Over the past decade Zhengzhou has been heavily affected by road collapse disasters, and the situation worsened after the extreme rainstorm on July 20, 2021. This context calls for risk assessment methods that delineate disaster hotspots and support municipal authorities in shifting from postevent investigation and handling to pre-event prevention and mitigation. Existing indicator weighting and machine learning approaches often have weak interpretability and limited transferability across cities. Traditional spatiotemporal kernel density estimation with fixed bandwidth (TSTKDE) is a de facto standard that captures temporal dynamics, but fixed bandwidths give unreliable estimates for inhomogeneous data, oversmoothing dense regions and undersmoothing sparse ones. To better analyze a decade of road collapse records in Zhengzhou, we develop an adaptive STKDE (ASTKDE) framework and evaluate its performance from multiple perspectives. Compared with TSTKDE, ASTKDE yields more reliable estimates across the entire domain, including the high-density upper tail, responds more sharply to the extreme event shock, produces more focused hotspot patterns, and improves short-horizon prediction of unseen events. In Zhengzhou, the old city district acts as a critical boundary, with disaster hotspots clustering within it and forming two major clusters and long-term hotspots aligning closely with the historic built-up area, indicating that early developed districts with aging pipelines are more collapse-prone, while problematic soils show weaker association at the city scale. Hotspots intensify as the flood season approaches, and the extreme rainstorm marks a watershed after which hotspot extent does not return to the pre-event state but expands beyond the old city district. These results suggest clear priorities for practice, provide reference for other cities facing similar challenges, and show that ASTKDE can be applied to other event types.
Jiao et al. (Sun,) studied this question.