Application of Artificial Neural Network for Snowmelt-Runoff (Case Study: Latyan Dam Watershed)

Flood is one of the natural disaster phenomena and flood prediction is very important. The rainfall-runoff process and flood are physical phenomena that these analyses are difficult due to the influence of various parameters. There are different methods and models for these phenomena analysis. This study is carried out for rainfall-runoff process simulation using artificial neural network (ANN) involving snow water equivalent (SWE) in Latyan watershed located in Tehran province. For this reason, 92 images of MODIS were obtained during three years from 2003-2004 to 2006-2007 from the NASA site. Snow cover areas (SCAs) were extracted from each images and SWE were computed during these years. The rainfall, temperature and stream flow were available during these years. Multilayer perception networks with back propagation algorithm were used for finding the structure of the networks. Results showed that ANN with 4-10-1 structure, 4 neurons in input layer, 10 neurons in middle layer and 1 neuron in output layer, with performance coefficient of 0.85, determination coefficient of 0.68 and root mean squared error of 0.04 as the best structure had good precision in runoff estimation and SWE was caused to increase the accuracy of the model.