricultural landscapes. Agriculture, Ecosystems and Environment, 108(3), 274–290. http://doi.org/10.1016/j.agee.2005.02.009 Schwarz, N., & Ernst, A. (2009a). Agent-based modeling of the diffusion of environmental innovations — An empirical approach. Technological Forecasting & Social Change, 76(4), 497–511. http://doi.org/10.1016/j.techfore.2008.03.024 Schwarz, N., & Ernst, A. (2009b). Agent-based modeling of the diffusion of environmental innovations — An empirical approach. Technological Forecasting and Social Change, 99  76(4), 497–511. http://doi.org/10.1016/j.techfore.2008.03.024 Senate of Canada. (2005). Water in the West: Under Pressure. Ottawa, Ontario. Simon, H. (1990). Invariants of Human Behaviour. Annual Review of Psychology, 41, 1–19. Simon, H. a. (2000). Bounded rationality in social science: Today and tomorrow. Mind & Society, 1, 25–39. http://doi.org/10.1007/BF02512227 Smajgl, A., Brown, D. G., Valbuena, D., & Huigen, M. G. a. (2011). Empirical characterisation of agent behaviours in socio-ecological systems. Environmental Modelling & Software, 26(7), 837–844. http://doi.org/10.1016/j.envsoft.2011.02.011 Spence, C., & Hedstrom, N. (2015). Attributes of Lake Okanagan evaporation and development of a mass transfer model for water management purposes. Canadian Water Resources Journal, 40(3), 250–261. Retrieved from http://www-tandfonline-com.ezproxy.library.ubc.ca/doi/pdf/10.1080/07011784.2015.1046140 Statistics Canada. (2012a). Focus on Geography Series, 2011 Census. Statistics Canada Catalogue no. 98-310-XWE2011004. Ottawa, Ontario. Retrieved from https://www12.statcan.gc.ca/census-recensement/2011/as-sa/fogs-spg/Facts-csd-eng.cfm?LANG=Eng&GK=CSD&GC=5935010 Statistics Canada. (2012b). The Canadian Population in 2011: Population Counts and Growth - Population and dwelling counts, 2011 Census. Retrieved from http://www12.statcan.gc.ca/census-recensement/2011/as-sa/98-310-x/98-310-x2011001-eng.pdf Statistics Canada. (2013a). Catalogue no. 98-313-XCB2011022. In 2011 Census of 100  Population. Retrieved from http://www.statcan.gc.ca/tables-tableaux/sum-som/l01/cst01/famil122g-eng.htm Statistics Canada. (2013b). Human Activity and the Environment Freshwater supply and demand in Canada 2010 — Updated. Retrieved from http://www.statcan.gc.ca/pub/16-201-x/16-201-x2010000-eng.pdf Steffen, W., Richardson, K., Rockström, J., Cornell, S. E., Fetzer, I., Bennett, E. M., … Sörlin, S. (2015). Planetary boundaries: Guiding human development on a changing planet. Science, 347(6223), 1259855–. http://doi.org/10.1126/science.1259855 Stern, P. C., Dietz, T., & Guagnano, G. A. (1995). The New Ecological Paradigm in social-psychological context. Environment and Behavior, 27(6), 723–743. Stewart, I. T., Cayan, D. R., & Dettinger. (2004). Changes in Snowmelt Runoff Timing in Western North America under a `Business as Usual’ Climate Change Scenario. Climatic Change, 62(1-3), 217–232. http://doi.org/10.1023/B:CLIM.0000013702.22656.e8 Stewart, I. T., Cayan, D. R., & Dettinger, M. D. (2005). Changes toward earlier streamflow timing across western North America. Journal of Climate, 18(8), 1136–1155. http://doi.org/10.1175/JCLI3321.1 Summit Environmental Consulting Inc. (2010). Okanagan Water Supply and Demand Project : Phase 2 Summary Report. Valbuena, D., Verburg, P. H., & Bregt, A. K. (2008). A method to define a typology for agent-based analysis in regional land-use research. Agriculture, Ecosystems & Environment, 128(1-2), 27–36. http://doi.org/10.1016/j.agee.2008.04.015 101  Valbuena, D., Verburg, P. H., Veldkamp,  a., Bregt, A. K., & Ligtenberg, A. (2010). Effects of farmers’ decisions on the landscape structure of a Dutch rural region: An agent-based approach. Landscape and Urban Planning, 97(2), 98–110. http://doi.org/10.1016/j.landurbplan.2010.05.001 Vano, J. A., Scott, M. J., Voisin, N., Stöckle, C. O., Hamlet, A. F., Mickelson, K. E. B., … Lettenmaier, D. P. (2010a). Climate change impacts on water management and irrigated agriculture in the Yakima River Basin, Washington, USA. Climatic Change, 102(1-2), 287–317. http://doi.org/10.1007/s10584-010-9856-z Vano, J. A., Voisin, N., Cuo, L., Hamlet, A. F., Elsner, M. M., Palmer, R. N., … Lettenmaier, D. P. (2010b). Climate change impacts on water management in the Puget Sound region, Washington State, USA. Climatic Change, 102(1-2), 261–286. http://doi.org/10.1007/s10584-010-9846-1 Wagner, J. R. (2008). Landscape Aesthetics, Water, and Settler Colonialism in the Okanagan Valley of British Columbia. Journal of Ecological Anthropology, 12(1), 22 – 38. http://doi.org/http://dx.doi.org/10.5038/2162-4593.12.1.2 Wagner, J. R., & White, K. (2009). Water and development in the Okanagan Valley of British Columbia. Journal of Enterprising Communities: People and Places in the Global Economy. http://doi.org/10.1108/17506200910999129 Weber, E. P., & Khademian, A. M. (2008). Wicked Problems, Knowledge Challenges, and Collaborative Capacity Builders in Network Settings. Public Administration Review, 68(2), 334–349. http://doi.org/10.1111/j.1540-6210.2007.00866.x Wilson, K. (1989). Irrigating the Okanagan: 1860 - 1920. The University of British 102  Columbia. Retrieved from file:///C:/Users/Jon/Downloads/UBC_1989_A8 W55.pdf Woolridge, M. (2002). Introduction to Multiagent Systems (2nd ed.). New York, NY, USA: John Wiley &amp; Sons, Inc. 103  Appendices Appendix A Simulation Resolution One of the great strengths of agent-based modeling (ABM) is the ability to integrate a large range of data types. This strength can lead to issues with scale, particularly with regards to resolution of different spatial datasets. A modelled environment exists as a finite series of pixels or discrete patches whereas GIS data is often found in continuous polygons. As the resolution of a simulation decreases each pixel or patch represents a larger area and so can encompass more GIS polygons. This can lead to small polygons being underrepresented as a patch takes on the traits of the largest polygon that it encompasses. As an example compare the prevalence of black patches between images in Figure 26. The black patches represent single/two unit houses which often occupy the smallest land parcels. 104   Figure 26: A comparison of a 200x200 resolution landscape (left) and a 50x50 resolution landscape (right). Different colours represent different land use types.  It is apparent that in Figure 26 a coarser resolution can lead to an underrepresentation of certain GIS polygons (those represented by purple for example) and therefore in relative terms over represent others (orange patches). This can create a statistical bias that is well known in the spatial analysis community as the modifiable areal unit problem. A table 105  describing the relationship between the resolution of the simulation and the area of habitable parcels being represented within the simulation – both in absolute and proportional terms – can be found in Table 8.  Table 8: The relationship between the resolution of a simulated city of Kelowna and the area of various residential land use types represented. The “GIS Polygon” column represents the true value and the percentages in parentheses represent how close a value is to this true value. “S2RESH” = single/two unit residential - hillside, “S2RES” = single/two unit residential, “MRL” = multi-unit residential (low density), “MRM” = multi-unit residential (medium density), “MRH” = multi-unit residential (high density), “MXR” = mixed use commercial/residential, “MXT” = mixed use tourism/residential. Resolution 200x200 400x400 800x800 GIS Polygon Proportional Area   S2RESH 0.0030(73.8%) 0.0027(65.5%) 0.0036(88.7%) 0.0041 S2RES 0.3161(45.5%) 0.5194(74.8%) 0.6564(94.5%) 0.6943 MRL 0.2944(204.8%) 0.2131(148.2%) 0.1570(109.2%) 0.1437 MRM 0.0542(137.2%) 0.0477(120.1%) 0.0503(127.4%) 0.0395 MRH 0.0168(299.5%) 0.0112(199.8%) 0.0068(121.3%) 0.0056 MRC 0 0.0001(26.8%) 0.0004(107.4%) 0.0004 MXR 0.1330(211.5%) 0.1039(165.3%) 0.0684(108.7%) 0.0628 MXT 0.1826(368.1%) 0.1020(205.5%) 0.0576(116.1%) 0.0496 Area   S2RESH 23237(19.8%) 37427(31.8%) 89904(76.4%) 117620 S2RES 2433620(12.2%) 7259337(36.3%) 16257332(81.4%) 19972260 MRL 2266881(54.8%) 2977825(72.1%) 3887475(94.1%) 4132922 MRM 417133(36.7%) 666695(58.7%) 1245845(109.7%) 1135219 MRH 129015(80.2%) 156236(97.1%) 168136(104.5%) 160847 MRC 0 1533(13.0%) 10887(92.5%) 11764 MXR 1023817(56.6%) 1452527(80.4%) 1692940(93.7%) 1807680 MXT 1406346(98.6%) 1425219(99.4%) 1426657(100%) 1426657  Total 7700049(26.8%) 13975266(48.6%) 24768289(86.1%) 28753205  It can be seen in the above table that increasing the resolution brings the value for the total area represented and the proportional area represented closer to the true value from the GIS polygons. It should be noted that in Table 8, for the 800x800 simulation the area 106  represented for “MRM” and “MRH” parcels is greater than 100%. This is due to a process used to compensate for the loss of small parcels at lower resolutions. Land parcels that were below the minimum threshold for lot size according to the Kelowna zoning laws were increased in area to this minimum (City of Kelowna, 2015). For example, “MRH” lots with an area below 1700 m2 were increased to 1700 m2. An artifact of this is that parcels that were established before these zoning laws came into effect would be increased in area and therefore could lead to an area greater than that seen in the GIS data.  