A0298
Title: Unbalanced regularized optimal transport to model natural disaster
Authors: Angelo Fama - University of Insubria (Italy) [presenting]
Paolo Pagnottoni - University of Insubria (Italy)
Abstract: Natural disasters have far-reaching consequences that vary across time and geography. A novel approach is presented to analyze the spatiotemporal distribution of disaster impacts using unbalanced regularized optimal transport. A framework is developed that models the distribution of quarterly mortality rates, spanning from 2000-01 to 2024-12, over a discretized geographic grid, accounting for imbalanced datasets where the total mass (e.g., deaths) changes across time intervals. The proposed method uses unbalanced optimal transport to handle shifts in population and disaster severity, while entropy regularization ensures computational efficiency and robustness to noise. Quarterly mortality distributions caused by climate-related disasters across a global grid are analyzed. Pairwise transport plans are computed to quantify spatial shifts in disaster impact, while Euclidean distances are used to evaluate structural similarities over time. Results reveal significant patterns of distributions across quarters, highlighting regions with increased vulnerability and potential spatiotemporal structures behind them.
