Hydrological model parameter regionalization: Runoff estimation using machine learning techniques in the Tha Chin...

Area of Science:

• Hydrology
• Environmental Modeling
• Water Resource Management

Background:

• Accurate hydrological modeling is crucial for understanding water resources, especially in ungauged catchments where data is scarce.
• Parameter estimation for hydrological models like the Soil and Water Assessment Tool (SWAT) presents a significant challenge for runoff prediction.

Purpose of the Study:

• To assess the SWAT model's capability for simulating hydrological processes in the Tha Chin River Basin.
• To evaluate the effectiveness of regionalizing hydrological parameters from a gauged basin (Mae Khlong River Basin) for ungauged catchment modeling.
• To compare the performance of different Machine Learning (ML) techniques in parameter regionalization for runoff prediction.

Main Methods:

• The Soil and Water Assessment Tool (SWAT) was calibrated and validated using historical hydrological data from the Mae Khlong River Basin (1993-2017 for calibration, 2018-2022 for validation).
• Machine Learning models, including Random Forest (RF), Artificial Neural Networks (ANN), and Support Vector Machine (SVM), were employed for parameter regionalization to predict runoff in the ungauged Tha Chin River Basin.
• Model performance was evaluated using statistical metrics, primarily the coefficient of determination (R²).

Main Results:

• The SWAT model demonstrated reasonable accuracy during calibration (R² = 0.85) and satisfactory performance during validation (R² = 0.64) for the Mae Khlong River Basin.
• Among the ML techniques, the Support Vector Machine (SVM) model achieved the highest runoff prediction accuracy for the ungauged catchment (R² = 0.63), slightly outperforming Random Forest (R² = 0.60) and Artificial Neural Networks (R² = 0.61).

Conclusions:

• The study confirms the viability of using the SWAT model coupled with ML-based parameter regionalization for enhancing runoff prediction accuracy in ungauged river basins.
• The findings underscore the potential of integrating hydrological models and advanced ML techniques to improve water resource management and hydrological forecasting.
• Further research is recommended to apply these integrated methodologies across diverse basins and enhance data acquisition strategies for improved model performance.