ANN-based mapping of monthly reference crop evapotranspiration by using altitude, latitude and longitude data in Fars province, Iran

Abstract

The main goal of this study was to evaluate the different feed-forward back-propagation artificial neural networks’ (ANNs) potential to estimate and interpolate the reference crop evapotranspiration (ET0) in Fars province of Iran. ET0 was calculated using the FAO-56 Penman–Monteith method over 24 synoptic stations. Then, altitude, latitude, longitude and the month’s number as inputs and the monthly ET0 as output (target) were used to train the ANNs. In addition, the three-layered ANNs optimized with different training algorithms including gradient descent back-propagation (gd), gradient descent with adaptive learning rate back-propagation (gda), gradient descent with momentum and adaptive learning rate back-propagation (gdx) and scaled conjugate gradient back-propagation (scg). The results indicated that scg algorithm with architecture (4 2 1) had more satisfactory results with the RMSE and R correlation coefficient equal to 18.538 mm and 0.967 in validation phase, respectively. Based on the mentioned architecture of scg algorithm, and input data form different parts of Fars province and surrounding areas, monthly ET0 maps were produced and annual one achieved by summation of monthly maps. The maps particularly annual one showed that highest values of ET0 could be found in the southern and especially southeastern regions, while the lowest values of ET0 could be seen in the northern parts. Contribution of geographic and topographic variables improved the accuracy and spatial details of the resulting maps. It is interesting to note that the fundamental capability of this model is the usage of just a few parameters for ET0 mapping. Since ET0 is a key parameter in water demand planning, therefore, the derived maps could be useful and applicable for many purposes mainly irrigation scheduling in Fars province, Iran.