The increased fidelity of higher resolutions comes at a cost. Not only does the landscape represent an increased area but the increase in available habitable parcels leads to a higher number of agents on the landscape and by extension a higher number of social links and interactions. These increases slow the simulations down. Therefore, a balance must be struck between the resolution desired and the time available for producing results. Due to computational and temporal limitations in this study, results were produced in a virtual landscape of 400 x 400 patches, corresponding to a patch length and width of 69 m.  The limited resolution at which the model was run should be considered in the interpretation of the results. 107  Appendix B Supplementary Agent Information Table 9: New Ecological Paradigm (NEP) Scale questions. All questions are answered on a seven-point scale ranging from “Strongly Agree” to “Strongly Disagree”. Even numbered questions’ scores are reversed to calculate the overall NEP score from an average of the 15 questions. Question 1. We are approaching the limit of the number of people the earth can support 2. Humans have the right to modify the natural environment to suit their needs 3. When humans interfere with nature it often produces disastrous consequences 4. Human ingenuity will insure that we do NOT make the earth unlivable 5. Humans are severely abusing the environment 6. The earth has plenty of natural resources if we just learn how to develop them 7. Plants and animals have as much right as humans to exist 8. The balance of nature is strong enough to cope with the impacts of modern industrial nations 9. Despite our special abilities humans are still subject to the laws of nature 10. The so-called “ecological crisis” facing humankind has been greatly exaggerated 11. The earth is like a spaceship with very limited room and resources 12. Humans were meant to rule over the rest of nature 13. The balance of nature is very delicate and easily upset 14. Humans will eventually learn enough about how nature works to be able to control it 15. If things continue on their present course, we will soon experience a major ecological catastrophe  108  Table 10: Personality traits by cluster group. Values are on a scale of 1 – 7. High NEP indicates eco-centric orientation, high FORESIGHT indicates high concern for future supply shortages, and high ENTITLEMENT indicates the belief that water should not have a cost  Cluster (N) NEP Mean/(standard deviation) FORESIGHT Mean/(standard deviation) ENTITLEMENT Mean/(standard deviation) P1 (83) “water wasters” 4.52 (1.19) 4.20(1.62) 3.58(1.10) P2 (359) “average heavy users” 4.95(0.60) 4.94(0.97) 2.28(0.66) P3 (44) “water savers” 6.25(0.28) 6.39(0.50) 1.40(0.41)    Table 11: List of agent attributes and descriptions Agent attribute Description houseID  ID number that links houses to the parcel they are on clustercentre?  variable used to designate agents as centres of clusters in simulations where agent personalities are being distributed in a clustered pattern totalwateruse  running total of both outdoor and indoor water use through a year totaloutdooruse  running total of outdoor water use in a year waterneed  the daily outdoor water required to irrigate avgusertotalwateruse  running total of the water that would have been used by a house that had average personality traits  avguserneed  the daily outdoor water thought to be required for irrigation by house of average personality traits monthly-wateruse  monthly counter of total water used 109  Agent attributes Description monthly-indoor  monthly counter of total water used indoors lastMI  a memory value of the last month's indoor use weekly-count-days-watered  the number of days within a week that an agent has watered the lawn. Utilized when there is a restricted number of days available to water dayswatered  total number of days watered during simulation upgrade-for-economic  tracts whether the agent upgraded for economic reasons spent  money spent by the agent lastbill  memory of the last months bill winterbill  memory of amount spent during winter months learning  the rate at which agents can learn about conservation programs and new information in the model occupancy  the number of occupants on a parcel irrigatoroccupancy  the number of occupants in the parcel irrigator's household desire-to-keep  the value that represents how strongly an agent would like to retain the irrigation rights of their outdoor property Personality  designates which personality type the agent adopts for the simulation. This value effects the starting values for the three personality traits (NEP, ENTITLEMENT, and FORESIGHT) NEP  "New Ecological Paradigm" Likert scale score. Influences water use and willingness to engage in conservation programs. Informed from survey results initialNEP  initial value for NEP score ENTITLEMENT  Likert scale score indicating whether an individual believes heavy users should pay more or not. Influences indoor use and willingness to engage in conservation programs. Latent variable informed from survey results. initialENTITLEMENT  initial value for ENTITLEMENT score FORESIGHT  Likert scale score. Influences water use and the way agents discount the perceived value of upgrading conservation technologies. Latent variable informed from survey results. initialFORESIGHT  initial value for FORESIGHT memNEP  stored value of the agents current NEP score. Used to restore agent's personality scores after they complete the calculations of wateruse using the average user values for their traits memENTITLEMENT  stored value of the agents current ENTITLEMENT score. Used to restore agent's personality scores after they complete the calculations of wateruse using the average user values for their traits memFORESIGHT  stored value of the agents current FORESIGHT score. Used to restore agent's personality scores after they complete the calculations of wateruse using the average user values for their traits  110  Table 12: List of patch attributes and descriptions Patch attribute Description parcelID  an ID number to link all patches that belong to the same parcel FURID  an ID number that identifies patches that represent GIS parcels of the type FUR (Future Urban Reserve) insim?  variable denotes whether a patch runs any procedures or not habitable?  determines whether a house can be placed on a given parcel or not parcelrep?  designates one patch of a parcel to be a parcel representative to speed up querying upgraded?  whether a parcel or patch has had irrigation technology improvements on it or not leased?  designates whether a parcel is leasing its outdoor water use or note infosource?  indicates a patch that is an area that will increase the rate of learning in surrounding patches clusterer?  variable used to designate centres of parcel clusters in simulations where there is an initial amount of landscape being upgraded in a clustered pattern FLU  future land use type area  total area of the parcel that a patch belongs to %cover  estimated % of cover on a parcel. (1 - %cover) is the % of irrigable land in the parcel maxdwellingcount  maximum number of houses on a parcel dwellingcount  current number of houses on a parcel watereq  the daily outdoor water requirements for a parcel concoeff  the coefficient that the outdoor water use is modified by. Changed when agents upgrade their irrigation technology avguserparcelwateruse  total water use on a parcel if the agents on parcel had the average of the initial trait values  parcelwateruse  total water use on a parcel  111  Table 13: List of global attributes and descriptions Global attribute Purpose gisIDlist  a list created from a .txt file containing all valid “parcelID”s found in the supplied GIS .shp file FURIDlist  a list created within the simulation that contains the “parcelID”s of any patches with the “FLU” type "FUR" (Future Urban Reserve) parcelIDlist  a list of the parcelIDs for each parcel represented in the simulation linklist  a list of the total number of links that each turtle has waterlist  a list sorted (based on turtle #) of the totalwateruse values for all agents building-dataset  the dataset that landscape variables from GIS data are loaded into precipitation-data  a list created from a .txt file containing all daily values of precipitation for an entire year in Kelowna, BC  area-data  a list created from a .txt file containing the area values for every parcel sorted by “parcelID”  occupancyfreq  a list of values representing the frequency of the number of individuals that reside in a house from 1 to 6+ individuals personalityfreq  a list of values representing the frequency of personality types that occur  irrigationseason?  