Temperature based generalized wavelet-neural network models to estimate evapotranspiration in India

Abstract

In this paper, generalized wavelet-neural network (WNN) based models were developed for estimating reference evapotranspiration (ETo) corresponding to Hargreaves (HG) method for different agro-ecological regions (AERs): semi-arid, arid, sub-humid, and humid in India. The input and target to the WNN models are climate data (minimum and maximum air temperature) and ETo (estimated from FAO-56 Penman Monteith method), respectively. The developed WNN models were compared with the various generalized conventional models such as artificial neural networks (ANN), linear regression (LR), wavelet regression (WR), and HG method to test the best performed model. The performance indices used for the comparison include root mean squared error (RMSE), Nash-Sutcliffe efficiency (NSE), the ratio of average output to the average target ETo values (Rratio), and relative percentage (RP). The WNN and ANN models were performed better as compared to LR, WR and HG methods. Further, the best performed WNN and ANN models were tested on locations, which were not included in training to test their generalizing capability. It is concluded that the WNN and ANN models were shown good generalizing capability for the tested locations as compared to HG method.