A hybrid framework: singular value decomposition and kernel ridge regression optimized using mathematical-based...

Area of Science:

• Hydrology and Water Resource Management
• Computational Science and Engineering
• Environmental Monitoring

Background:

• Precise monitoring and timely alerting of river water levels are critical for safeguarding communities in river basins.
• Accurate river water level forecasting is essential for effective flood control and water resource management.

Purpose of the Study:

• To develop a novel hybrid model for highly accurate river water level forecasting.
• To evaluate the performance of the proposed model for one- and three-day ahead predictions in Canadian rivers.

Main Methods:

• A hybrid model integrating singular value decomposition (SVD), kernel-based ridge regression (SKRidge), multivariate variational mode decomposition (MVMD), light gradient boosting machine (LGBM) for feature selection, and Runge-Kutta optimization (RUN) for parameter tuning.
• The L-SKRidge model combines linear and regularization techniques with adaptive algorithms to capture complex data patterns and prevent overfitting.
• Application of the L-SKRidge model to forecast water levels in the Brook and Dunk Rivers, Canada.

Main Results:

• The L-SKRidge model demonstrated superior forecasting efficiency in both the Brook and Dunk Rivers.
• Achieved high accuracy with R = 0.970 and RMSE = 0.051 for the Brook River, and R = 0.958 and RMSE = 0.039 for the Dunk River.
• Outperformed other hybrid and standalone forecasting frameworks in accuracy and robustness.

Conclusions:

• The L-SKRidge method provides an acceptable and accurate capability for river water level prediction.
• The model's effectiveness supports academics and policymakers in developing hydraulic control strategies and advancing sustainable water resource management.