Previsão de Seca na Bacia Hidrográfica do Açude Epitácio Pessoa/PB com Base em Redes Neurais Artificiais

Abstract

Droughts are natural events associated with below-average precipitation periods in a given region, which can occur in various parts of the planet and cause impacts on agriculture, water supply, tourism, ecosystems and on the provision of environmental services. Between 1994 and 2013, these events affected more than one billion people in the world. In Brazil, droughts occur frequently in the Brazilian Northeast region, due, among other factors, to the high spatial and temporal variability of rainfall. Considering the vulnerability of the Brazilian semi-arid region to these event, this work aims to evaluate the applicability of artificial neural networks (ANNs) in the identification and prediction of droughts using the standardized precipitation index (SPI) in the Epitacio Pessoa watershed, located in the municipality of Boqueirão - a semi-arid region of the State of Paraíba – Brazil. At first, the SPI values 24, 12, 6, and 3 were calculated for the historical series of three selected pluviometric stations. After that, RNAs were created where several architectures were applied to evaluate those that better described the relationship between the variables. The backpropagation algorithm was used to adjust the coefficients involved in the ANNS as an alternative method to the calculation of the SPI - through the construction of ANNs without lag between the data, and its predictive capacity - through the use of RNAs with data lag. The results showed a good capacity of RNAs of 10 neurons in the intermediate layer for the calculation of SPI in all stations studied. However, the construction of networks with data lags proved unable to predict droughts.