Flash drought early warning by a two-stage coupled framework integrating Causal-ConvLSTM in the Han River basin

Flash droughts, as rapidly evolving extreme hydrological events, pose severe threats to agricultural production and water security. Reliable early warning of flash droughts is thus critical for proactive agricultural drought mitigation and optimal irrigation scheduling. However, existing prediction methods suffer from two major limitations: (1) rapid error propagation in long-series soil moisture forecasting, which undermines prediction accuracy and physical interpretability; and (2) failure to account for the strong seasonal heterogeneity of flash droughts, with fixed discriminant thresholds unable to balance missed detection and false alarms. To address these gaps, this study develops a two-stage coupled framework for 15-day flash drought early warning in the Han River Basin. First, a Causal-Convolutional Long Short-Term Memory (Causal-ConvLSTM) model integrated with causal priors is established to forecast daily root-zone soil moisture, driven by season-specific factors identified via the Boosted Regression Tree (BRT) model. Second, the three-pentad (i.e., 15-day) cumulative soil moisture deficit is innovatively extracted as the core discriminant feature, and a Light Gradient Boosting Machine (LightGBM) classifier is constructed to predict flash drought occurrence probability. Finally, the Youden index is used to determine seasonal optimal discriminant thresholds, adapting to seasonal differentiation. Results show that: (1) the proposed framework achieves an Area Under the Curve (AUC) of 0.788 and an overall accuracy of 0.715 on the held-out test period (2015–2022), demonstrating competitive performance; the embedded Causal-ConvLSTM model attains an R² of 0.89 at 1–5 day lead times, decreasing to 0.71 at 6–10 days and 0.57 at 11–15 days (1–15 day mean: 0.73), indicating that error propagation is suppressed relative to the benchmark models but that forecast skill still decays with increasing lead time; (2) the framework successfully identified the severe summer 2022 flash drought event 15 days in advance, robustly capturing its spatiotemporal evolution; and (3) seasonal performance is consistently reliable across all seasons, with spring and summer showing the highest skill. This study provides a reliable and operational tool for flash drought early warning, enabling farmers and water managers to schedule irrigation proactively and refine water resources allocation in the Han River Basin and similar semi-humid regions.