Abstract:  Urban road collapse is a serious safety issue in urban areas that can result in casualties and transportation disruptions, threatening urban operations and social development. Accurately predicting urban road collapse and analyzing its spatial and temporal dynamics are of great significance for urban safety. In this study, we focused on Futian District in Shenzhen City, Guangdong Province as the study area. We developed a model to predict the susceptibility of urban road collapse by integrating multi-source remote sensing data with the random forest algorithm. Additionally, we examined the primary indicators affecting the prediction performance of the model and the primary factors influencing the vulnerability of urban roads to collapse. The spatiotemporal prediction results of urban road collapse susceptibility show that the urban road collapse susceptibility prediction model, constructed by combining optical data and radar data, can predict the spatial and temporal changes of road collapse susceptibility  accurately. The predicted coefficient of determination is 0.65, indicating high prediction accuracy. From 2017 to 2022, the risk of roadway collapse in Futian District exhibited an overall increasing trend. There was a decrease in the area of zones with very low and low susceptibility, and an increase in the area of zones with medium and high susceptibility. The results of the random forest feature importance analysis indicate that the texture features extracted from image data make a significant contribution to the prediction model. Based on the geodetector results, it is known that population, GDP, and underground facilities are three key factors influencing urban roadway collapse.

Key words:  , road collapse, random forest, multi-source remote sensing data, spatial and temporal changes, Futian District, Shenzhen City, Guangdong Province