An innovative approach for predicting groundwater TDS using optimized ensemble machine learning algorithms at two...

Area of Science:

• Environmental Science
• Hydrogeology
• Machine Learning

Background:

• Groundwater salinization is a critical issue in coastal aquifers globally, driven by human activities and natural factors.
• Accurate assessment of groundwater quality is essential for sustainable water resource management, particularly in regions like Oman.
• Seawater intrusion is a primary cause of salinization, necessitating effective monitoring and prediction methods.

Purpose of the Study:

• To develop and evaluate a novel ensemble trees-based (ETB) machine learning strategy for predicting groundwater total dissolved solids (TDS).
• To assess the efficacy of Catboost regression (CBR), Extra Trees regression (ETR), and Bagging regression (BA) models in identifying seawater intrusion.
• To enhance predictive accuracy through a two-level modeling approach, utilizing individual models as inputs for an ensemble model.

Main Methods:

• Employed a two-level ensemble modeling strategy using machine learning algorithms: Catboost regression (CBR), Extra Trees regression (ETR), and Bagging regression (BA).
• Level 1 models (ETR, CBR) used limited input variables (Cl, K, Sr) from groundwater samples.
• Level 2 utilized a Bagging regression (BA) model, incorporating outputs from Level 1 models to improve predictions of groundwater TDS.

Main Results:

• Individual ETR and CBR models at Level 1 demonstrated satisfactory performance in predicting TDS using key chemical indicators.
• The Level 2 Bagging regression (BA) model significantly enhanced prediction accuracy, achieving R² = 0.995 and NSE = 0.996.
• The BA model at Level 2 outperformed individual models in predictive accuracy, generalization capabilities, and accurately identifying saline and non-saline groundwater locations.

Conclusions:

• The proposed two-level ETB machine learning approach effectively predicts groundwater TDS with high accuracy, serving as an early warning system for water quality degradation.
• This method offers a robust tool for monitoring and managing groundwater resources, crucial for ensuring water sustainability in coastal aquifers.
• The study highlights the potential of ensemble machine learning techniques in addressing complex hydrogeological challenges like seawater intrusion.