Optimising hurricane shelter locations with smart predict-then-optimise framework

Hurricanes pose an escalating threat to global communities, underscoring the urgent need for robust disaster response strategies. A pivotal component of these strategies involves the establishment of secure shelters. However, the inherent vulnerability of these shelters to hurricane damage frequently undermines their utility. This study introduces a Predict-then-Optimise (PTO) framework designed to support relief agencies in selecting optimal locations for emergency shelters, with an emphasis on minimising potential damage during hurricanes. Employing a two-phase approach, the framework initially predicts potential hurricane-induced damage losses, subsequently utilising these predictions to optimise shelter placement strategies. Nevertheless, conventional PTO methods in shelter planning may lead to suboptimal decisions, primarily because of potential discrepancies between predicted and actual damage losses, given the inherent uncertainties and complexities of hurricane impacts. To address these limitations, our study introduces an advanced smart Predict-then-Optimise (SPO) framework. This SPO framework more cohensively integrates the prediction and optimisation phases, thereby facilitating an adaptive and resilient response to the dynamic challenges posed by hurricanes. We demonstrate the effectiveness of this methodology through a case study in Miami-Dade County, Florida, where the SPO framework successfully identified optimal shelter locations, significantly reducing exposure to high-risk areas.