Accurate medium- to long-term runoff forecasting is of great significance for flood control, drought mitigation, comprehensive water resource management, and ecological restoration. However, runoff formation is a complex process influenced by various natural and anthropogenic factors, resulting in nonlinearity, nonstationarity, and long prediction periods, which complicate forecasting efforts. Traditional statistical models, which primarily focus on individual runoff sequences, struggle to integrate multi-source data, limiting their predictive accuracy. This narrow approach overlooks the multifaceted variables influencing runoff, resulting in incomplete and less reliable predictions. To address these challenges, we selected and integrated Random Forest (RF), Support Vector Regression (SVR), and Multilayer Perceptron Regression (MLPR) to develop two coupled intelligent prediction models—RF-SVR and RF-MLPR—due to their complementary strengths. RF effectively removes collinear and redundant information from high-dimensional data, while SVR and MLPR handle nonlinearity and nonstationarity, offering enhanced generalization capabilities. Specifically, MLPR, with its deep learning structure, can extract more complex latent information from data, making it particularly suitable for long-term forecasting. The proposed models were tested in the Yalong River Basin (YLRB), where accurate medium- to long-term runoff forecasts are essential for ecological management, flood control, and optimal water resource allocation. The results demonstrate the following: (1) The impact of atmospheric circulation indices on YLRB runoff exhibits a one-month lag, providing crucial insights for water resource scheduling and flood prevention. (2) The coupled models effectively eliminate collinearity and redundant variables, improving prediction accuracy across all forecast periods. (3) Compared to single baseline models, the coupled models demonstrated significant performance improvements across six evaluation metrics. For instance, the RF-MLPR model achieved a 3.7%–6.5% improvement in the Nash-Sutcliffe efficiency (NSE) metric across four hydrological stations compared to the RF-SVR model. (4) Prediction accuracy decreased with longer forecast periods, with the R² value dropping from 0.8886 for a 1-month forecast to 0.6358 for a 12-month forecast, indicating the increasing challenge of long-term predictions due to greater uncertainty and the accumulation of influencing factors over time. (5) The RF-MLPR model outperformed the RF-SVR model, demonstrating a superior ability to capture the complex, nonlinear relationships inherent in the data. For example, in terms of the R² metric, the RF-MLPR model’s performance at the Jinping hydrological station improved by 6.5% compared to the RF-SVR model. Similarly, at the Lianghekou station, for a one-month lead prediction period, the RF-MLPR model’s R² value was 7.9% higher than that of the RF-SVR model. The significance of this research lies not only in its contribution to improving hydrological prediction accuracy but also in its broader applicability. The proposed coupled prediction models provide practical tools for water resource management, flood control planning, and drought mitigation in regions with similar hydrological characteristics. Furthermore, the framework’s flexibility in parameterization and its ability to integrate multi-source data offer valuable insights for interdisciplinary applications across environmental sciences, meteorology, and climate prediction, making it a globally relevant contribution to addressing water management challenges under changing climatic conditions.

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

The water process refers to the movement and changes in water on Earth, encompassing changes among its three states and its spatial movement. This process is vital for human society as it directly influences water resources, environmental sustainability, and climate regulation. Previous studies have used various related factors to analyze the water process but have not explained the rationale behind selecting these factors from the perspective of pathways. Based on this, the paper explores the construction and application of a top-down water process knowledge graph to clarify the changing process of water movement and the sources of influencing factors. Firstly, we define the concept of the water process and classify its entities based on the concept of water boundaries. Secondly, we identify key knowledge components of the water process, including water bodies, processes, and influencing factors. Finally, we construct and analyze a knowledge graph of the water process and its influencing factors. Results show that (1) the paths of water process help us understand the movement and change process of the water bodies; (2) the number of paths increases with the length of the connection between entities, reflecting the complexity of water process relationships; and (3) tracing these pathways can help identify their influencing factors, providing a data foundation for applying deep learning algorithms in water process research.