Estimativa dos parâmetros de funções de pedotransferência para os solos do Rio Grande do Sul

• Ano de defesa: 2019
• Autor(a) principal: Kayser, Luiz Patric
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Santa Maria; Brasil; Engenharia Agrícola; UFSM; Programa de Pós-Graduação em Engenharia Agrícola; Centro de Ciências Rurais
• Palavras-chave em Português: Redes neurais artificiais; Regressão linear múltipla; Van Genuchten; Curva de retenção de água no solo; Soil water retention curve; Artificial neural networks; Multiple linear regression; CNPQ::CIENCIAS AGRARIAS::ENGENHARIA AGRICOLA
• Link de acesso: http://repositorio.ufsm.br/handle/1/17297

The soil water retention curve is an important information for the rational management of irrigation. Due to the difficulty in generating it through traditional methods, there is a need to create alternative methods, such as pedotransfer functions, that generate the curve indirectly, using data that can be acquired more easily and quickly. In this context, the present work aims to estimate the parameters of the Van Genuchten model through pedotransfer functions with physical-hydro data for soils of the State of Rio Grande do Sul using Multiple Linear Regression and Artificial Neural Networks. To estimate the parameters θr, α and n of the Van Genuchten equation, the levels of sand, silt, clay, soil density (ds), particle density (dp) and organic matter (Mo) were used as independent variables. Multiple linear regression analyzes were performed using the stepwise (Forward) procedure of the IBM SPSS Statistics 25 software, while the artificial neural networks were generated using the multiple layer perceptron function of the same software. The results obtained in the estimation of the parameters α, θr and n can be considered good with both estimation methodologies. In the multiple linear regression the values of the coefficient of determination were higher than 0.9 in most of the proposed models, and the root mean square error presented values lower than 0.008. Using artificial neural networks, the R² values were higher than 0.9 in all the proposed models, and the RMSE presented values lower than 0.0012. With this, we can affirm that the use of Multiple Linear Regression and Artificial Neural Networks was efficient in generating Pedotransfer Functions, for the data base used, estimating the parameters of the Van Genuchten equation with high predictive capacity of the proposed models. that the second methodology had higher values of R² and lower values of RMSE.