Keywords: artificial neural networks, evapotranspiration, emergency models, exogenous data.

The combination of the two separate processes whereby water is lost from the soil surface by evaporation and by transpiration is referred to as evapotranspiration. Accurate evapotranspiration estimations are crucial for optimization of irrigation water use in arid and semiarid regions, highly conditioned by water shortages and rising costs. The standard methods that provide accurate enough evapotranspiration estimations cannot be applied in most cases due to low data availability or reliability. So, the development of more efficient models for those cases where scant or no climatic records are available is mandatory. This paper presents a new artificial neural network (ANN) model to estimate reference evapotranspiration when no local climatic data are available, taking advantage of ancillary data records from secondary similar stations, which work as exogenous inputs.

The study was carried out in thirty weather stations of the Valencia region, in the Mediterranean coast of Spain. The proposed models consider 2 types of variables: a local variable, namely theoretical extraterrestrial radiation, and several exogenous variables from similar secondary stations, namely exogenous reference evapotranspiration. A criterion was established to deal with the selection and hierarchization of these secondary stations. This criterion was based on a continental characterization of the study region, carried out through the calculation of the mean Gorezynski continentality index. Fifteen ANN types with 1 up to 15 exogenous inputs, respectively, arranged according to that hierarchization were trained for each station. The ANNs used correspond to multilayer feed-forward networks using back-propagation and supervised training with the Levenberg-Marquardt algorithm.

In general, the model performance improves the more ancillary stations are considered. Nevertheless, the improvement rates decrease, stop or even might become negative if too many exogenous inputs are considered. The proposed models present better performance indicators than two commonly used methods based on local temperature measurements: the calibrated Hargreaves equation [1] and the four-input ANN [2]. The presented models can be used as emergency models when more precise models cannot be applied, because there are not enough climatic measurements for their application. The model can be also used for time series gap infilling, given that it demands no local climatic record.

1

S.S. Zanetti, E.F. Sousa, V.P.S. Oliveira, F.T. Almeida, S. Bernardo, "Estimating evapotranspiration using artificial neural network and minimum climatological data", J. Irrig. Drain. Eng., 133(2), 83-89, 2007. doi:10.1061/(ASCE)0733-9437(2007)133:2(83)

2

G.H. Hargreaves, Z.A. Samani, "Reference crop evapotranspiration from ambient air temperature", Appl. Eng. Agric., 1(2), 96-99, 1985.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £125 +P&P)