Exploration of the spatiotemporal characteristics and triggering factors of flash flood in China

• Uncovering spatiotemporal patterns of flash floods. • Developing an optimized XGBoost-SHAP framework to quantify disaster-driver contributions. • Evaluating the impact and interactions of key factors on flash flood patterns. Flash floods are a major natural hazard, increasingly intensified by the growing frequency of extreme weather events, causing substantial casualties and economic losses. This study first compiled flash flood disaster data from 2017 to 2021 and analyzed its spatiotemporal distribution trends. Then, XGBoost was introduced to quantitatively identify key influencing factors of flash floods, with a global interpretation conducted using SHAP (Shapley Additive Explanations). On this basis, a three-dimensional trend analysis was conducted to examine the spatial heterogeneity of flash flood drivers, thereby elucidating the key factors that govern their spatial characteristics. The results indicate that: (1) Flash flood disasters in China have exhibited an upward trend, with the lowest occurrence in 2019. Flash floods are primarily concentrated in the flood season and are most prevalent in the southwestern region, exhibiting a high consistency with the spatial distribution of heavy rainfall. (2) The explainable machine learning achieves high accuracy, with meteorological and topographical factors serving as the primary drivers of flash flood occurrence.(3) Regional variations in flash flood occurrences are significantly modulated by key environmental factors, particularly annual precipitation, elevation, and slope. Specifically, flash floods are notably suppressed in low-precipitation, high-elevation regions, while moderate elevations and steep slopes significantly enhance their occurrence. These findings offer essential theoretical and technical references for flash flood disaster prevention and management.