Unveiling multifaceted resilience: A heterogeneous graph neural network approach for analyzing locale recovery patterns

Abstract

Resilience, denoting the capacity to swiftly recover to a state of normalcy subsequent to the occurrence of a disaster, constitutes a multifaceted phenomenon necessitating in-depth investigation. This study undertakes the quantification of resilience pertaining to specific locales through the utilization of heterogeneous data encompassing visitation patterns, demographic particulars, and points of interest (POI). A heterogeneous graph neural network is applied to model the resilience of these locales in Galveston, TX, USA. Our model outperforms regression models and other homogeneous baseline methodologies. Subsequent analysis unveils discernible resilience patterns intertwined with metrics such as visitation frequencies, visitors’ travel behaviors, and geographical attributes. In comparison to resilience investigations solely predicated upon visitation counts, our approach captures a more extensive array of information, thereby yielding a comprehensive understanding of the locale’s resilience.