Modeling the dynamics of drought resilience in South-Central United States using a Bayesian Network

Abstract

Droughts impact both the natural and human communities worldwide. Understanding the complex process underlying community resilience to drought is crucial to coping with the hazard. This study examined the dynamics of resilience to drought hazard for 503 counties in South-Central USA using a Bayesian Network (BN) approach. We first applied the Resilience Inference Measurement (RIM) framework and found 10 out of 52 variables contributed most to the resilience level of the county. We then applied the bootstrapped Hill-Climbing algorithm to Bayesian Network learning to identify the significant links among two resilience outcomes (population change, agricultural damage) and resilience predictors. The final BN, which included eight predictors, was used to find the probabilities of population decline and agricultural damage conditional upon different levels of hazard intensity (drought incidence) and two levels of resilience predictors (at Years 2000 and 2015). The study reveals that an increase in drought incidence will likely lead to higher agricultural damage, but it will unlikely curb population growth. This is probably due to the higher impact of drought hazard on agriculture than the general population. These probabilities and associated findings could be used as decision-support tools for stakeholders and practitioners in the communities affected by drought.