indicates whether it is the time of year that people irrigate outdoors or not year  the number of years that have elapsed since the start of the simulation day  the day of the year precipitation  the value of precipitation for the current tick/day days-left-in-week  counter that keeps track of the remaining numbers of days in a week. Important in scenarios that have a restricted number of days of watering total-residential  the total amount of water used by all agents within the simulation govt-cost  the amount that is spent on system maintenance and government payed irrigation leases prevquantity  amount that represents the total water used in the previous month NEPchange  a list containing the values for the difference of (“initialNEP”- “NEP”) for each agent. Sorted by turtles inherent "who" variable value ENTITLEMENTchange  a list containing the values for the difference of (“initialENTITLEMENT”-  “ENTITLEMENT”) for each agent. Sorted by turtles inherent "who" variable  FORESIGHTchange  a list containing the values for the difference of (“initialFORESIGHT”-  “FORESIGHT”) for each agent. Sorted by turtles inherent "who" variable  avgNEP  the mean value of each “initialNEP”score for all agents avgENTITLEMENT  the mean value of each “initialENTITLEMENT”score for all agents avgFORESIGHT  the mean value of each “initialFORESIGHT”score for all agents  first-tier-price  the cost of water within the first tiered volume of the current simulation second-tier-price  the cost of water within the second tiered volume of the current simulation third-tier-price  the cost of water within the third tiered volume of the current simulation remainder-price  the cost of water that surpasses all the tiered volumes of the current simulation  112   Figure 27: User interface of the model. A = Initialization and reset buttons, B = System options, C = Policy options, D = Display options, E = Map output, F = Additional outputA B D C E F 113  Appendix C ODD +D Technical Description Overview  1. Purpose  This model is meant to be used as a tool to explore the effects of social complexity and policy implementation on urban water use. The model is set in Kelowna, BC Canada. Modelling a real location allows for the incorporation of GIS data and social survey data whose context may or may not be more broadly applicable to other locations.   2. Entities, state variables, and scales  Agents:  There is one agent type represented in this model, the home owner/water user.   Agent state variables:  Described in Appendix B  Landscape/Patches:  The landscape of this model is functionally constructed of parcels of land represented by collections of patches   Patch variables:  Described in Appendix B  114  Temporal/Spatial Scales:  This model’s time steps represent a single day within an entire year that is simulated. The spatial extent of the model represents the city limits of Kelowna, BC Canada.   Global Environment Variables:  “gisIDlist” - a list created from a .txt file containing all valid *parcelID*s found in the supplied GIS .shp file  “FURIDlist” - a list created within the simulation that contains the parcelIDs of any patches with the FLU type “FUR” (Future Urban Reserve)  “parcelIDlist” - a list of the parcelIDs for each parcel represented in the simulation  “linklist” - a list of the total number of links that each turtle has  “waterlist” - a sorted list (sorted based on turtle #) of the totalwateruse values for every house  “building-dataset” - the dataset that landscape variables from GIS data are loaded into  “precipitation-data” - a list created from a .txt file containing all daily values of precipitation for an entire year in Kelowna, BC  “area-data” - a list created from a .txt file containing the area values for every parcel sorted by parcelID  “occupancyfreq” - a list of values representing the frequency of the number of individuals that reside in a house from 1 to 6+ individuals  “personalityfreq” - a list of values representing the frequency of personality types that occur  115  “irrigationseason?” - indicates whether it is the time of year that people irrigate outdoors or not  “year” - the number of years that have elapsed since the beginning of the simulation  “day” - the day of the year  “precipitation” - the value of precipitation for the current tick/day  “days-left-in-week” - counter that keeps track of the remaining numbers of days in a week. Important in scenarios that have a restricted number of days of watering  “total-residential” - the total amount of water used by all agents within the simulation  “govt-cost” - the amount that is spent on system maintenance and government payed irrigation leases  “prevquantity” - amount set monthly that represents the total water used in the previous month  “NEPchange” - a list containing the values for the difference of (initialNEP - NEP) for each agent. Sorted by turtles inherent “who” variable value  “ENTITLEMENTchange” - a list containing the values for the difference of (initialENTITLEMENT - ENTITLEMENT) for each agent. Sorted by turtles inherent “who” variable value  “FORESIGHTchange” - a list containing the values for the difference of (initialFORESIGHT - FORESIGHT) for each agent. Sorted by turtles inherent “who” variable value  “avgNEP” - the mean value of each initialNEP score for all agents  “avgENTITLEMENT” - the mean value of each initialENTITLEMENT score for all agents  “avgFORESIGHT” - the mean value of each initialFORESIGHT score for all agents  116  “first-tier-price” - the cost of water within the first tiered volume of the current simulation  “second-tier-price” - the cost of water within the second tiered volume of the current simulation  “third-tier-price” - the cost of water within the third tiered volume of the current simulation  “remainder-price” - the cost of water that surpasses all the tiered volumes of the current simulation   3. Process overview and scheduling  Every day the following sequence of events and processes relevant to agent actions occur  Environmental variables for the day are updated  Houses assess indoor water need and use water indoors  If it is irrigation season, houses that are irrigators will assess water need outdoors and use water outdoors  Weekly events are run every 7 days 4.1 If houses do not have upgraded irrigation and it is irrigation season and they are an irrigator, they will assess the possibility of upgrading their irrigation 4.2 If leasing irrigation rights is possible yet not been implemented by a house and it is irrigation season, they will assess the possibility of leasing their right 4.3 If the scenario allows for social influence between agents, influence occurs  Monthly events are run every 30 days 5.1 Agents pay for their water use 117  5.2 The water infrastructure is maintained 5.3 Usage is visualized  The simulation progresses one time step   Design concepts  4. Theoretical and Empirical Background  See literature review   5. Individual Decision-Making  Within this model houses that represent collections of people sharing a domicile make decisions on their residential water use. Residential parcels/lots can contain multiple domiciles and although each domicile on a parcel will make decisions on indoor water use, only one residential unit within a parcel will act as the irrigator and use water outdoors.  The model user acts as the regulatory body and makes decisions on growth patterns and water conservation measures in the UI. The household agents then make their residential water use decisions in the confines of the defined environment.  In accordance with the Theory of Planned Behavior (TPB) (Ajzen 1991) 3 things contribute to an agent’s intention to make a decision: their attitudes towards the behavior, subjective norms, and their perceived behavioral control.   118  Effects of Attitudes  Each agent is given 3 personality traits that affect their water use decisions. It is assumed that individuals that share a living space share similar attitudes towards water use and therefore share personality trait values within the model. This was done for simplicity purposes and to reduce computational demand.  The 3 personality traits are titled “ENTITLEMENT”, “FORESIGHT”, and “NEP”. These variables are informed with data from water attitude surveys conducted by Dr. John Janmaat. Each is a value on a spectrum from 1 to 7.  