Abstract

Soil erosion is a serious environmental problem in South Asia in general and Nepal in particular, where more than 80 percent of the land area is mountainous and tectonically active. Deforestation, overgrazing and intensive farming, unscientific cultivation, haphazard construction and intensive monsoon have accelerated the erosion problem in Nepal. About 1.63 mm of top soil is dislodged from total land of Nepal every year. The present paper focuses on the suspended sediment load assessment in Kankai Mai River in Nepal at Mainachuli. The area of Kankaimai watershed is 1180 km2 and lies between 87°35’ to 88°10’ latitude and 26°37’ to 27°05’ longitude. Elevation of the watershed varies from 3636 m to 125 m over a length of 90 km with average slope of 4%.

Vegetational and geomorphologic analysis of the watershed reveals that Kankaimai watershed is fairly good watershed with moderately peak flow of shorter duration. About 80% of the watershed area is covered with vegetation. The observed daily sediment yield and runoff data are used to develop the ANN model. ANN models developed in this study is consisted of three layers, uses sigmoid transfer function and back propagation algorithm for calculating the weightage. Four sets of inputs consisting of runoff and sediment yield of different lag times are considered to develop the models. The model that gives the maximum correlation and minimum RMSE is selected. Prediction of sediment yield with runoff of the same day as input produces better results then the other inputs. Regression models are also proposed using the same sets of inputs. It is found that ANN produces better sediment yield than regression model.

Keywords: Artificial Neural Network, Sediment Yield Prediction, Nepal, Kankaimai, Regression model