Machine-Learning Methods for Water Table Depth Prediction in Seasonal Freezing-Thawing Areas.

Area of Science:

• Hydrology
• Environmental Science
• Agricultural Engineering

Background:

• Accurate groundwater hydrology predictions are vital for environmental sustainability in arid agricultural regions with seasonal freezing and thawing.
• Water-saving measurements are critical in these areas, necessitating reliable water table depth modeling.

Purpose of the Study:

• To develop and evaluate advanced machine learning models for accurate long-term water table depth prediction.
• To assess the impact of temperature and water diversion on groundwater levels in arid regions.
• To predict future water table depths under water-saving scenarios and identify environmental risks.

Main Methods:

• Developed a Controlled Auto-Regressive and Ridge Regression (CAR-RR) model for water table depth modeling.
• Integrated temperature data into the model to improve accuracy during freezing-thawing periods.
• Further developed an integrated CAR-SVR model to capture nonlinear dynamics and enhance prediction accuracy.

Main Results:

• The CAR-RR model outperformed Support Vector Regression (SVR) and Multiple Linear Regression (MLR) in validation.
• Including temperature data during the freezing-thawing period significantly improved model accuracy.
• Water diversion was identified as the most critical factor influencing water table depth.

Conclusions:

• The advanced CAR-SVR model provides accurate predictions for water table depth in arid agricultural areas.
• A water diversion volume below 3.64 billion m³ poses a significant risk of environmental problems.
• The study highlights the importance of managing water diversion for groundwater sustainability.