A comparative study for estimating reference evapotranspiration using artificial neural network and empirical techniques

Abstract

Evapotranspiration (ET) constitutes a major component of hydrologic cycle and its accurate estimation is essential for crop water balance studies. The present study deals with the estimation of daily reference crop evapotranspiration (ET ) using empirical/ 0 conventional techniques, e.g., FAO-56 Penman-Monteith (PM) model, Blaney-Criddle model, Hargreaves-Samani model and non-conventional technique, e.g., artificial neural network (ANN) for Mohanpur area situated in Nadia district, West Bengal. The study also compared the performance of the ANN technique in ET estimation with the 0 empirical techniques of estimating the same. A feed forward back propagation type ANN structure with a learning rate of 0.6 was used. The networks were trained with the different daily climatic data (maximum and minimum temperature, maximum and minimum relative humidity, solar radiation, and wind speed) as input and the FAO-56 PM estimated ET as output. The performances of the different ANN architectures with 0 the different climatic data set as input were evaluated on the basis of standard error estimation (SEE) and coefficient of determination. The ET values estimated by ANN, 0 Blaney-Criddle and Hargreaves-Samani models were compared with the PM estimated ET . The results inferred that the ANN architecture with maximum number of input 0 variables performed well in estimating ET . The study also revealed that ANN gave 0 better result than the two empirical techniques, i.e. Blaney-Criddle model and Hargreaves- Samani model with the same input conditions. Overall, the results of the study suggested that in sparse data conditions reference ET can be simulated efficiently using ANN and even better results could be obtained than the conventional ET estimation 0 techniques.