Determinants of flash drought recovery rates: the role of precipitation patterns and surface heterogeneity

Predicting the onset of flash droughts remains a formidable challenge for current drought early warning systems. However, understanding the timing and mechanisms of flash drought recovery is equally critical for effective flash drought risk management. Despite this, limited research has focused on the recovery phase across different stages of the flash drought full life cycle. This study integrates multi-source data to investigate the spatiotemporal dynamics, key drivers, and underlying mechanisms of flash drought recovery processes across nine agroclimatic regions in China from 1980 to 2020. Integrating Geodetector and interpretable machine learning (XGBoost-SHAP), we quantified the relative contributions of precipitation patterns and land surface characteristics to flash drought recovery rates (FDRR). Additionally, a composite analysis was conducted to clarify the role of extreme precipitation events and assess the risk of cascading dry-to-wet transitions associated with flash drought recovery. Results indicate that after the year 2000, flash drought events with both high onset and high recovery rates have significantly increased in China’s monsoon region. Extreme precipitation events drive flash drought recovery in only 0–20 % of cases across most regions, except in the central and western Qinghai-Tibet Plateau, where the probability reaches 40–60 %, highlighting a high-risk zone for cascading dry-to-wet transitions. Encouragingly, XGBoost-based simulations of FDRR in this region achieved high accuracy (R² = 0.912). SHAP-based interpretation and interaction detection confirmed that precipitation characteristics and land surface conditions are the primary determinants of FDRR. Key factors include cumulative precipitation, maximum precipitation, drought severity, terrain slope, and soil clay content. In China’s flash drought hotspot, the Middle-Lower Yangtze Plain, the primary moisture source during flash drought recovery predominantly stems from large-scale circulation systems rather than local evapotranspiration or moisture recycling. Our findings highlight the feasibility of developing flash drought recovery forecasting models, which could improve risk management strategies for meteorological and agricultural agencies. If predicting flash drought onset remains challenging, accurately forecasting recovery may be crucial in mitigating flash drought risks.