Authors: Taylor Anderson*, Simon Fraser University, Suzana Dragicevic, Simon Fraser University
Topics: Geographic Information Science and Systems
Keywords: Geographic information systems, network science, agent-based modelling, model testing, model validation
Session Type: Paper
Start / End Time: 11:10 AM / 12:25 PM
Room: Grand Ballroom 1, Sheraton, IM Pei Tower, Second Floor Level
Presentation File: No File Uploaded
Agent-based models (ABM) are bottom-up approaches capable of representing a variety of processes that form complex spatio-temporal systems. In order to be useful for knowledge discovery and decision-making, ABMs need to be fully validated to determine how well they represent real-world processes and whether or not they are fit for their purpose. Typically, model validation procedures compare emergent spatial patterns with independent data sets. However, these methods are limited in their ability to determine whether the simulated interactions and processes that form the emergent patterns are represented correctly. This research study proposes a novel network-based approach for ABM validation. First, a network-based ABM (N-ABM) is developed by abstracting the system and the processes simulated by the ABM into sets of dynamic measurable spatial networks. Next, graph theory is applied in order to quantify the properties of the abstracted spatial networks across multiple scales using a set of network measures. Finally, the properties of the simulated networks are compared with the observed properties of the real networks and the degree to which the simulated network measures correspond with empirical network measures for the same system can be evaluated. The proposed network-based approach for ABM validation is general so that it can be used to validate a variety of developed ABMs and thus increase the confidence that both the simulated processes and emergent spatial patterns are realistic.