ENTITLEMENT : the opinion that those who use more water should pay for it. Agents with scores closer to 1 believe that those who use water should pay for it, while agents with scores closer to 7 believe that heavy users should not be subject to a high cost  FORESIGHT : the opinion that there is the possibility that the region will suffer water shortages in the future. Agents with scores closer to 1 believe there is little to no chance of water shortages, while agents with scores closer to 7 believe there is a significant chance of water shortages  NEP : the New Ecological Paradigm (Dunlap et al., 2000) score represents the dichotomy between eco-centric and anthropocentric. Agents with scores closer to 1 place greater value on human development, while agents with scores closer to 7 place greater value on nature  It is assumed that these variables have an effect on the water use decisions of an individual.  Details on how each attribute specifically effects which decisions can be found in the Submodels section of this document   119  Effects of Subjective Norms  Subjective norms are incorporated into the model in 3 distinct ways  Agents will evaluate the perceived cost of upgrading their irrigation technology dependant on the prevalence of this behavior in their social network. If nobody in their network has upgraded their irrigation, the upgrade is seen as 10% more costly. Each person in their network past the first that does have the upgrade will grant a 10% reduction in perceived cost. This process occurs on a weekly basis  Agents have the potential to lease their irrigation rights dependant on the number of individuals in their social network that have done so. There is a 5% chance of conforming for each individual in an agent’s network that has leased their rights. This process occurs on a weekly basis  Agents have the potential to influence the attribute values of those in their social network. If an agent has sufficiently similar trait scores to another in its social network it has a 50% chance to slightly modify the other’s score to be closer to their own. This process occurs on a weekly basis   Effects of Perceived Behavioral Control  An agent’s perceived behavioral control is represented by a “learning” variable within the model. The higher the value of this variable, the more likely an individual is to learn of their ability to either lease their watering rights or upgrade their irrigation technology. Even if the agent is morally obligated to modify their behavior, they are unable to do so if there is no perceived behavioral control (e.g. they have not learned that they can).  120  The initial value for the variable is set in the UI and there are two simulation events that can increase this value. If an agent is in proximity to an education source then they will increase their learning score relative to how close they are to the source. Additionally, if someone within the agent’s social network has the irrigation technology upgraded on the parcel within which they are residing, the agent will increase their learning score by 5%.   6. Learning  As described in the section Effects of Perceived Behavioral Control there is a direct learning variable for each agent. This variable has effect in decisions of leasing water rights or upgrading irrigation technology. Learning is simulated by comparing randomly generated numbers with the learning variable of an agent, therefore each agent has the potential to learn of their options but those with higher learning scores will have a more likely chance.  Within the procedure of evaluating leasing or upgrading, an agent must first pass an intention function and then a learning function. This order of events is modeled after the TPB in which even if there is intent to do an action, the lack of perceived behavioral control can halt that action  The agents also learn in a broader sense from their social connections. Through social interactions agents can change their attitude scores, representing the ability to learn from the opinions of those in your social network. This change in attitude has the potential to effect the agents’ water use behavior    121  7. Individual Sensing  Endogenous variables  Agents can assess their own last month’s bill/usage and the bill/usage from the winter months without error. There is no cost associated with acquiring any of this information.   Exogenous variables  Agents can sense the day, precipitation, and water prices without error. They also sense variables of those in their social network without error including others’ personality traits, and whether they are leased or upgraded. They can also sense the area, %cover, and water requirement of the parcel that they reside on without error. There is no cost associated with acquiring any of this information.   8. Individual Prediction  While the agents are assessing upgrading their irrigation technology, they make predictions on the perceived benefit and the perceived cost of the upgrade.  Since the perceived cost is likely easier to predict, it is predicted based on the actual cost of the upgrade modified by the prevalence of that upgrade in the agent’s social network. “Cost” in this sense incorporates not only the monetary cost but the perceived social cost of modifying one’s outdoor space.  The perceived benefit of an upgrade is more speculative in nature and therefore affected by error. The agent assess how much they spent on outdoor water use by comparing their previous 122  month’s bill with their winter water bill. They then estimate the amount of money they would have saved by dividing this difference in the bills by ((random-float 0.333) + 0.333). The upgraded irrigation cuts outdoor watering need by half but this inclusion of randomness accounts for agents potentially erroneous prediction of the exact cost savings and so agents estimate savings between 33.3% and 66.6% savings. The agents use this predicted value to also estimate the savings for the next two billing periods   9. Interaction  There are four ways in which agents directly interact with each other.  (1) when an agent upgrades their landscape they increase the learning variable of their link neighbours by 0.05 (a 5% increase in the likelihood the linked agent will learn about conservation options)  (2) agents who have upgraded landscapes effect their link neighbours perceived cost of upgrading their own landscape  (3) if greater than half of an agent’s social network has leased irrigation rights then there is a chance for an agent to also participate in the program due to social pressures  (4) agents ‘communicate’ on a weekly basis and have the opportunity to change each other’s personality traits if they are relatively similar but not equal. These interactions depend on the agents’ trait values and thresholds set in the UI.  Interactions occur through social networks which are created in initialization. The number of connections and types of connections that each agents creates are imposed in the UI. The social 123  networks are heterogeneous between agents as any agent may have had connection made with it before or after it gets the chance to create its own connections, leading to agents having a ranging number of connections.   10. Collectives  The only collectives that are formed in this model are the social networks that each agent has. The aggregation has an effect on the agent in the sense that it is likely that the agent will move towards the average value of their social network as long as there are individuals in the network with sufficiently similar attitudes as the agent to ‘convince’ them. Agent’s similarly have an effect on the aggregate in that they contribute to this mean group value for other agent’s social networks   11. Heterogeneity  Agents differ in both their initialized attributes and also in the way that they run procedures based on these traits. For example, agents that have particularly eco-centric NEP scores may upgrade their landscape for moral reasons, a procedure that could not be accessed by other agents with different NEP scores.  The agents are also heterogeneous in their spatial context. Not only are the spatial attributes of the parcels they reside on different, but where they are placed has an effect on who they make social connections with as agents make a certain number of connections based on proximity.  124  The decision making processes are largely heterogeneous as well. Many of the decisions have a component that is dependent on personality trait values (which are heterogeneous) and also a random component that adds additional heterogeneity   12. Stochasticity  Randomness is incorporated in both the initialization procedures and in the running procedures of the model to account for processes that are either partly random or to account for extraneous drivers that likely impact the system but that are not directly modeled.   Randomness in Initialization  Parcel occupancy is determined in a semi-random way. For each dwelling on a parcel a group of between 1-6 occupants is added to the dwelling based on the distribution observed in census data for the area (Statistics Canada, 2011a).  