Characterising the high spatial heterogeneity of urban flood resilience under extreme rainfall using mobile signal data

Characterisation of the spatial heterogeneity of urban flood resilience (UFR) under extreme rainfall events can support decision-making in urban planning and emergency management. Traditional flood-resilience assessment methods based on socioeconomic or infrastructure statistics, face challenges in capturing urban operational efficiency (UOE; e.g. traffic mobility), especially its loss and recovery time. To overcome such challenges, we propose a new framework of UFR heterogeneity that: (1) utilises 800-m spatial and day-time resolution mobile signal big data as a dynamic metric for UOE; (2) presents a resilience matrix to quantitatively measure both the loss and recovery of UOE, achieving a unified UFR representation during extreme events; and (3) offers a spatial attention-based machine learning method to attribute the heterogeneity of UFR. An empirical study of the July 2021 extreme rainfall event in Zhengzhou, China, reveals that robust/high resilience, medium resilience, and low resilience grid cells account for 25%, 67%, and 8% of the samples; among them, the administrative core mainly comprises robust and high-resilience categories, whereas the county boundary exhibits low-resilience. Infrastructural transportation characteristics (28%), social-economic factors (25%), and spatial attributes (21%) are tracked as critical drivers by the XGBoost-SHAP framework.