Vol. 7, Issue 7 (2018)

Author(s):
Mohammad Zeb Bashir, Dr. Vikram Singh and Dr. Arpan Sherring

Abstract:
To predict daily runoff for watershed, using ANN based Multi Layer Perceptron technique, a study was conducted on Nekpur watershed. Four parameters i.e; daily rainfall, previous day rainfall, next previous day rainfall, and previous day runoff were used in Neuro Solution Version 5.0 software for prediction of daily runoff. The ten years data of rainfall and runoff (1994-2003) were collected from DVC Hazaribagh, Jharkhand. Out of 10 years data, 8 years data (1994-2001) were used for training period. Fifteen ANN models for 5 neuron combination were run during the training period to predict daily runoff. The Statistical parameter i.e; Root Mean Square Error (RMSE), Correlation Coefficient(r), and Coefficient of Efficiency (CE), were used to compare observed and predicted runoff for selection of best fitted model-neuron combination for each model. Selected models were again used to generate predicted daily runoff and compared for different error and correlation coefficient. The result of the study indicates that minimum root mean square error (RMSE=0.00104) and maximum correlation coefficient (R²=0.888) was observed for Model 2-Neuron 4 combination. It was concluded that ANN model can be used effectively for prediction of daily runoff for Nekpur watershed.