Stochasticity is incorporated into the creation of an agent’s social network. “Close-social-connections” are created with agents that are closest to the agent making the connections. If there are multiple agents that tie for closest, a link will be made with one at random. The "random-social-connections” are made by randomly linking with any other agent.  If an “info-source” is being created, the source is placed on a habitable patch either randomly or if “target-high-density-areas?” is selected in the UI, randomly amongst only the habitable patches with the highest number of houses surrounding them while preventing too many sources from being close together.  125  The agents are assigned a personality type semi-randomly. The selection probability is based on the proportion of personality types observed in the cluster analysis performed on survey data.  Each agent is assigned personality variables values randomly from a normal distribution of values dependant on the distribution of these values in the survey data for each personality type.  The area covered within a parcel is also randomly assigned from a normal distribution of values with a maximum of 0.5, minimum of 0.1, and mean of 0.25. These values were determined through Kelowna zoning bylaws on maximum parcel coverage   Randomness in Procedures  General  The human decision making process incorporates many interacting components that are not fully understood and therefore difficult to model. Randomness was incorporated in parts of the decision-making procedures to attempt to account for, at least in part, some of the additional factors that might play a role in residential water use decisions and the associated actions within this model.   Assessing Water Required  If the current days precipitation is enough to meet the water requirements of the parcel an agent resides on, there is a partially random chance the agent will not water if personality-effects-behavior? is turned on. In the case that it is turned off, all agents will not water when there is 126  enough precipitation. When turned on though, the likelihood of not watering is dependent on an agent’s NEP score as moderated by the following if statement:  if random-float 1 < ((8 - NEP) / 7) [report 0]  The effect of this is that those individuals that are less concerned with ecological values (low NEP score) are less likely to be prudent with shutting of their water on days that there is precipitation  This same if-statement is used in the case that there is some precipitation but not enough to meet the demand of the parcel. The rate of agents adjusting their additional watering needs to the appropriate level is also dependant on this semi-random qualifier.  Finally, when assessing the amount of water needed on the agent’s parcel, the agent has the potential to consider the cost of additional watering and this process also incorporates a random element through the following:  if random-float 1 < consider-cost [report 0]  For details on consider-cost, see the submodel description below. In short it is a value from 0 to 0.5 that increases once the previous month’s bill surpasses the winterbill by a threshold defined in the UI and maxes out at 0.5 once the bill passes a second max-cost-aversion-at threshold defined in the UI   Restricted Watering Frequencies  If there is a restriction on the frequency of watering allowed in the scenario then houses are assigned days to water each week in a semi-random fashion. This was done to allow for the 127  restriction to any number of days per week in place of rules that are dependent on house number (data which was unavailable). Although this may change the pattern of spatial water use at a small temporal scale (intraweek), the effect on a longer time scale (year) will be relatively negligible.  While watering restrictions are in place, agents still have the ability to disobey the restrictions at a rate specified in the UI. The way this is implemented in the model differs depending on if personality-effects-behavior? is turned on in the UI.  If it is turned off, the agents homogeneously have an equal rate of disobeying the restrictions as set in the UI’s rate-of-compliance. If it is turned on, then only agents with ENTITLEMENT scores that are above the population average will have a chance of disobeying at a rate that preserves the overall population’s rate as defined in the UI.   Checking Neighbors for Leased Irrigation Rights  When checking the link neighbors of an agent for leased irrigation rights, a value is compared again to a random-float 1 value to determine if the agent will lease their property rights because of social influence. Details can be found in the submodel description section.   Assessing upgrades and leases  Randomness is incorporated in both of these procedures to account for factors in the Theory of Planned Behavior. A qualifier must be passed to turn intention into action (if random-float 1 < 128  0.2 then intention != action) and an agent must also learn of their perceived behavioral control (if random-float 1 > learning then the agent does not have perceived behavioral control)  Within the upgrading procedure the agent has the potential to upgrade for moral reasons and this is partially random. If the agent has sufficiently high NEP score (oriented towards ecological values) then there is a chance of them upgrading regardless of cost at a rate related to their ENTITLEMENT score (if random-float 7 >= ENTITLEMENT they have the opportunity to upgrade)  Randomness also plays a role in the agents’ assessment of the perceived benefit of upgrading their landscape. Details can be found in the submodel description   Social Influence  When agents are influencing each other, if an agent has the potential to influence another there is only a 50% chance that it actually will   Emergence  Clusters of houses can emerge as water savers. This is due to the spatial interactions that occur between close houses. As one house in a grouping changes behavior for ‘moral’ reasons (i.e. due to personality traits) this can reduce the perceived social cost of changing behavior for connected houses and therefore increases their likelihood of changing as well.    129  Details  14. Implementation Details  This model has been programmed and implemented in NetLogo and is available from the author by contacting JBepple@gmail.com.  15. Initialisation  When this model is opened in Netlogo the following occurs as a “startup” procedure:  GIS data is projected onto the landscape and the .shp file is read into lists. The parcelID and FLU are imprinted on the landscape from the dataset. Area values are then assigned based on parcelID. Patches that don’t have FLU = “FUR” set their FURID to -1 so as to not interfere with “FUR” specific procedures. Finally the maxdwellingcount is set for the patches  If the Scenario = “Restrictive Growth” the rezone-for-restricted-growth procedure is run. If the Scenario = “Urban Sprawl” or “Densification” then the segment-urban-reserve procedure is run, effectively transitioning “FUR” parcels to liveable parcels.  Parcels with FLU = “S2RES”, “S2RESH”, “MRM”, “MRH”, “MRL”, “MXR”, “MXT”, or “MRC” are set as habitable? = true parcels. Parcel representatives are assigned, one patch for every parcel.  Parcels that do not meet the minimum area for their land use type (according to Kelowna Zoning Bylaws) are corrected to reflect the minimum lot size. Any parcels that do not have parcelID that match up with actual habitable parcels, as found in the full resolution .shp GIS files, are removed from the simulation.  130  Coverage data is then added for any habitable patch.  After “startup” procedures are finished running and the user presses the “Initialize” button the following occurs:  Agent, patch, and global variables and lists are created or set at initial values for later modification or use.  The prices are set within the simulation dependant on whether the user has selected “low” “medium” or “high” price structures.  Watereq is set for patches as (area * (1 - %cover) * lawn-irrigation-requirement)  If set-%upgraded is turned on in the UI then the set-initial-upgraded procedure is run  The occupancyfreq list is created with the occupancy rates found in Kelowna.  Turtles are then created. There is a maximum of 1 turtle per parcel. If the user has defined a number of turtles higher than the maximum, additional turtles will instead add +1 dwellingcount to an already existing turtle and therefore add to the overall occupancy on the parcel without adding an additional agent. This was done to reduce computational load. The location the turtles are placed within the parcel changes between simulations and the number of additional dwellings added to a parcel also changes between simulations.  Agent occupancy is set by drawing a number from the occupancyfreq list for every dwelling that the agent represents. This value changes between simulations  Links are then created in a semi-random way that changes linkages between agents in different simulations  131  Personalities and trait values are then assigned to each agent through the assign-personality submodel.  