Prediction of Suspended Sediment Yield Using Soft Computing Approaches of Mahanadi River

Abstract

Understanding the mass balance between the ocean and the land requires estimating sediment yield. Direct measurement of suspended sediment is challenging considering the time and money required. The suspended sediment yield is influenced by several variables, all of which have non-linear and complex interrelationships. This study proposed artificial intelligence algorithms such as Artificial Neural Networks (ANN) to estimate the suspended sediment yield in Mahanadi River Basin. The hydro-climatic parameters, namely precipitation, stage, and discharge as the direct influencing parameters, temperature, and soil moisture were taken as the indirect influencing parameters to estimate the suspended sediment yield at the Tikarapada gauging station of the Mahanadi River Basin the AI models were compared with the conventional mathematical models of Sediment Rating Curve (SRC) and Multiple Linear Regression (MLR) models. The results demonstrated that the ANN model l has the highest coefficient of determination for the testing data the and least root mean squared error value (RMSE = 0.005) for the testing data. It was followed by ANN, MLR, and finally SRC. ANN model had the least underestimated values of the estimated yield, and these two models were very close to estimating the peak sediment yield values. SRC was the least accurate model, which heavily underestimated the peak yield values. Hence, the ANN model will be beneficial to estimate the yield where suspended sediment yield values are unavailable.