Agent-based models to couple natural and human systems for watershed management analysis

Coupled Human and Natural System;Agent-based Model;Multi-agent System;Parallel Computing;Web Application;Hadoop;Socio-hydrological Model;Global Sensitivity Analysis;Polynomial Chaos Expansion;Behavioral Uncertainty;Bounded Rationality;Probabilistic Graphical Model;Directed Information Graph;Boosted Regression Tree

Hu, Yao

关键词: Coupled Human and Natural System;    Agent-based Model;    Multi-agent System;    Parallel Computing;    Web Application;    Hadoop;    Socio-hydrological Model;    Global Sensitivity Analysis;    Polynomial Chaos Expansion;    Behavioral Uncertainty;    Bounded Rationality;    Probabilistic Graphical Model;    Directed Information Graph;    Boosted Regression Tree;   

Others  :  https://www.ideals.illinois.edu/bitstream/handle/2142/90865/HU-DISSERTATION-2016.pdf?sequence=1&isAllowed=y

美国|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

This dissertation expands conventional physically-based environmental models with human factors for watershed management analysis. Using an agent-based modeling framework, two approaches, one based on optimization and the other on data mining-are applied to modeling farmers' pumping decision-making processes in the High Plains aquifer within the hydrological observatory area. The resulting agent-based models (ABMs) are coupled with a physically-based groundwater model to investigate the interactions between farmers and the underlying groundwater system. With the optimization-based approach, the computational intensity arises from the execution of the resulting coupled ABM and groundwater model. This dissertation develops a computational framework that utilizes multithreaded programming and Hadoop-based cloud computing to address the computational issues. The framework allows multiple users to access and execute the web-based application of the coupled models simultaneously without an increase in latency via computer network. In addition, another computational framework to combine Hadoop-based Cloud Computing techniques with Polynomial Chaos Expansion (PCE) based variance decomposition approach is developed to conduct global sensitivity analysis with the coupled models, and influential behavioral parameters which are used to simulate agents’ behavior are identified. Being different from the optimization-based approach, which assumes all agents are rational, the data-driven approach attempts to account for the influences of agents’ bounded rationality on their behavior. A directed information graph (DIG) algorithm is used to exploit the causal relationships between agents’ decisions (i.e., groundwater irrigation depth) and time-series of environmental, socio-economical and institutional variables, and a machine learning technique, boosted regression tree (BRT) is applied to converting these causal relationships to agents’ behavioral rules. It is found that, in comparison with the optimization-based approach, crop profits and water tables as the result of agents’ pumping behavior derived using the data-driven approach can better mimic the actual observations. Thus, we can conclude that the data-driven approach using DIG and BRT outperforms the optimization-based approach when capturing agents’ pumping behavioral uncertainty as the result of bounded rationality, and for simulating real-world behaviors of agents.