The agent’s winterbill is set as the monthly-flat-rate plus either (30 * daily-indoor-use * (ENTITLEMENT / 2.418) * irrigator-occupancy * first-tier-price / 1000) if personality effects behavior, or (30 * daily-indoor-use * occupancy * irrigator-occupancy * first-tier-price / 1000) if it doesn’t. In the situation that personality effects behavior the agent’s ENTITLEMENT score is divided by 2.418 (the mean value for this trait in the population) because there was a strong statistically significant (p<0.0001) relationship between a respondents ENTITLEMENT score and their number of self-reported indoor water conservation behavior (higher score had less conservation behavior).  The list of precipitation values in then created from a .txt file.  Finally, if the scenario calls for “education centres” then they are distributed throughout the landscape either randomly or in areas with the highest densities of agents if the target-high-density-areas? is on in the UI. Once the patches to be exposed to educational efforts are selected, the become-info-source procedure is run.   16. Input data  This model utilizes GIS datafiles to inform the landscape variables within a simulation. Some variables from the .shp file are kept as .txt files (e.g. the area). Due of the way that Netlogo averages values for patches that overlap with multiple polygons, storing values such as area in .txt files can ensure that the whole, true value for a parcelID is being represented.  132  The precipitation data is also stored in a .txt file that is read into the simulation and stored in the root directory with the model.   17. Submodels  set-environmental-variables  First the days-left-in-week variable is set for use in watering frequency accounting in scenarios that include restricted watering frequencies.  At the end of every year a set of lists are written. A “waterlist” is written that contains the totalwateruse for each house sorted by house number. This allows for an assessment of the distribution of water use across houses. A list is then written for each of the 3 personality variables to record the amount that the agent’s personality has changed from the initial values. These trait lists are also sorted by house number.   assess-and-use-indoors  The basic indoor use value is set within the UI. When a house uses water indoors it simply multiplies the UI set value by its occupancy score to get a total value. If personality-effects-behavior? is turned on in the UI, this indoor use value is further modified by the households ENTITLEMENT score.  Within the survey data used to inform this model, the questions that contribute to the ENTITLEMENT score were found to be strongly correlated (p < 0.0001) with indoor water conservation behavior. In the model this is implemented by modifying the amount used indoors 133  by a factor of (ENTITLEMENT/ 2.418) (2.418 represents the mean ENTITLEMENT in the data)  Once a final value for the indoor water use has been determined by an agent the value is accounted for in the monthly-indoor and totalwateruse of the agent and in the patch variable parcelwateruse and global variable total-residential.  The agent then switches their personality to that of a resident with average trait values and goes through the above procedures again to determine that amount that would have been used on that parcel by a homogeneous average agent. This indoor use value is then stored in the patch variable avguserparcelwateruse to compare the parcels water use to the amount that would have been used by an average user.   assess-and-use-outdoors  Agents that are irrigators and are not on properties with leased irrigation rights go through the following procedures  If restricted-watering? is off in the UI the agent proceeds directly to the “begin-outdoor-use” procedure.  If restricted-watering? is on in the UI the agent performs 2 actions.  (1) If the current day of the week matches a day on the agents permitteddayslist the agent will continue to the begin-outdoor-use procedure  (3) If the agent has been denied the ability to water this turn they then have the opportunity to be noncompliant and water anyways. This happens in one of two ways depending on if personality-134  effects-behavior? is on in the UI or not. If it is off the houses will generate a random number from 1 to 100 and if that value is above the rate-of-compliance set in the UI, then they will proceed to the “begin-outdoor-use” procedure, otherwise they will set their waterneed for the day to 0.  If personality-effects-behavior is on then only agents with ENTITLEMENT scores above the average will have the opportunity to be noncompliant. They do so randomly at a rate that preserves the expected population frequency set in the UI.   begin-outdoor-use  The first thing irrigating agents do in this procedure is assess their waterneed. In doing so they have the opportunity to (1) forgo watering if there is rain, (2) reduce their watering if there is rain, (3) reduce their watering based on personality traits, (4) reduce their watering need based on how efficient their irrigation technology is, and (5) consider the cost of their irrigation.  (1) The amount of precipitation expected on the uncovered portion of the parcel is compared against the watering requirements on the patch. If there is more precipitation than is needed for irrigation then the agent does one of two things depending on if personality-effects-behavior? is on or not. If it isn’t on then all agents will not water outdoors. If it is on, they will set their outdoor water use to 0 if the following function produces “true” for them:  if random-float 1 < ((8 - NEP)/7)  135  The agent’s rate of not watering is modified by their NEP score because it is assumed that those individuals that are more eco-centric (high NEP score) will be more prudent about not watering on days that it is not necessary.  (2) If there is rain but not enough to fulfil the water requirements of a patch then one of two things will happen depending on if personality-effects-behavior? is on or not. If it is off then all agents will reduce the amount that they water outdoors by the amount of precipitation they get. If it’s on then the rate of agents reducing their water use by the precipitation amount is modified by their NEP score in the same way described in (1) above.  (3) If there is no rain, or if there is rain an agent still has a non-zero value for their assessed outdoor irrigation needs then there is the opportunity to reduce use for moral reasons if personality-effects-behavior? is on in the UI. If the agent has a particularly high NEP score (is quite eco-centric) or has a high FORESIGHT score (strongly believes there could be water shortages in the future) then they will reduce the amount of water they perceive to be needed by 25%  (4) Agents will reduce their water use by 50% if they have upgraded irrigation  (5) Finally agents will consider the cost of outdoor water use. They compare the amount they spend in the winter to the amount they spent on their last bill. Once the amount reaches the consider-cost-at value (representing the percentage of the winter bill that must be exceeded before the agent will consider the cost) the function will produce a value of 0. This value increases linearly up to a value of 0.5 for individuals at or above the max-cost-aversion-at value (percentage above winter bill that will produce the strongest response in reduction of watering). 136  The produced value will be compared against a random-float 1 value and if the random value is less than the produced value then the agent will set their water needs to 0.  After an appropriate amount of watering has been determined, the agent will use the water, account for it in their agent variables, patch variables, and global variables, and then rerun the procedures with average personality trait values for comparison purposes.   assess-upgrade  Assessment of the desire to upgrade the irrigation technologies of a parcel occur differently depending on if personality-effects-behavior? is on or not but in both cases the decision, at least in part, depends on the agents perception of the cost and benefit of upgrading their landscape, described below.  It also depends on random qualifiers implemented to account for the intention of an action not always leading to the action itself as per the TPB (i.e. if random-float 1 < 0.2 then intention != action) and the necessity of the act of learning of the agent’s perceived behavioral control as per the TPB (if random-float 1 > learning then the agent does not learn of their potential for behavioral control.   