Analysis of human flow during a natural disaster utilizing trajectory-free mobile network data: a case study of...

Understanding population movement during natural disasters is critical for emergency response, resource allocation, and risk mitigation. This study proposes a kernel-based flow extraction framework to infer directional human mobility from aggregated, trajectory-free mobile network data. By integrating kernel density estimation (KDE) and a modified gravity model, the method generates spatiotemporal vector fields that capture flow dynamics without requiring individual tracking. The framework is applied to a case study of the 2025 M_L 6.4 (M_wg 5.8) Dapu earthquake in southern Taiwan, using mobile communication data aggregated in 500 m × 500 m grids at 10-minute intervals. The analysis reveals significant mobility changes following the seismic event, with directional shifts transitioning from outward dispersal to inward convergence, especially in densely populated zones. Temporal peaks in anomalous flow patterns were detected within the first hour post-event, with variations across seismic intensity levels and population densities. These findings highlight the spatial heterogeneity of behavioral responses to disasters and underscore the utility of the proposed approach for identifying critical zones of disruption. The framework offers a scalable, privacy-preserving tool for real-time disaster monitoring and supports the development of early warning systems and urban resilience planning.