Utilization of breakthrough curve for Estimation of Dispersion Coefficients of Subsurface Aquifer using Machine Learning

To forecast the future direction of pollutant movement, the advection dispersion equation is employed with positive time increments. Solving the Advection-dispersion equation necessitates prior knowledge of the dispersion coefficient, posing the challenge of estimating these coefficients within a subsurface aquifer. The task of deducing longitudinal and transverse coefficients for the subsurface aquifer emerges as an inverse problem. In this study, an innovative approach using Artificial Neural Network (ANN) modeling technique is introduced to estimate the dispersion coefficient of the subsurface aquifer. Training data is based on pollutant concentration over time, commonly referred to as a breakthrough curve. By focusing solely on the section of the breakthrough curve near the peak concentration, the required sampling duration are minimized. This is in contrast to previous research where the complete breakthrough curve was employed to estimate dispersion coefficients. Notably, both longitudinal and transverse dispersion coefficients are concurrently estimated in this study. The developed ANN model is trained using the Levenberg-Marquardt algorithm and is assessed under two and three-dimensional scenarios, incorporating both error-free and error-prone concentration data with a margin of up to 10%. The outcome of the performance evaluation underscores the robustness and efficacy of the ANN model in predicting longitudinal and transverse dispersion coefficients for subsurface aquifers.