Comparative Performance of Multiple Linear Regression and Artificial Neural Network Based Models in Estimation of Evaporation
Neeraj Kumar *
Department of Soil and Water Conservation Engineering, College of Technology, GBPUA&T, Pantnagar, Uttarakhand, 263145, India
Ganesh Upadhyay
Department of Soil and Water Conservation Engineering, College of Technology, GBPUA&T, Pantnagar, Uttarakhand, 263145, India
Pankaj Kumar
Department of Soil and Water Conservation Engineering, College of Technology, GBPUA&T, Pantnagar, Uttarakhand, 263145, India
*Author to whom correspondence should be addressed.
Abstract
Evaporation is an integral part of water cycle. The measurement of evaporation plays a significant role in water management planning, irrigation requirement and to know the water availability in storage system. Considering the complexity in estimation of evaporation by empirical formulas, this study was undertaken to develop regression and neural network based models for estimation of evaporation from climatic variables. The parameters viz. average temperature (), wind speed (W), average relative humidity () and sunshine hours (S) were used as predictors and evaporation was considered as response variable. Mean squared error (MSE) and correlation coefficient (r) were used to judge the performance of developed models. The multiple linear regression (MLR) model exhibited MSE 1.12 and 0.92 whereas with artificial neural network (ANN) model, MSE was found to be 0.56 and 0.68 in training and testing phase, respectively. In training period, correlation coefficient was 0.92 for MLR model as compared to 0.96 with ANN model. The correlation coefficient in testing phase was found to be 0.95 and 0.97 for MLR and ANN model, respectively. The developed ANN model outperformed MLR model in estimation of evaporation from climatic variables.
Keywords: Evaporation, climatic variables, gamma test, multiple linear regression, artificial neural network, performance