Artificial neural network models for estimation of potential evapotranspiration in a semi-arid region of Raichur, Karnataka

Accurate estimates of evapotranspiration by employing efficient and proven soft computing techniques that involve least number of influencing variables are important to tackle present water crisis. In the present study, Artificial Neural Network (ANN) models were developed to predict the potential evapotranspiration (PET) in Raichur, Karnataka, using six input parameters viz., maximum and minimum temperatures, maximum and minimum relative humidity, sunshine hours and wind speed. The models were trained with Bayesian Regularization (BR) and Gradient Descent training algorithms with Momentum and Adaptive Learning Rate Back Propagation (GDX). The results revealed correlation coefficient of 0.99 between actual and predicted PET for ANN-BR model with 0.1448 mm root mean square error for validation period, which indicated a better performance over the ANN-GDX model. Therefore, ANN-BR model was chosen for predicting PET in the study area.