Urban-Rural Disparity in Disaster Resilience: Harnessing Nighttime Light Data and Interpretable Machine Learning

Nighttime Light (NTL) data, offering insights into cross-regional human activities and infrastructure changes, has gained widespread use in disaster monitoring. This study explores the application of NASA's Black Marble daily data in monitoring post-typhoon responses. In a landscape characterised by a high mix of urban built-up areas and peri-urban villages, we investigate differences in nighttime light recovery across various land-use types after disasters. Combining interpretable machine learning, we explore the reasons behind these disparities by comparing Shapley values and specific Accumulated Local Effects (ALE) between regions, evaluating high importance of individual predictive factors and identifying potential non-linear patterns and threshold effects.
  Our findings reveal more instances of sustained nighttime light decline in rural areas (residential and agricultural land), while urban areas exhibit increased nighttime light during disasters. These differences primarily relate to infrastructure features, especially roads. Meteorological factors, such as precipitation probability and wind speed, impact NTL predictions in urban and rural areas. Post-disaster relief activities significantly influence NTL changes in rural settlements. Additionally, the occurrence of extreme weather increases the likelihood of cascading disasters. Our study finds that disaster impact zones in coastal areas extend deeper into the mainland, posing threats to adjacent mountainous regions and elevating the risk of secondary disasters like landslides.
  In conclusion, this study provides a regional assessment of resilience differences and influencing mechanisms using nighttime light data. It offers valuable information for policymakers to identify key factors influencing typhoon disaster resilience, enabling them to mitigate systemic risks and enhance overall system resilience. The significance of this research extends to serving as a valuable reference for data-driven recovery quantification from typhoon hazards and other crises.