Multiple kernel fusion: A novel approach for lake water depth modeling

Area of Science:

• Hydrological modeling
• Machine learning applications in environmental science
• Water resource management

Background:

• Lake water depth is critical for reservoir management and hydrological analysis.
• Traditional hydrological models often struggle with complex, multi-variable interactions.
• Multiple kernel fusion (MKF) offers a novel approach to integrate diverse data sources.

Purpose of the Study:

• To apply and evaluate the Multiple Kernel Fusion (MKF) approach for lake water depth simulation.
• To investigate the predictive capabilities of MKF using hydro-meteorological variables.
• To compare MKF performance against established regression and neural network models.

Main Methods:

• Utilized 40-year hydro-meteorological data (groundwater, streamflow, precipitation, evaporation) near Lake Urmia.
• Employed kernel regression to assess individual parameter importance.
• Implemented MKF to learn optimal weights for combining multiple base kernels in regression.

Main Results:

• MKF demonstrated superior predictive accuracy for lake water depth compared to Kernel Ridge Regression (KRR), Support Vector Regression (SVR), Back Propagation Neural Network (BPNN), and Auto Regressive (AR) models.
• Individual hydro-meteorological parameters showed limited predictive power alone.
• Optimal combination of all input parameters via MKF significantly enhanced model accuracy (RMSE = 0.098 m, R² = 0.987, NSE = 0.986).

Conclusions:

• MKF provides a powerful and accurate method for hydrological modeling, specifically for lake water depth prediction.
• The approach effectively integrates multiple hydro-meteorological variables for improved forecasting.
• MKF shows significant potential for addressing complex challenges in hydrological science.