Snowmelt Runoff Prediction by Using Artificial Neural Network and Adaptive Neuro-fuzzy Inference System in Taleghan Watershed

Snowmelt runoff prediction is one of the important challenges in watershed management. The present research was carried out for snowmelt runoff prediction using artificial neural network (ANN) and adaptive neuro-fuzzy inference system (ANFIS) in the Taleghan watershed located in Alborz province. For this research, 38 MODIS instrument images have been obtained from Iranian space agency for years of 2003, 2004, 2005 and 2006. Snow cover area (SCA) was extracted from all images. Then, snow water equivalent volume was computed using SCA and snow depth and density for mentioned years. Daily rainfall, temperature, snow water equivalent variables were used as inputs and daily discharge as output to multilayer feedforward perceptrons using backpropagation algorithm and compared with artificial neural fuzzy interference system (ANFIS). The results reveal that for the Galink hydrometry station, it was found that ANN with RMSE=0.133 m3/s and R2=0.71 in the validation stage are superior in snowmelt runoff forecasting than the ANFIS with RMSE=0.84 m3/s and R2=0.52, respectively. For this station, the ANN models without daily snow water equivalent as input are superior than the ANN models with daily snow water equivalent as input. An increase in the number of inputs from one previous time period to three consecative previous time period proved to be an excellent alternative to perform high quality daily snowmelt runoff prediction. In the other part of study, these comparisons were performed for three other gauging stations, it was found that the ANN in the validation stage is superior in snomelt runoff forecasting than the ANFIS. The ANN models with daily snow water equivalent as input are superior than the ANN models without daily snow water equivalent as input and also an increase in the number of inputs from one previous time period to three consecative previous time period proved to be an excellent alternative to perform high quality daily snowmelt runoff prediction for two stations.