A Long-term Spatial Runoff and Flood Prediction Method in Higher Accuracy

When predicting future long-term runoff using hydrological models, the large uncertainty associated with general circulation models (GCMs) pose significant limitations. Additionally, current accurate long-term runoff predictions are restricted to specific locations with gauge stations, hindering basin-wide water resource planning and management. To address these challenges, this study proposes a hybrid Hydrological model, Empirical Orthogonal Function analysis, Gaussian Process Regression (HEG) model, which demonstrates higher accuracy in daily runoff prediction across the entire basin compared to the traditional multi-model ensemble mean method, with KGE improved by 0.09~0.11, and NSE improved by 0.08~0.32). Moreover, to enhance the estimation of future extreme flood risks which are of great concern of the public but are often predicted with high uncertainty, the model incorporates uncertainty interval information into prediction and is called HEGU model. Evaluations conducted in the topographically and climatically diverse Brahmaputra River Basin confirm the effectiveness of the HEGU model. The relative error of peak discharge (REPD) is reduced to an average of ~46% of that obtained through the ensemble mean method, while the correlation coefficient (CC) for flood volume estimation during the monsoon period increases from -0.054 to 0.645. Furthermore, the HEGU model demonstrates the potential to improve overall runoff prediction accuracy across the basin when the data quality of extremely few grids in the high-fidelity dataset is enhanced. The enhancement can be achieved through the incorporation of additional runoff gauge stations, remote sensing data, and other data augmentation techniques. These findings underscore the practical significance of the HEGU model, indicating its high effectiveness and applicability in real-world future hydrological projection and water resource management scenarios.