Leveraging multi-source data and teleconnection indices for enhanced runoff prediction using coupled deep learning models DOI Creative Commons
Jintao Li, Ping Ai, Chuansheng Xiong

и другие.

Scientific Reports, Год журнала: 2025, Номер 15(1)

Опубликована: Апрель 27, 2025

Accurate medium- to long-term runoff forecasting is crucial for flood control, drought resilience, water resources development, and ecological improvement. Traditional statistical methods struggle utilize multifaceted variable information, leading lower prediction accuracy. This study introduces two innovative coupled models-SRA-SVR SRA-MLPR-to enhance by leveraging the strengths of deep learning approaches. Stepwise Regression Analysis (SRA) was employed effectively handle high-dimensional data multicollinearity, ensuring that only most influential predictive variables were retained. Support Vector (SVR) Multi-Layer Perceptron (MLPR) chosen due their strong adaptability in capturing nonlinear relationships extracting latent hydrological patterns. The integration these significantly improves accuracy model stability. By integrating 80 atmospheric circulation indices as teleconnection variables, models tackle critical challenges such data, dynamics. Yalong River Basin, characterized complex processes diverse climatic influences, serves case validation. results show that: (1) Compared baseline single models, SRA-MLPR reduced RMSE (from 798.47 594.45) 26% MAPE 34.79 22.90%) 34%, while achieving an NSE 0.67 0.76) improvement 13%, particularly excelling extreme scenarios. (2) inclusion not enriched feature set but also improved stability, with demonstrating enhanced capability relationships. (3) A one-month lag identified optimal predictor basin-scale runoff, providing actionable insights into temporal (4) To interpretability, SHAP (SHapley Additive exPlanations) analysis quantify contribution predictions, revealing dominant climate drivers interactions. indicate Northern Hemisphere Polar Vortex East Asian Trough exert significant control over dynamics, influence modulated large-scale oscillations North Atlantic Oscillation (NAO) Pacific Decadal (PDO). (5) models' scalability validated through modular design, allowing seamless adaptation contexts. Applications include forecasting, optimized reservoir operations, adaptive resource planning. Furthermore, demonstrates potential generalizable tools basins varying geographic conditions. highlights robust across indices, proposed stability offering valuable prevention, mitigation, management. These methodological advancements bridge gap between approaches, a scalable framework accurate interpretable hydrological, climatological, environmental predictions. Given escalating brought about change, findings this make contributions sustainable management, decision-making support, disaster preparedness at global level.

Язык: Английский

Leveraging multi-source data and teleconnection indices for enhanced runoff prediction using coupled deep learning models DOI Creative Commons
Jintao Li, Ping Ai, Chuansheng Xiong

и другие.

Scientific Reports, Год журнала: 2025, Номер 15(1)

Опубликована: Апрель 27, 2025

Accurate medium- to long-term runoff forecasting is crucial for flood control, drought resilience, water resources development, and ecological improvement. Traditional statistical methods struggle utilize multifaceted variable information, leading lower prediction accuracy. This study introduces two innovative coupled models-SRA-SVR SRA-MLPR-to enhance by leveraging the strengths of deep learning approaches. Stepwise Regression Analysis (SRA) was employed effectively handle high-dimensional data multicollinearity, ensuring that only most influential predictive variables were retained. Support Vector (SVR) Multi-Layer Perceptron (MLPR) chosen due their strong adaptability in capturing nonlinear relationships extracting latent hydrological patterns. The integration these significantly improves accuracy model stability. By integrating 80 atmospheric circulation indices as teleconnection variables, models tackle critical challenges such data, dynamics. Yalong River Basin, characterized complex processes diverse climatic influences, serves case validation. results show that: (1) Compared baseline single models, SRA-MLPR reduced RMSE (from 798.47 594.45) 26% MAPE 34.79 22.90%) 34%, while achieving an NSE 0.67 0.76) improvement 13%, particularly excelling extreme scenarios. (2) inclusion not enriched feature set but also improved stability, with demonstrating enhanced capability relationships. (3) A one-month lag identified optimal predictor basin-scale runoff, providing actionable insights into temporal (4) To interpretability, SHAP (SHapley Additive exPlanations) analysis quantify contribution predictions, revealing dominant climate drivers interactions. indicate Northern Hemisphere Polar Vortex East Asian Trough exert significant control over dynamics, influence modulated large-scale oscillations North Atlantic Oscillation (NAO) Pacific Decadal (PDO). (5) models' scalability validated through modular design, allowing seamless adaptation contexts. Applications include forecasting, optimized reservoir operations, adaptive resource planning. Furthermore, demonstrates potential generalizable tools basins varying geographic conditions. highlights robust across indices, proposed stability offering valuable prevention, mitigation, management. These methodological advancements bridge gap between approaches, a scalable framework accurate interpretable hydrological, climatological, environmental predictions. Given escalating brought about change, findings this make contributions sustainable management, decision-making support, disaster preparedness at global level.

Язык: Английский

Процитировано

0