Recent crises, such as the COVID-19 pandemic and western-US wildfires, underscore complex and non-obvious interrelationships between human actions and impacts. In the context of complex spatially networked environments, understanding human actions and their spatial-temporal impacts is especially urgent to develop policy to mitigate harms caused by global climate change.
Agent-based models (ABMs) are powerful tools for modeling complex human dynamics at an individual level. Spatial agent-based models can be scaled up to handle large study areas, massive agent numbers, and complex agent-environment interactions needed to address real world challenges.
However, addressing both large-scale spatial contexts and individual-level population heterogeneity is computationally intensive and requires high-performance computing (HPC) resources. As HPC resources are both expensive and can be challenging to access, work towards modeling that is both reproducible and approachable to interdisciplinary researchers is critical to increase participation towards more robust modeling outcomes.
Leveraging the power of cyberGIS, advanced cyberinfrastructure combined with geospatial computational capabilities, the goal of this dissertation is to create a robust and accessible environment for research on spatially explicit ABMs. This dissertation aims to realize this goal by 1) assessing the importance of spatial parameters in evacuating modeling, 2) creating a scalable and accessible software framework for spatial network-based evacuation modeling, and 3) demonstrating adaptive load balancing for optimizing HPC resource utilization to enable scalable modeling of emergency evacuation.
Contributions from this dissertation include the following: Varying evacuee route decisions in simulated evacuations lead to differences in evacuation clearance timing and spatial temporal locations of traffic congestion can occur from, which supports the need for better data on evacuee route choice. A major contribution from this dissertation is CyberGIS-ABM, a software framework to provide integration between ABMs, spatial networks, and parallel computing.
This dissertation also demonstrates how CyberGIS-Compute and science gateways can be leveraged to run CyberGIS-ABM on HPC resources allowing access to the simulation with fewer technological barriers. Furthermore, this dissertation builds on prior evacuation specific partitioning approaches to implement dynamic load balancing to support effective resource use in cases of spatially autocorrelated damage to evacuation routes. This research will work to make large scale ABMs of heterogeneous agents along spatial networks accessible to interdisciplinary researchers and support effective computing for emergency evacuation models.