Resuspension Velocity Prediction of Fine Sediment using Radial Basis Function and Recurrent Neural Network

Abstract

Siltation, originating from urbanization and large-scale development, constitutes a form of water pollution precipitated by the presence of fine sediment, primarily silt and clay. As runoff from sloping terrain carries eroded soil into water bodies, it gives rise to turbid water, detrimentally impacting water quality. Despite numerous research efforts to investigate sedimentation concerns, a comprehensive understanding of siltation problems is still limited. Hence, this study aims to formulate a mathematical model to predict the resuspension velocity of fine sediment in water bodies. Two distinct techniques were employed to construct the predictive model, namely radial basis function (RBF) and recurrent neural network (RNN). The input variables included particle size, flow rate, y-axis movement, dmax, and d/d_max, while resuspension velocities served as the output. To ensure robust training, the experimental data were partitioned into a ratio of 80:20, with 80% allocated for training and the remainder for testing. The efficacy of the developed AI models was assessed using metrics such as mean absolute error (MAE), root-mean-square error (RMSE), and coefficient of determination (R²). RBF appears as the model with better performance, with MAE of 0.0003, RMSE of 0.0003 and R² of 0.6584.