摘要：The sediment yield is important factor concern with erosion rate from the catchment which is caused the problems of reducing the storage capacity of reservoirs, creating delta at mouth of rivers and reduces capacity of streamflow, etc. There are several models developed for estimation of sediment yield like USLE, RUSEL and physical based models like SWAT, but they required rigours series of data. In present study artificial neural network model is non linear Black box model used to forecast the sediment yield of Kal river in Maharashtra using the streamflow, stream flow lag by one or two day, rainfall and sediment yield lag one or two day as input to the model. In present study multi layers feed forward back propagation neural network model with one to three input layers, one hidden layers and one output layers were developed. The models were adopted by changing numbers of neuron in hidden layers and epoch. The models performance was evaluated by statistical indices such as R, RMSE, CE, VE, MAD, and MAPE. The study reveal that, ANN model with single input as streamflow and 10 neuron in hidden layer found R values 0.92 and 0.85 during training and cross validation respectively and other indices such as RMSE, CE, VE, MAD and MAPE were 91.58 tons/day, 84.16 per cent, 2.28 per cent, -4.52 per cent and 98.07 per cent during training period where 110.35 ton/day, 76.82 per cent, 0.1 per cent, 10.62 per cent and 20.91 per cent during cross validation period, respectively. It is also observed that, the performance of model increase with increases input parameter and changing combination inputs parameters. The linear regress model developed to compare the performance, found the ANN model performance were better and overall ANN model performance were satisfactory for prediction of sediment yield.
