Analysis of flash droughts in China using machine learning

[摘要] The term “flash drought” describes a type of droughtwith rapid onset and strong intensity, which is co-affected by bothwater-limited and energy-limited conditions. It has aroused widespreadattention in related research communities due to its devastating impacts onagricultural production and natural systems. Based on a global reanalysisdataset, we identify flash droughts across China during 1979–2016 by focusing on the depletion rate of weekly soil moisture percentile.The relationship between the rate of intensification (RI) and nine relatedclimate variables is constructed using three machine learning (ML)technologies, namely, multiple linear regression (MLR), long short-termmemory (LSTM), and random forest (RF) models. On this basis, thecapabilities of these algorithms in estimating RI and detecting droughts (flashdroughts and traditional slowly evolving droughts) were analyzed.Results showed that the RF model achieved the highest skill in terms of RIestimation and flash drought identification among the three approaches.Spatially, the RF-based RI performed best in southeastern China, with anaverage CC of 0.90 and average RMSE of the 2.6 percentile per week, while poorperformances were found in the Xinjiang region. For drought detection, allthree ML technologies presented a better performance in monitoring flashdroughts than in conventional slowly evolving droughts. Particularly, theprobability of detection (POD), false alarm ratio (FAR), and criticalsuccess index (CSI) of flash drought derived from RF were 0.93, 0.15, and0.80, respectively, indicating that RF technology is preferable in estimatingthe RI and monitoring flash droughts by considering multiple meteorologicalvariable anomalies in adjacent weeks to drought onset. In terms of themeteorological driving mechanism of flash drought, the negativeprecipitation ( P ) anomalies and positive potential evapotranspiration (PET)anomalies exhibited a stronger synergistic effect on flash droughts comparedto slowly developing droughts, along with asymmetrical compound influencesin different regions of China. For the Xinjiang region, P deficit played adominant role in triggering the onset of flash droughts, while insouthwestern China, the lack of precipitation and enhanced evaporativedemand almost contributed equally to the occurrence of flash drought. Thisstudy is valuable to enhance the understanding of flash droughts andhighlight the potential of ML technologies in flash drought monitoring.

[学科分类] 妇产科学