Sistema web e mobile para estimativa de evapotranspiração de referência utilizando redes neurais artificiais

• Ano de defesa: 2013
• Autor(a) principal: Ferraz, Rafael Camargo
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Santa Maria; BR; Engenharia Agrícola; UFSM; Programa de Pós-Graduação em Engenharia Agrícola
• Palavras-chave em Português: NeuroEvap; Android; Dispositivos móveis; Penman-Mointeith; Tecnologia da informação; Mobile; Information technology; CNPQ::CIENCIAS AGRARIAS::ENGENHARIA AGRICOLA
• Link de acesso: http://repositorio.ufsm.br/handle/1/3610

The reference evapotranspiration (ETo) is a component of great importance in several areas, as well as studies in agricultural and water resources management. Several methods of determination are studied, the Penman-Monteith widely used as standard. The main disadvantage of this method is the fact that the meteorological data required are usually not easily available, or have a high cost for determination. The insertion of artificial neural networks (ANN) in these studies provides satisfactory results with fewer input variables. This paper's main objective is to develop a web and mobile tool for determining ETo models using artificial neural networks for the State of Rio Grande do Sul The data used for modeling were collected from the stations of the National Meteorological Institute between 2008 and 2012. It was adopted as the standard ETo values estimated by the Penman-Mointeith and then compared with those obtained with the RNA's. The scheduling algorithm was defined as the best neural network architecture, considering the indide performance and error evaluation. Both applications were developed under Linux with free tools and SQLite database. It can be seen that the estimates made with the RNA's have better performance when compared with the known empirical methods, varying its index of determination (R²) between 0,856 and 1,0. Obtained outperform models with daily solar radiation as input. Two model was chosen to implement the system architecture, the first solar radiation with insertion of the second liquid architecture makes use of extraterrestrial radiation. This definition was chosen because of the lack or high cost for obtaining solar radiation data net. It was concluded that artificial neural networks are able to predict the quality of reference evapotranspiration for the State of Rio Grande do Sul enabling applications on web and mobile.