It Matters “How”, Not Just What, Factors Are Included: a Case Study of a Migration Agent-Based Model

Abstract:

Studies of coupled natural-human systems (CNHSs) incorporate natural and social factors in diverse environmental issues. Existing CNHS literature has usually discussed what natural and social factors must be chosen for model development. The field of CNHS, however, has little knowledge of “how” these factors should be included in the models (factor configuration). In fact, factor configuration may be more influential to emergent behaviors of models than individual factors alone. Establishing a wrong factor configuration can even produce misprediction and inhibit successful management against unexpected disasters. From this point of view, we question: i) is the factor configuration critical to the emergent pattern in the CNHS models?; and ii) will the system respond differently in each factor configuration when disturbed by a disaster? This research particularly focused on the effect of factor configuration in the environmentally induced migration as a case study of CNHS problems. We developed a proof-of-concept agent-based model (ABM) with several factor configurations to answer our research questions. Two measures were used to capture transient migration patterns: spatial distribution of populations and the mixing of cultural groups. We analyzed what underlies distinct migration patterns by looking back into the theoretical foundations of varying factor configurations. Moreover, we experimented with a disaster scenario of reducing water availability in one region and compared different responses in each factor configuration. We observed that transient migration patterns varied depending on the factor configuration. Each factor configuration also exhibited different responses to the disturbance. Ultimately, we expand our ABM results in the context of CNHSs. Our findings imply that CNHS models must set up a factor configuration that is theoretically reasonable to predict correct transient patterns and maximize the value of the model.