Evaluating Water Demand Using Agent-Based Modeling

The supply and demand of water resources are functions of complex, inter-related systems including hydrology, climate, demographics, economics, and policy. To assess the safety and sustainability of water resources, planners often rely on complex numerical models that relate some or all of these systems using mathematical abstractions. The accuracy of these models relies on how well the abstractions capture the true nature of the systems interactions. Typically, these abstractions are based on analyses of observations and/or experiments that account only for the statistical mean behavior of each system. This limits the approach in two important ways: 1) It cannot capture cross-system disruptive events, such as major drought, significant policy change, or terrorist attack, and 2) it cannot resolve sub-system level responses. To overcome these limitations, we are developing an agent-based water resources model that includes the systems of hydrology, climate, demographics, economics, and policy, to examine water demand during normal and extraordinary conditions. Agent-based modeling (ABM) develops functional relationships between systems by modeling the interaction between individuals (agents), who behave according to a probabilistic set of rules. ABM is a "bottom-up" modeling approach in that it defines macro-system behavior by modeling the micro-behavior of individual agents. While each agent's behavior is often simple and predictable, the aggregate behavior of all agents in each system can be complex, unpredictable, and different than behaviors observed in mean-behavior models. Furthermore, the ABM approach creates a virtual laboratory where the effects of policy changes and/or extraordinary events can be simulated. Our model, which is based on the demographics and hydrology of the Middle Rio Grande Basin in the state of New Mexico, includes agent groups of residential, agricultural, and industrial users. Each agent within each group determines its water usage based on its own condition and the condition of the world around it. For example, residential agents can make decisions to convert to or from xeriscaping and/or low-flow appliances based on policy implementation, economic status, weather, and climatic conditions. Agricultural agents may vary their usage by making decisions on crop distribution and irrigation design. Preliminary results show that water usage can be highly irrational under certain conditions. Results also identify sub-sectors within each group that have the highest influence on ensemble group behavior, providing a means for policy makers to target their efforts. Finally, the model is able to predict the impact of low-probability, high-impact events such as catastrophic denial of service due to natural and/or man-made events.