Estimação da evapotranspiração de referência no estado do Rio de Janeiro usando redes neurais artificiais.

Abstract

This work was performed with the aim of proposing an artificial neural network (ANN) to estimate the reference evapotranspiration (ETo) as a function of geographic position coordinates and air temperature in Rio de Janeiro State. Data used for the network training were collected from 18 Historical series of climatic elements. Starting from historical series the ETo was calculated by Penman-Monteith (FAO-56) method and used as a reference for network training. ANNs of multilayer perceptron type (MLP) were trained to estimate ETo as a function of latitude, longitude, altitude, average air temperature, thermal amplitude and day of the year. After training with different network configurations the one showing best performance was selected, and was composed by only one intermediary layer (with twenty neurons and sigmoid logistic activation function) and one output layer (with one neuron and linear activation function). According to the results obtained on the test stage we can conclude that, considering only geographical positioning coordinates and air temperature it is possible to estimate daily ETo in Rio de Janeiro State by using an ANN.