Least square support vector machine-based variational mode decomposition: a new hybrid model for daily river water...

Area of Science:

• Environmental Science
• Hydrology
• Machine Learning

Background:

• Machine learning models are increasingly used for predicting river water temperature (T_w) based on air temperature (T_a).
• Existing models often utilize the nonlinear relationship between T_a and T_w, proving to be robust tools.
• The potential of Variational Mode Decomposition (VMD) to further enhance these predictive models requires investigation.

Purpose of the Study:

• To evaluate the contribution of Variational Mode Decomposition (VMD) in improving machine learning model performance for river water temperature prediction.
• To compare the effectiveness of six different machine learning models under various input scenarios.

Main Methods:

• Acquired and analyzed measured river water temperature and air temperature data from five Polish stations (1987-2014).
• Applied six machine learning models: K-nearest neighbor's regression (KNNR), least square support vector machine (LSSVM), generalized regression neural network (GRNN), cascade correlation artificial neural networks (CCNN), relevance vector machine (RVM), and locally weighted polynomials regression (LWPR).
• Evaluated models using three scenarios: T_a only, T_a + periodicity, and VMD-decomposed T_a (intrinsic mode functions) as inputs.

Main Results:

• Models using only T_a achieved high accuracy (R² ≈ 0.910, NSE ≈ 0.915), with similar performance across all six models.
• Incorporating periodicity significantly improved predictions (R² ≈ 0.956, NSE ≈ 0.955), with LSSVM showing slight advantages.
• VMD preprocessing led to substantial accuracy improvements (up to 40.50% reduction in RMSE, 39.12% in MAE) for some models, though GRNN and KNNR did not benefit.

Conclusions:

• Variational Mode Decomposition (VMD) is a highly capable preprocessing technique for enhancing machine learning-based river water temperature prediction.
• The effectiveness of VMD as a preprocessing step can vary depending on the specific machine learning model employed.
• Combining VMD with appropriate machine learning models offers a promising approach for more accurate hydrological forecasting.