Analysis of flash droughts in China using machine learning

Abstract. The term “flash drought” describes a type of drought with rapid onset and strong intensity, which is co-affected by both water-limited energy-limited conditions. It has aroused widespread attention in related research communities due to its devastating impacts on agricultural production natural systems. Based global reanalysis dataset, we identify flash droughts across China during 1979–2016 focusing the depletion rate weekly soil moisture percentile. relationship between intensification (RI) nine climate variables constructed using three machine learning (ML) technologies, namely, multiple linear regression (MLR), long short-term memory (LSTM), random forest (RF) models. On this basis, capabilities these algorithms estimating RI detecting (flash traditional slowly evolving droughts) were analyzed. Results showed that RF model achieved highest skill terms estimation identification among approaches. Spatially, RF-based performed best southeastern China, an average CC 0.90 RMSE 2.6 percentile per week, while poor performances found Xinjiang region. For detection, all ML technologies presented better performance monitoring than conventional droughts. Particularly, probability detection (POD), false alarm ratio (FAR), critical success index (CSI) derived from 0.93, 0.15, 0.80, respectively, indicating technology preferable considering meteorological variable anomalies adjacent weeks onset. In driving mechanism drought, negative precipitation (P) positive potential evapotranspiration (PET) exhibited stronger synergistic effect compared developing droughts, along asymmetrical compound influences different regions China. region, P deficit played dominant role triggering southwestern lack enhanced evaporative demand almost contributed equally occurrence drought. This study valuable enhance understanding highlight monitoring.