Integrating Global Sensitivity Approaches to Deconstruct Spatial and Temporal Sensitivities of Complex Spatial Agent-Based Models

Abstract

Spatial agent-based models (ABMs) can be powerful tools for understanding individual level decision-making. However, in an attempt to represent realistic decision-making processes, spatial ABMs often become extremely complex, making it difficult to identify and quantify sources of model sensitivity. This paper implements a coastal version of the economic agent-based urban growth model, CHALMS, to investigate both space- and time-varying sensitivities of simulated coastal development dynamics. We review the current state of spatially- and temporally-explicit global sensitivity analyses (GSA) for environmental modeling in general, and build on the innovative but nascent efforts to implement these approaches with complex spatial ABMs. Combined variance- and density-based approaches to GSA were used to investigate the partitioning, magnitude, and directionality of model output variance. Time-varying GSA revealed sensitivity of multiple outputs to storm frequency and cyclical patterns of sensitivity for other input parameters. Spatially-explicit GSA showed diverging sensitivities at landscape versus (smaller-scale) zonal levels, reflecting trade-offs in residential housing consumer location decisions and spatial ‘spill-over’ interactions. More broadly, when transitioning from a conceptual to empirically parameterized model, sensitivity analysis is a helpful step to prioritize parameters for data collection, particularly when data collection is costly. These findings illustrate unique challenges of and need to perform comprehensive sensitivity analysis with dynamic, spatial ABMs.