Characterizing public response to unforeseen cascading fuel shortage: Through the lens of human mobility-based explainable machine learning models

Abstract

Climate disasters unfold multitudes of effects, from societal and commercial disruptions to fuel and power shortages. These consequences escalate further in cascading disasters, where individuals are more likely to respond unwarrantedly due to the lack of preparation and situational awareness. A noticeable gap exists in comprehending the linkages between public responses to such disasters and socioeconomic and spatial disparities, which are critical to the provision of effective guidance and situational information to those affected. Based on mobile phone data and various socioeconomic, built environment, and geographical variables, this study systematically examines human mobility-based public responses during a cascading fuel shortage crisis. The spatiotemporal analysis uncovered a significant increase in visits to gasoline stations during and after the crisis and a decrease in mean distance traveled at the Census Block Group level. Furthermore, mobility prediction models were constructed using the random forest regression algorithm, which can adequately forecast visits and mean distance traveled to gasoline stations across different crisis stages. The Shapley Additive Explanations analysis reveals how various factors (e.g., educational attainment and distance to the coast) influenced public responses. These findings reinforce the importance of tailored disaster response education and situational awareness to ensure equitable resource access during cascading disasters.