Enhancing local-scale groundwater quality predictions using advanced machine learning approaches.

Area of Science:

• Environmental Science
• Hydrogeology
• Data Science

Background:

• Traditional groundwater quality assessment relies on time-consuming and expensive laboratory tests, hindering real-time, local-level monitoring.
• Existing spatial groundwater quality models often lack accuracy at local scales due to complex hydrogeological and anthropogenic factors.

Purpose of the Study:

• To identify robust machine learning algorithms for accurate groundwater quality forecasting at local monitoring sites.
• To utilize easily measurable water quality parameters for rapid assessment, reducing reliance on extensive sampling and lab work.

Main Methods:

• Calculated the Entropy-weighted Water Quality Index (EWQI) using extensive historical data (2014-2021) from 977 wells.
• Employed Random Forest (RF), eXtreme Gradient Boosting (XGB), and Deep Neural Network (DNN) models to predict EWQI.
• Utilized easily measured parameters (pH, total hardness, total dissolved solids) as input variables for model training and local-scale validation.

Main Results:

• All three machine learning models achieved over 90% accuracy (R²) in predicting EWQI during training and local-scale validation.
• Models demonstrated minimal prediction errors when using pH, total hardness, and total dissolved solids as input.
• The study successfully predicted EWQI at the village level, closely matching actual values.

Conclusions:

• Machine learning models can effectively forecast groundwater quality using basic, readily measurable parameters.
• This approach provides a reliable and efficient method for local groundwater quality representation, bypassing costly laboratory analyses.
• The developed models offer a valuable tool for timely and accurate groundwater quality management at the local level.