Spatiotemporal Patterns of Drought and Flood Abrupt Alternation and Their Driving Factors in China from 1951 to 2020

Abstract Climate change has intensified variations in the terrestrial water cycle, increasing the occurrence of extreme events such as droughts, floods, and compound events. Using long-/short-cycle drought–flood abrupt alternation (DFAA) indices, combined with the random forest regression and the Shapley additive explanations method, this study aims to explore the spatiotemporal variation and driving mechanisms of DFAA indices, i.e., drought-to-flood (DTF) and flood-to-drought (FTD) events, in the 218 tertiary river basins of China from 1951 to 2020. Results show that regions with significant decreases in the long-cycle DFAA index are mainly in northern China, where FTD events tend to occur, with their gravity center shifting northward by approximately 328 km. Conversely, regions with an increasing DFAA index are mostly in southern China, where DTF events are prone to happen, with their gravity center shifting southward by approximately 325 km. A similar migration pattern is observed for short-cycle DFAA events: FTD hotspots move northward in May–June, and DTF hotspots move southward in June–July. Relative humidity and downward shortwave solar radiation are the main drivers of DFAA index variation: Higher relative humidity or weaker downward shortwave solar radiation contributes more positively to the DFAA index, favoring the occurrence of DTF events. In addition, the influence of teleconnection indices on the DFAA index is generally weaker than that of local variables, which mainly affect local anomalies by modulating large-scale atmospheric circulation. These findings help to understand the spatiotemporal characteristics of DFAA events in China and their underlying causes, providing valuable insights for decision-makers to formulate response policies. Significance Statement Based on a 70-yr analysis across China’s river basins, this study reveals that regions prone to shifts from drought to flood have moved southward, while flood-to-drought hotspots have migrated northward. Using a machine learning framework, we demonstrate that these compound extreme events are primarily driven by meteorological factors, particularly relative humidity, and are further intensified by large-scale climate patterns. These spatiotemporal shifts pose severe threats to water security, agriculture, and ecosystems. Our findings provide a crucial scientific basis for developing targeted early warning systems and adaptive watershed management strategies, thereby helping society better anticipate and mitigate the cascading impacts of these complex hydrological extremes in a changing climate.