Estimating Reference Evapotranspiration using Artificial Neural Networks

The objective of this paper is to investigate the potential of artificial neural networks (ANNs) for estimation of monthly crop reference evapotranspiration (ET_o) and to compare the performance of ANNs with Penman-Monteith (P-M) method used to estimate ET_o. Several issues associated with the use of ANNs are examined including learning methods and number of neurons in the hidden layer. ANNs were trained and tested using P-M estimated ET_o. Three learning algorithms namely standard Backpropagation, Backpropagation with momentum and Conjugate gradient were used to train the networks. The training and testing results were compared with each other. The best ANN architecture for estimation of monthly ET_o was obtained for Mahanadi reservoir project command area located at Raipur (Chhattisgarh) in India. The networks were trained with monthly climate data (maximum and minimum temperature, maximum and minimum relative humidity, wind speed, and solar radiation) as input and the P-M estimate as output. The best ANN architecture for estimation of monthly ET_o was selected on the basis of statistical parameters like square estimates of error (SEE), adjusted square estimates of error (ASEE) and model efficiency. ANN model trained with the Conjugate gradient algorithm was found to be the best with minimum SEE of 0.225 mm/day, ASEE of 0.195 mm/day and maximum model efficiency of 98.8%.