Estimation and Forecasting of Suspended Sediment Yield in Mahanadi River Basin: Application of Artificial Intelligence Algorithms

Rivers are dynamic geologic agents on earth and act as the main pathways for transport of continental materials to the ocean. Estimation and forecasting of suspended sediment yield (SSY) are essential towards understanding the mass balance between the ocean and land. The volume of SSY transported in a river provides important information about its morphodynamics, the hydrology of its drainage basin, and the erosion and sediment delivery processes operating within that basin. Estimation and forecasting of SSY is always a key factor during the evaluation of reservoir project, protection of fish and wildlife habitats, understanding of flood capacity, hydroelectric equipment longevity, planning and management of any river system. The SSY depends on number of variables and their inter-relationships which are highly nonlinear and complex in nature. Direct measurement of SSY is difficult and needs sufficient time and money. Additionally, it is difficult to estimation and forecasting of the SSY by using traditional mathematical models because they are incapable to handle the complex non-linearity and non-stationarity process. Thus, the aim of this study is to develop hybrid simple genetic algorithm based artificial neural network (GA-ANN) and genetic algorithm based multi objective optimization with artificial neural network (GA-MOO-ANN) models to estimate and forecast the SSY at eleven locations in the Mahanadi river basin, which is one of the largest rivers in India. All parameters associated with the models are optimized simultaneously using simple genetic algorithm and multi-objective genetic algorithm for estimation and forecasting of the SSY. Temporal information of monthly rainfall, temperature, water discharge, and suspended sediment yield during 1990-2010 along with rock type, relief and catchment area are used at all eleven gauging stations in Mahanadi river for the development of models. It is observed that water discharge and SSY show wide fluctuations whereas temperature and rainfall do not show much variation among different gauge stations in the basin. The performances of GA-MOO-ANN and GA-ANN models were compared to traditional artificial neural network (ANN), multiple linear regression (MLR), auto regressive (AR), multi variate auto regressive (MAR) and sediment rating curve (SRC) method for evaluating the estimation and forecasting capability of models on testing data set. The results suggested that the hybrid GA-MOO-ANN and GA-ANN models exhibited satisfactory performance and provided better results than the traditional ANN, MLR, AR and MAR models. The models are unable to estimate and forecast the SSY at gauge stations having very small catchment areas whereas performing satisfactory on locations having moderate to large catchment area. The models provide the best result at Tikarapara, the gauge station location in the extreme downstream, having the largest catchment area. If measurements of suspended sediment are not available in any river then the modelling approach can be potentially used for the estimation and forecasting of SSY at gauge or ungagged locations