Fig 1.
Location of the YLRB and distribution of 4 hydrological stations.
(The data required to create this figure comes from the National Catalogue Service for Geographic Information: www.webmap.cn).
Fig 2.
Monthly runoff observation values and variation in 4 hydrological stations in the YLRB.
The asterisk * in the figure is used solely as an example for validation dataset verification. In the subsequent algorithm training process, the selection of the validation dataset is performed through random sampling, following the K-fold cross-validation method.
Fig 3.
The basic process of feature selection of the random forest algorithm.
Fig 4.
Schematic diagram of the basic principles of SVR.
Fig 5.
SVR and MLPR algorithm flow optimized by GridSearchCV.
Table 1.
Pseudo code for coupling model execution process.
Fig 6.
Flowchart of the integrated framework developed for runoff prediction.
Fig 7.
The spatial distribution map of the correlation between atmospheric circulation indices and runoff at four hydrological stations in the YLRB.
Fig 8.
Correlation heatmap of atmospheric circulation indices and monthly average runoff at LHK hydrological station with different lag periods.
Fig 9.
Results of RF algorithm screening factors.
Table 2.
Major hyperparameters of RF-SVR.
Table 3.
Major hyperparameters of RF-MLPR.
Table 4.
The ablation study presents the comparison of the coupled model and the baseline model in terms of the same evaluation metrics at the LHK and JP hydrological stations.
Table 5.
The ablation study presents the comparison of the coupled model and the baseline model in terms of the same evaluation metrics at the GD and ET hydrological stations.
Fig 10.
Predicted plots of runoff datasets for LHK and JP hydrological stations.
Subplots A and C show the predicted results of the RF-SVR and RF-MLPR models at the LHK hydrological station, respectively. Subplots B and D depict the predicted results of the RF-SVR and RF-MLPR models at the JP hydrological station.
Fig 11.
Regression plots of test data from LHK and JP hydrological stations.
Subplots A and C display the predictive results of the RF-SVR model and RF-MLPR model, respectively, at the LHK hydrological station. Subplots B and D correspond to the predictive results of the RF-SVR model and RF-MLPR model at the JP hydrological station. Subplots E and G correspond to the predictive results of the RF-SVR and RF-MLPR models at the GD hydrological station, respectively. Subplots F and H correspond to the predictive results of the RF-SVR and RF-MLPR models at the ET hydrological station, respectively.
Fig 12.
Predictive graphs of the RF-SVR and RF-MLPR models for the comprehensive dataset at the GD and JP hydrological sites.
Subplots A and C depict the forecasted outcomes obtained using the RF-SVR and RF-MLPR models, respectively, at the LHK hydrological station. Subplots B and D correspond to the results of predictions made at the JP hydrological station using the RF-SVR and RF-MLPR models.
Fig 13.
Evaluation values of RMSE and MAE for the prediction results of the RF-SVR and RF-MLPR models at 4 hydrological stations in the YLRB.
Fig 14.
Evaluation values of MAPE and MSLE for the prediction results of the RF-SVR and RF-MLPR models at 4 hydrological stations in the YLRB.
Fig 15.
Taylor diagrams for each model for different station test periods.
Subplots A, B, C, and D correspond to the Taylor diagrams of prediction results for the LHK, JP, GD, and ET hydrological stations, respectively.
Fig 16.
R-Square values of the two coupled models for different prediction period.
Fig 17.
The runoff errors of RF-SVR and RF-MLPR at different prediction period.
Table 6.
The evaluation metrics of the coupled model at the LHK hydrological station for different prediction period.