Machine Learning Maps Flash Drought Drivers Across the Yellow River Basin

Flash droughts, which develop and intensify far more rapidly than conventional slow-onset droughts, have become a growing concern for agriculture and natural ecosystems worldwide.

The research team, led by Ting Liu with colleagues including Wei Wei, Dang Lu and Ashraf Dewan, analyzed flash drought events in the basin from 1980 to 2023. Rather than relying on a single data source, they used three reanalysis datasets, ERA5, ERA5-Land and GLDAS, and applied the Standardized Evaporative Stress Ratio to detect and characterize flash drought episodes across more than four decades.

The results show that flash droughts identified across the datasets occur predominantly in the middle and upper reaches of the basin. The three datasets agree strongly on how often these events occur, how long they last on average, and where they are distributed spatially, but they diverge notably when it comes to drought intensity, a finding that underscores how sensitive flash drought characterization can be to the underlying data product.

During the growing season, which the study defines as April through September, a different pattern emerges. Flash droughts detected by ERA5 and ERA5-Land concentrate mainly in the middle and lower reaches of the basin, while GLDAS identifies a higher prevalence of flash drought in the upper reaches.

Seasonality also varies by dataset. The authors report these seasonal differences as part of a broader effort to establish where and when flash drought risk is concentrated in the basin.

Beyond mapping where flash droughts occur, the study tackles a harder question: what causes them. The researchers trained a machine learning model on meteorological data and applied both binary classification and regression analysis to probe the factors influencing flash drought occurrence and dynamics. This dual approach allowed them to separate the variables most associated with whether a flash drought happens from those most influential in shaping its behavior.

The binary classification results point to maximum temperature, minimum temperature and net shortwave radiation as the primary drivers of flash drought in the basin.

The analysis also reveals that the driving mechanisms differ across the basin’s three main reaches. The upper reaches are thermally driven, meaning temperature-related factors dominate. The middle reaches are shaped by the synergy of heat and water, while the lower reaches are governed by the combined influence of humidity and weather systems.

The regression analysis reinforces the central role of the surface energy balance. Evapotranspiration, net shortwave radiation and maximum temperature stand out as the most influential factors affecting flash drought dynamics, linking rapid drought intensification to the exchange of moisture and energy between the land surface and the atmosphere.

The authors present the work as a foundation for improving flash drought risk management in the Yellow River Basin, where rapid-onset water shortages can strike agriculture with little lead time. Because the study is based on reanalysis data and statistical modeling rather than direct forecasts, its findings describe historical patterns and associations between meteorological factors and flash drought occurrence. Even so, by showing that the drivers of flash drought shift systematically from thermal control in the upper reaches to humidity and weather-system control downstream, the research offers a spatially explicit basis for targeting monitoring efforts where they are most likely to pay off.

Subject of Research: Flash drought identification and driver analysis in the Yellow River Basin using reanalysis datasets and machine learning

Article Title: Flash drought identification and affecting factors interpretation based on reanalysis data and machine learning approach at river basin scale: a case study in the Yellow River Basin

Article References: Liu, T., Wei, W., Lu, D., Gao, F., Zhou, L., Zhou, J., Song, Y., Sherif, M., Wang, X., & Dewan, A. (2026). Flash drought identification and affecting factors interpretation based on reanalysis data and machine learning approach at river basin scale: a case study in the Yellow River Basin. Climate Dynamics, 64(11), Article 471. https://doi.org/10.1007/s00382-026-08416-5

Image Credits: AI Generated

DOI: 10.1007/s00382-026-08416-5

Keywords: flash drought, Yellow River Basin, reanalysis data, ERA5, GLDAS, machine learning, Standardized Evaporative Stress Ratio, evapotranspiration, net shortwave radiation, drought drivers, Climate Dynamics, Flash