Advanced Soft Computing Ensemble for Modeling Contaminant Transport in River Systems: A Comparative Analysis and Ecological Impact Assessment

The paper applies soft computing techniques to contaminant transport modeling in river systems and focuses on the Monocacy River. The research employed various techniques, including Artificial Neural Networks (ANN), Adaptive Neuro-Fuzzy Inference Systems (ANFIS), Support Vector Regression (SVR), and Genetic Algorithms (GA), to predict pollutant concentrations and estimate transport parameters. The ANN, particularly the Long Short-Term Memory architecture, had more superior performance: the lowest RMSE of 0.37, and the highest R-squared was 0.958. The RMSE obtained by the ANFIS model was 0.40, with an R-squared value of 0.945. It provided a balance with accuracy and interpretability. SVR performance with RBF kernel was robust; it has attained an RMSE of 0.42 and R-squared of 0.940, along with very fast training times. The flow velocities and the longitudinal dispersion coefficients at different reaches were estimated to be in the range of 0.30 to 0.42 m/s for average flow velocity and 0.18 to 0.31 m²/s for the longitudinal dispersion coefficient. In addition, the potentially affected fraction of species due to peak concentrations was used to reflect the assessment of ecological impact, which had values ranging from 0.07 to 0.35. For the time-varying estimation, there is supposed to be a variation in the dispersion coefficient and the decay rate over 48 hours, from 0.75 to 0.89 m²/s and from 0.10 to 0.13 day⁻¹, respectively. The research demonstrates the potential of soft computing approaches for modeling complex pollutant dynamics and further provides valuable insights into river management and environmental protection strategies.