The prediction of aquifer groundwater level based on spatial clustering approach using machine learning

Area of Science:

• Hydrology and Water Resources Management
• Artificial Intelligence in Environmental Science

Background:

• Effective water resources management hinges on understanding water availability and aquifer status.
• Simulation models are crucial tools for addressing complex groundwater challenges.

Purpose of the Study:

• To simulate groundwater levels and assess aquifer quantitative status using data-based models.
• To identify key variables influencing groundwater level prediction.
• To evaluate the performance of Group Method of Data Handling (GMDH), Bayesian Network (BN), and Artificial Neural Network (ANN) models.

Main Methods:

• Selected five observation wells in the Birjand aquifer using spatial clustering.
• Developed 10 prediction scenarios incorporating variables like past groundwater levels, exploitation, precipitation, and climate data.
• Employed GMDH, BN, and ANN models for groundwater level simulation and performance evaluation.

Main Results:

• The GMDH model demonstrated superior prediction performance, achieving an R² of 0.97.
• Key predictors for GMDH included previous month's groundwater level, aquifer exploitation, and precipitation.
• Hydrograph simulation over 6 years indicated a critical groundwater level and a predicted loss of 1.2 meters.

Conclusions:

• The GMDH model is highly effective for simulating groundwater levels and assessing aquifer status.
• The Birjand aquifer is in a critical condition, necessitating immediate management interventions.
• Data-driven modeling provides valuable insights for sustainable groundwater resource management.