Suspended Sediments Modeling & Its Prediction Using AI and Conventional Methods

Abstract

Estimating sediment transport in canals is a tough task because sediment movement is dynamic and complex. Flow velocity, channel geometry, and sediment characteristics keep changing, making it hard to predict accurately. Human activities like dredging or construction also make sediment estimation more complicated. To manage sedimentation effectively, we need accurate modeling techniques and continuous monitoring. Our study focuses on four methods: the conventional sediment rating curve, the statistical model Seasonal Auto Regressive Integrated and Moving Average with exogenous variables (SARIMAX), Feed Forward Artificial Neural Network (ANN), and Long-and-short term memory (LSTM). We're using the Upper Chenab Canal at Head Marala as a case study. Among these methods, ANN performed the best with a R^2 value of 0.886 followed by LSTM with a R^2 value of 0.84. It excelled due to its flexibility in analyzing diverse data. In contrast, the sediment rating curve performed the worst because it depends only on one variable. We chose the best model based on the R^2value and validation performance. The best forecasting model was then used to challenge the barrage operation/SOPs, which helped reduce sediment load.