Estimating Evapotranspiration Using Artificial Neural Network and Minimum Climatological Data S. S. Zanetti 1 ; E. F. Sousa 2 ; V. P. S. Oliveira 3 ; F. T. Almeida 4 ; and S. Bernardo 5 Abstract: The objective of this study was to test an artificial neural network ANN for estimating the reference evapotranspiration ETo as a function of the maximum and minimum air temperatures in the Campos dos Goytacazes county, State of Rio de Janeiro. The data used in the network training were obtained from a historical series September 1996 to August 2002 of daily climatic data collected in Campos dos Goytacazes county. When testing the artificial neural network, two historical series were used September 2002 to August 2003 relative to Campos dos Goytacazes, and Viçosa, State of Minas Gerais. The ANNs multilayer perceptron type were trained to estimate ETo as a function of the maximum and minimum air temperatures, extraterrestrial radiation, and the daylight hours; and the last two were previously calculated as a function of either the local latitude or the Julian date. According to the results obtained in this ANN testing phase, it is concluded that when taking into account just the maximum and minimum air temperatures, it is possible to estimate ETo in Campos dos Goytacazes. DOI: 10.1061/ASCE0733-94372007133:283 CE Database subject headings: Climatic data; Forecasting; Evapotranspiration; Neural networks. Introduction Water is becoming a rare and expensive resource in most civilized areas throughout the world. The increased competition of water between urban and agricultural areas is noticeable. It is estimated that more than half of the world population depends on agricul- tural irrigated products Lima et al. 1999. Irrigated agriculture is exposed to a high danger of failing when it is inadequately designed and managed. According to Sousa et al. 2002, an irrigation system is well designed and managed when the application of water is rationally performed at the amount and frequency desirable for the complete development of the crop, by discerningly using the available water resources. Therefore, the quantification of the cropping evapotranspiration is very important. According to Kumar et al. 2002, evapotranspiration is a complex and nonlinear phenomenon, because it depends on the interaction of several climatic elements solar radiation, wind speed, air humidity, and temperature, as well as on the type and growth stage of the crop. According to Pereira et al. 2002, the selection of a method for estimating the evapotranspiration depends on several factors. One of these factors is the availability of meteorological data, as the complex methods requiring a high number of variables have applicability only when all necessary data are available. When there is availability of data, Allen et al. 1998 recommend the application of the Penman-Monteith PM as the sole standard method for the definition and computation of the reference evapo- transpiration ETo. Mendonça et al. 2003 compared the ETo measured in lysim- eter daily and average values at 3, 7, and 10 days in Campos dos Goytacazes with ETo estimated by the PM method. The research- ers found that the PM method satisfactorily estimates ETo. The best estimates are for average values for 7 and 10 days. Although the meteorological variables necessary for the appli- cation of the PM method are not always universally available, in particular those related to the solution of the aerodynamic term wind speed and the deficit of water vapor pressure in the air. So, the methods for estimating ETo as a function of the climatic ele- ments that might be obtained on a more practical way, such as the air temperature and the extraterrestrial radiation, are very impor- tant Hargreaves and Samani 1985; Samani 2000. A tool that can be used to estimate ETo is the artificial neural network ANN. The ANNs are parallel distributed systems, which are composed of simple processing units that calculate some mathematical functions. Such units are arranged into one or more layers and interlinked by a high number of connections. In most models, these connections are associated to weights that store the knowledge represented in the model after the learning process. The behavior of these networks is based on a biological structure conceived by nature: the human brain Braga et al. 2000; Haykin 2001; Kovács 2002. The ANNs have been successfully used to model the relation- ships involving the complex temporary series in several areas of knowledge. According to Galvão et al. 1999, because of their 1 CNPq Grantee, Northern Rio de Janeiro State Univ., Campos dos Goytacazes, Rio de Janeiro, Brazil. 2 Professor, Northern Rio de Janeiro State Univ., Campos dos Goytacazes, Rio de Janeiro, Brazil. 3 Professor, Federal Center of Technological Education of Campos, Campos dos Goytacazes, Rio de Janeiro, Brazil. 4 Professor, Northern Rio de Janeiro State Univ., Campos dos Goytacazes, Rio de Janeiro, Brazil. 5 Professor, Northern Rio de Janeiro State Univ., Campos dos Goytacazes, Rio de Janeiro, Brazil. Note. Discussion open until September 1, 2007. Separate discussions must be submitted for individual papers. To extend the closing date by one month, a written request must be filed with the ASCE Managing Editor. The manuscript for this paper was submitted for review and pos- sible publication on March 11, 2005; approved on June 1, 2006. This paper is part of the Journal of Irrigation and Drainage Engineering, Vol. 133, No. 2, April 1, 2007. ©ASCE, ISSN 0733-9437/2007/2-83–89/ $25.00. JOURNAL OF IRRIGATION AND DRAINAGE ENGINEERING © ASCE / MARCH/APRIL 2007 / 83 Downloaded 07 Feb 2012 to 200.137.65.103. Redistribution subject to ASCE license or copyright. Visit http://www.ascelibrary.org