Development of Statistical Models for Assessing Pollutant Transport and Intensity in Local Water Systems

Introduction: Assessing pollutant transport and intensity in local water systems is essential for effective Groundwater (GW) pollution risk management. Traditional vulnerability assessment methods often fail to adequately account for pollutant concentrations, especially under varying hydrogeological conditions. Objective: In this research, a statistical model is developed to integrate pollutant concentration data, enhancing GW vulnerability assessments and addressing the limitations of traditional methods. Method: Using nitrate nitrogen as an indicator of GW Pollution Intensity (GPI), the model identifies key influencing features, such as diffusion coefficient, emission concentration, soil density, hydraulic conductivity, GW recharge rate, soil porosity, aquifer depth, and land use type through Principal Component Analysis (PCA). These factors are used to develop an Intelligent Random Forest (Int-RF) model for GPI prediction at contaminated sites. Result: The Int-RF method attained an overall accuracy of 95.3%, with a Mean Absolute Error (MAE) of 0.38 and a Root Mean Square Error (RMSE) of 0.42. The method’s efficiency is also assessed using cross-validation, resulting in a mean R² value of 0.94, indicating strong predictive capability. A comparison with traditional simulation-based methods demonstrated an improvement in prediction efficiency and a reduction in error. The system’s forecasts directly aligned with real measured GPI rates, with a coefficient of 0.94 and a p-value < 0.01 from significance testing, confirming the reliability of the system. Conclusion: The findings indicate that the Int-RF system is a practical and efficient tool for GW source management and land use planning, especially in regions with limited data.