Answers are grounded in this paper and cite their sources.
Study Region: The Beimiaoji catchment, a flood-prone tributary catchment of the upper Huai River Basin, China, was examined using daily runoff and gauge precipitation (2015–2024), ERA5-Land temperature and potential evapotranspiration, and catchment-context rasters. Study Focus: This study evaluated an HBV-guided encoder–decoder LSTM (HG-ED-LSTM) for daily runoff estimation under observed hydroclimatic forcing. The framework integrates antecedent hydroclimatic information, separately encoded HBV runoff sequences, and catchment-context features through gated fusion to correct the target-day HBV estimate. Performance was assessed using hv-blocked five-fold validation, benchmark comparisons, ablation experiments, and high-flow diagnostics covering nine representative and all 35 eligible peak-centered windows. New Hydrological Insights for the Region: In the Beimiaoji catchment, HBV provided a process-based runoff baseline but exhibited a negative aggregate bias (PBIAS = −23.79%). Incorporating HBV-guided residual correction improved NSE from 0.721 to 0.811, reduced RMSE by 17.9%, and brought PBIAS to −0.22%. Across 35 high-flow windows, pooled RMSE decreased by 33.6%, while mean peak-magnitude error declined by 50.8%. The pooled signed event-volume bias also improved from −28.98% to −3.43%, indicating a closer agreement with observed high-flow runoff volumes. Peak-timing performance varied between the representative and expanded event sets, highlighting the complementary value of magnitude- and timing-based evaluations. Overall, these findings demonstrate the potential of HBV-guided residual learning to improve daily runoff estimation and high-flow representation in the upper Huai River Basin under observed forcing.
No takes yet. Share an insight, caveat, or question.
Qiang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: