Characterizing flash drought patterns in Eastern China based on high-resolution and long-term soil moisture fusion data

Abstract

Influenced by global environmental change, the increasing flash droughts have raised the demand for reliable root zone soil moisture (SM) data for drought monitoring. However, existing model-based products exhibit uncertainties and limitations in accuracy and spatial resolution, in-situ or satellite datasets fail to provide long-term records for comprehensive flash drought analysis. This study has developed a high-resolution (0.0625°) Soil Moisture Integrated Fusion (SMIF) dataset spanning from 1979 to 2022 for flash drought analysis in Eastern China. The fusion framework is constructed by using an advanced LightGBM machine learning algorithm, integrating multi-source SM data including observations from 1321 in-situ stations, high-resolution and long-term land model simulation, and reanalyzed and satellite retrievals, as well as long-term auxiliary meteorological variables. Independent validation reveals a 25% to 88% improvement in KGE for temporal prediction accuracy compared to existing SM products at both site and regional scales, such as ERA5 and GLDASv2.1 reanalysis data. As a result, SMIF performs well in capturing spatiotemporal evolution of flash droughts, including Yangtze River mega flash drought events in 2013 and 2022. Compared with the 44-year SMIF dataset, ERA5 overestimates flash drought frequency by 48% in Sichuan Basin and 31% in Yangtze and Huai River basins, while GLDASv2.1 underestimates flash drought frequency by 49% in northern regions. The newly generated SMIF dataset effectively reduces the limitations of current root zone SM products in terms of accuracy, spatial resolution, and temporal coverage for flash drought characterization, which is useful for application in agriculture, water resource management, and environmental sustainability.