Inter-comparison of neural network and conventional techniques for estimating evapotranspiration

The evapotranspiration is one of the basic components of the hydrologic cycle and is essential for estimation of irrigation water requirements. This paper shows the utility of artificial neural network (ANN) for estimation of daily reference crop evapotranspiration (ET_o) and compares performance of ANN with conventional techniques like pan evaporation method and multiple linear regression. In this study out of many feed forward neural network architectures, multilayer perceptron has been considered. The neural network were trained with daily mean climatic data (solar radiation, maximum and minimum temperature, mean relative humidity, wind speed and sunshine hours) as input and ET_o estimated by FAO56 Penman-Monteith (PM) model as the desired output. The best ANN architecture was selected on the basis of mean absolute error, mean square error and maximum correlation coefficient. The ANN architecture 5-4-1 (five four and one neurons in the input, hidden and output layers respectively) was trained and tested. Same data set, which was used in ANN, was used for developing multiple linear regression (MREG) model and also for validation. Based on the results it can be concluded that ANN can estimate ET_o better than the conventional techniques. Based on ANN model, evapotranspiration requirements of major crops can be estimated using crop coefficients as provided by Dooronbos and Pruitt (1977). The developed ANN model will be of immense use for estimating water requirements of different crops of north-western Shivalik foothill region of India.