Integrating machine learning with the Minimum Cumulative Resistance Model to assess the impact of urban land use on road waterlogging risk

Abstract

Urban waterlogging risk frequently manifests primarily on roadways, owing to their low topographical elevation and high impermeability. Accurately assessing the influence of surrounding land use on this risk is crucial for developing effective strategies. This study integrates machine learning models with the Minimum Cumulative Resistance Model to assess the impact of urban land use on road waterlogging risk, quantifying the interactions and diffusion patterns among various land units. The results indicate that: 1) The random forest classifier effectively identified 97 % of the test waterlogging points as low-resistance-cost areas. The primary factors influencing road waterlogging include the distance from the road (0.38), the stormwater drainage capacity (0.16), and vegetation coverage (0.12). 2) The diffusion resistance of waterlogging risk has been categorized into 10 levels. The resistance values for the highest risk level range from −263 to −17, which accounts for approximately 9.6 % of the study area. 3) The regions with high-risk concentration consist of six main sections, with minimum cumulative resistance differences ranging from −263 to 1072. These high-risk areas exhibit a gradual concentration towards the northeast. 4) A total of 456 potential transfer paths characterized by high waterlogging risk were identified, with lengths varying from 6 to 641 m, and their intersections with roads were delineated. The methodologies developed in this study facilitate a more precise evaluation of the effects of urban lands on road waterlogging risk, elucidating the mechanisms of risk propagation and yielding significant insights for the enhancement of management practices and mitigation strategies.