Tang et al. (2026) Improved flash drought forecasting and attribution: A spatial-temporal causality-aware deep learning approach

Identification

• Journal: Journal of Hydrology
• Year: 2026
• Date: 2026-01-12
• Authors: Sijie Tang, Shuo Wang, Jiping Jiang, Yi Zheng
• DOI: 10.1016/j.jhydrol.2026.134945

Research Groups

• State Key Laboratory of Climate Resilience for Coastal Cities, Department of Land Surveying and Geo-Informatics, The Hong Kong Polytechnic University, Hong Kong, China
• Otto Poon Research Institute for Climate-Resilient Infrastructure and Research Institute for Land and Space, The Hong Kong Polytechnic University, Hong Kong, China
• School of Environmental Science and Engineering, Southern University of Science and Technology, Shenzhen 518055, China
• Shenzhen Municipal Engineering Lab of Environmental IoT Technologies, Southern University of Science and Technology, Shenzhen 518055, China
• State Environmental Protection Key Laboratory of Integrated Surface Water-Groundwater Pollution Control, Southern University of Science and Technology, Shenzhen 518055, China

Short Summary

This study introduces a novel deep learning framework, integrating a spatial-temporal causality-aware module into a CNN-LSTM architecture, to improve flash drought forecasting and attribution in China's Greater Bay Area. The framework significantly enhances prediction accuracy, particularly for flash drought onset, and reveals new insights into critical drought drivers, including the previously underrecognized role of downward longwave radiation.

Objective

• To develop and evaluate a novel deep learning framework that integrates a spatial–temporal causality-aware (STC) module into a CNN-LSTM hybrid architecture to enhance flash drought prediction in China’s Greater Bay Area (GBA).
• To improve the predictability of flash droughts and mitigate potential risks by providing insights into their drivers and mechanisms.

Study Configuration

• Spatial Scale: China's Greater Bay Area (GBA).
• Temporal Scale: Forecasting of rapid-onset flash drought events, with a focus on onset prediction.

Methodology and Data

• Models used: Deep learning framework integrating a spatial–temporal causality-aware (STC) module into a CNN-LSTM hybrid architecture.
• Data sources: Input data for the model include flash drought drivers such as soil moisture, downward longwave radiation, and precipitation.

Main Results

• The causality module enhanced model generalization (GA = 0.90) and performance (NSE = 0.83).
• The accuracy of flash drought onset prediction was substantially increased (F1 score = 0.33) compared to baseline models.
• Explainable AI analyses revealed that incorporating causality strengthened the predictive contributions of key flash drought drivers, including soil moisture memory, downward longwave radiation, and precipitation.
• Downward longwave radiation emerged as a critical, previously underrecognized predictor of soil moisture variability in humid subtropical climates.
• Distinct mechanisms were identified for slow and flash droughts: slow droughts are dominated by initial soil moisture and persistent shortwave radiation, while flash droughts are driven by rapid energy imbalances and longwave radiation.
• Anthropogenic activities in China’s GBA were found to intensify the complexity of drought mechanisms, increasing both prediction difficulty and regional vulnerability to hydrological extremes.

Contributions

• Introduction of a novel deep learning framework with a spatial-temporal causality-aware module for improved flash drought forecasting.
• Significant enhancement in flash drought onset prediction accuracy through the integration of causality.
• Identification of downward longwave radiation as a critical and previously underrecognized driver for soil moisture variability in humid subtropical regions.
• Elucidation of distinct mechanisms underlying slow versus flash droughts, providing deeper attribution insights.
• Highlighting the impact of anthropogenic activities on drought complexity and vulnerability in the Greater Bay Area.

Funding

[No funding information provided in the excerpt.]

Citation

@article{Tang2026Improved,
  author = {Tang, Sijie and Wang, Shuo and Jiang, Jiping and Zheng, Yi},
  title = {Improved flash drought forecasting and attribution: A spatial-temporal causality-aware deep learning approach},
  journal = {Journal of Hydrology},
  year = {2026},
  doi = {10.1016/j.jhydrol.2026.134945},
  url = {https://doi.org/10.1016/j.jhydrol.2026.134945}
}