perceived-cost  First, the agent sets the cost for upgrading as the value in the UI for cost-of-upgrade multiplied by the area of the parcel being assessed. If social-influence? is turned off in the UI, then the agent stops here in perceiving the cost.  137  The agent modifies the cost of upgrading their landscape in one of two ways based on the number of its social connections that have upgraded their landscape. If none of its social connection have the upgrade then the cost of the upgrade is multiplied by 1 + social-influence-on-cost. If any of the houses social connections have the upgrade, the cost is reduced by the following amount:  cost * (social-influence-on-cost * #upgraded)  At a social-influence-on-cost value of 0.1 the effect of this function is to reduce the perceived cost to the agent by 10% for each of its connections already with the upgrade.   perceived-benefit  First the agent determines the amount that they spent in the previous billing cycle on outdoor water by subtracting their winterbill amount from their lastbill. The agent then estimates the amount they would save by multiplying their last month’s outdoor usage cost by ((random-float 0.333) + 0.333) . Therefore the agent assesses the amount they will save in the future from upgrading as between 33% - 66% of their previous outdoor watering bill (actual improvement is 50%).  The agent then predict the potential long term benefits of upgrading their landscape. The do this by applying an economic temporal discounting equation to the expected value to be saved for an additional two billing periods after the next. The general form of this equation is :  TotalCurrentValueOfBenefit(@t=x) = CurrentValueOfBenefit(t0) + CurrentValueOfBenefit(t1) + CurrentValueOfBenefit(t2) + …. CurrentValueOfBenefit(tx)  138  where…  CurrentValueOfBenefit(tx) = CurrentValueOfBenefit(t0) * Discount(x)  and…  Discount(x) = 1 / (1 + discount-rate) ^ x  This equation is used to sum the expected value to be saved in the next immediate billing period plus the discounted value of the amount saved over the following two billing periods   assess-lease  In scenarios where leasing an agent’s irrigable area is an option, leases can be implemented one of two ways. Either the irrigator pays the lease to have the governing body to assume responsibility for irrigation, or the governing body pays the irrigator an amount for the opportunity to control the outdoor irrigation.  The procedure changes depending on who the lease-payer is but in both cases before an agent can consider leasing they must learn about the option to do so. This is done by comparing a random-float 1 value to their learning score. If their random value is below their score, then they move on to their next qualifying function.      139  lease-payer = “households”  In this case the decision to lease the watering rights is a function of the amount the agent has spent on outdoor watering, the lease-amount, a factor termed desire-to-keep, and the agents learning score.  The agent then assess whether the amount they spent outdoors (estimated by deducting their winterbill form their previous months bill) is less than the lease-amount that they would have to pay multiplied by the desire-to-keep factor.   lease-payer = “government”  With the governing body paying the land owner to take over the responsibilities and costs of outdoor irrigation, leasing is more attractive in this scenario. If the agent learns of their ability to lease then the decision to lease is made if the amount they received is greater than their desire-to-keep factor multiplied by 10. Since the desire-to-keep is no longer being used as a factor to adjust a monetary value, we multiply by 10 here to account for additional factors that may be at play in this decision making process   check-leased-neighbors  If agents do not have their property leased they also have the potential to lease their property as a result of social influence. Once half of an agents neighbors have converted to leased irrigation, the agent has a 10% chance of converting to leased irrigation for each additional household after the first half in their social network with leased property.  140   influence-social  If social-influence? is turned on in the UI then the agent runs through their influential procedures once a week. Each of the agents personality traits has the opportunity to be changed by those in the agent’s social network. The same procedural framework is used for each personality trait (TRAIT refers to NEP/ENTITLEMENT/FORESIGHT in the following equation).  if (TRAIT - ((standard deviation of TRAIT) * threshold-of-influence / 100)) < (social connection's TRAIT) and (social connection's TRAIT) < (TRAIT + ((standard deviation of TRAIT) * threshold-of-influence / 100))    [   if random-float 1 > 0.5   [      let tempnumber (TRAIT - (social connection's TRAIT))     set TRAIT (TRAIT - (tempnumber * strength-of-influence / 100))     if TRAIT < 1 [set TRAIT 1]     if TRAIT > 7 [set TRAIT 7]   ] ] 141  In the first “if” statement, the equation is functionally assessing whether the social connection has a trait value that is within a certain range of its own, modeled after the Relative Agreement model (Deffuant et al 2002). The range is determined by +/- the standard deviation of the trait modified by the threshold-of-influence set in the UI. At a threshold-of-influence value of 100, the range is +/- 1 standard deviation, while at a value of 200 the range is +/- 2 standard deviations etc.  When an agent is communicating with another in its social network and is within the threshold of being able to be influenced, the agent has a 50% chance to actually be influenced. If influence does occur, the strength of the effect is determined by the difference between to two agents’ traits and the strength-of-influence set in the UI. At a strength-of-influence level of 100, the agent being influenced will shift their trait the entire difference between the two agent’s traits to make themselves have the same value as their influencer, while at a strength-of-influence value of 50, the agent being influenced will only shift their trait closer to their influencer by half the difference.  The final step of this procedure is a failsafe to ensure that values do not exceed the maximum or reduce past the minimum of the personality traits.   settle-bills  Once a month the agents will settle the bill for their monthly wateruse. Depending on if the irrigation rights are leased or not this occurs in one of two ways.   142  bill-lease  If the lease-payer is the household agent then the agent’s spent value is increased by the value of the lease-amount and the govt-cost is reduced by that amount. If the lease-payer is the governing body then the reverse occurs for accounting.  At the end of this procedure the agent sets their lastbill as the value that they spent on water last month (which will only be the indoor amounts), store their monthly use in the lastMI variable and then reset their monthly-indoor and _monthly-wateruse_to 0   bill-user  Agents add the amount of their bill to their spent variable, set their lastbill as the value of the bill, set their lastMI as their monthly-indoor amount and then reset their monthly-indoor and monthly-wateruse amounts to 0. The following reporter is used to calculate the bill   to-report bill  The agent first sets a temporary value “X” as the monthly-wateruse (total amount that was used by all the residents within the household). The agent then removes the total amount used indoors, monthly-indoor (amount used by all occupants that make up the occupancy value). The indoor use from the individual agent paying the bill ((monthly-indoor / occupancy) * irrigator-occupancy) is then added back to X. X now represents the entire outdoor water use plus the indoor use of the agent paying the bill but excludes the indoor use from the remainder of the occupancy which would be paying their own bills for water.  143  The agent then takes the value X and separates this volume into the amount allowed for first, second, and third tier pricing with any remainder that surpasses these allowed volumes going into a “remainder” category. Each tier volume is then multiplied by that tier’s price and then summed together with the monthly-flat-rate to get the total cost of the bill   maintain-system  The totalwateruse of all houses is summed and stored as a temporary value. This temporary value then has prevquantity removed from it (the sum of totalwateruse of all the houses from the last time maintain-system was run). The temporary value is then multiplied by the maintenance-cost set in the UI divided by 1000 and added to govt-cost. Finally, prevquantity is set as the current totalwateruse of all houses   set-initial-upgraded  Temporary numbers “x” and “y” are first created. X represents the amount of landscape upgraded so far, and is initially defined as 0. Y represents the total amount of area to be upgraded as governed by the UI designated %upgraded before the following procedures begin. This procedures then runs one of two ways depending on if the upgraded parcels are being clustered or not.  (1) if the upgraded parcels are NOT being clustered, while x < y, a parcel that has an area < (y - x) will be upgraded. Once a parcel has been upgraded, its area is added to the value x. If there are no parcels that have an area < (y - x) then the procedure stops which results in the final amount 144  of landscape upgraded as equal to or less than (by <1% generally) the amount designated by %upgraded.  (2) if the upgraded parcels are being clustered, first a number of intial cluster centres are designated according to the amount set in the UI by number-of-clusters. The area of each of these parcels is added to x. While x < y the cluster centres search for additional parcels to upgrade within a radius that starts at 1 and increases by 1 each time there are no upgradeable parcels found within the search radius. If parcels are found within the radius that is being searched that have areas < (y - x) then they are upgraded and their areas are added to the value of x. If the area of the non-upgraded parcels in the search radius is greater than (y - x) then the parcel with the smallest area in the simulation is upgraded and its area is added to x.   rezone-for-restricted-growth  A list is created of the parcelID of each separate land parcel that has FLU = “FUR” (Future Land Use = Future Urban Reserve). A parcel representative is set for each parcelID on the list and then these parcels are given the new FLU of “Undeveloped”.  Then 1% of the Single / Two Unit Residential (“S2RES”) and the Single / Two Unit Residential Hillside (“S2RESH”) are converted to Medium Density Multiple Unit Residential (MRM)      145  segment-urban-reserve  A list is created of the parcelID of each separate land parcel that has FLU = “FUR” (Future Land Use = Future Urban Reserve). A value is created called “tempticker” for use later in the procedure.  For each of the parcels, the minimum and maximum X and Y coordinates are found and stored as temporary values. The minimum values for each are set as the initial values for local variables in this procedure called “lastx” and “lasty” An increment is then set as 2 patches in the Scenario “Urban Sprawl” or 4 patches in the Scenario “Densification” and the initial values for the local variables “nextx” and “nexty” are set as the value of “lastx” and “lasty” plus the increment.  In the “Densification” scenario, a portion of each parcel is converted to “PARK” equal to % designated in the user interface.  While the value of “lastx” < the maximum X coordinate of the parcel and the value of “lasty” < the maximum Y coordinate of the parcel the following procedure will run:    ask patches with [parcelID = ? and pxcor >= lastx and pxcor < nextx and pycor >= lasty and pycor < nexty]    [     set pcolor tempticker      set FURID tempticker    ]     set tempticker tempticker + 2 146     if tempticker mod 10 <= 1    [       set tempticker tempticker + 2    ]    ifelse (lastx + 2) < (max X coordinate of parcel)     [      set lastx nextx      set nextx nextx + increment    ]    [      set lastx (min X coordinate of parcel)      set nextx lastx + increment      set lasty nexty      set nexty nexty + increment    ] For each of the parcels that are Future Urban Reserve the procedure will break each parcel into squares of size “increment” x “increment” starting at the lowest X and Y coordinates of the parcel. Each square is assigned a colour and a new ID, a “FURID”, based on the “tempticker” which is updated each time a new square is created within the parcel. The procedure first 147  proceeds along the lowest values of Y and from left to right for values of X. Once the maximum X coordinate will be exceeded by the next iteration, the procedure reverts back to the minimum value of X and increases the Y values being processed.  Once all parcels have been iterated through the procedure, the finish-urban-reserve procedure is called   finish-urban-reserve  Local variables are set within the procedure for the minimum and maximum lot sizes of “S2RES”, “S2RESH”, and “MRM” parcels. The minimum is set as the minimum value already found within the area of interest. If the value is below 201 for “S2RES” or “S2RESH” properties then the minimum value is set as 290 as this is the minimum lot size for this type of land use found in the Kelowna Zoning Bylaws. The maximum area size for the lots is then set as the mean lot size value of all parcels of that type found in the GIS files available in Kelowna’s online Open Data Catalogue in 2015. The mean was selected as the future maximum because it is assumed that the size of new lots in general will decrease as the population of the city increases. The current distribution of lot size is skewed to high values (“S2RES” mean = 1003, median = 838, 3rd quartile = 1036 | “MRM” mean = 4687, median = 2936, 3rd quartile = 5090).  Each square with unique FURID calculates the proportion of the original parcel’s area that it represents. If the square represents an area that is less than the minimum area for the land use type that it will turn into (“S2RES” in Urban Sprawl scenarios and “MRM” in Densification scenarios) the square will run through the dissolve-property procedure. If the square represents 148  an area larger than the maximum for the land use type it will turn into it runs the split-property procedure.  Once all squares have either split, been dissolved, or been left alone depending on where in the allowable range of areas the square lies, the finish-urban-reserve procedure continues  A list of all squares with unique FURID values is created. One patch for each FURID calculates the proportion of area that that FURID represents of the original parcel. This area value is assigned to itself and other patches that share the FURID along with a new unique parcelID.  The patches are then converted from FLU = “FUR” to FLU = “S2RES” (urban sprawl scenarios) or to FLU = “MRM” (densification scenarios) and the maximum dwelling count is set for these newly habitable patches.   dissolve-property  If there are patches that share the parcelID with the square being dissolved but not the FURID (i.e. represent a different square within the parcel from the segment-urban-reserve procedure) then the square proceeds with the following (otherwise nothing happens and an ERROR message is displayed in the Command Center):  The square will find the closest other square within the original parcel (shares parcelID but not FURID) and convert the patches within its square to the new adjacent squares FURID     149  split-property  If there are 2 or more patches that make up the square with too large of area then the patches run through the following (otherwise nothing happens and an ERROR message is displayed in the Command Center):  The minimum and maximum X and Y coordinates are found of the square being split. The square is then cut in half along its longest dimension and the new half is assigned a new FURID   assign-personality  Assigning personality types and traits variables for the agents occur one of two ways depending on if the water saving agents are being clustered or not. In both cases agents can be assigned one of three personality types: P1 (the water wasters), P2 (the average heavy user), or P3 (the water savers)  (1) If the water saving agents are NOT being clustered then agents will be randomly assigned a personality type. The proportion of each agent type is either what was observed in our data or if the set-%watersavers option is turned on in the UI then the prevalence of P3 agents are fixed in the UI and the relative ratio of P1:P2 is preserved for the remaining agents.  (2) If the water saving agents are being clustered then variables “x” and “y” are created in the procedure. X represents the current amount of P3 agents and therefore starts at 0. Y represents the total number of agents to be converted to P3 personality type as desginated by the %watersavers variable in the UI. Next a number of agents equal to the number-of-clusters variable set in the UI are converted to personality type 3 to act as the centre of the clusters. While 150  (x < y) the agents that are the cluster centres will search in an increasing radius for other agents without a personality type to convert to P3 agents.   become-info-source  Landscape patches that run this procedure multiple the learning variable of agents in a radius of 5 patches around them by a rate that deteriorates with distance from the source following the equation: (1 + (2 / ([distance from source] + 1)))  Agents within a 5 patch radius then decrease their ENTITLEMENT score by 5% and increase their FORESIGHT and NEP scores by 5%.  

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

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"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.24.1-0300421/manifest

Comment

Related Items