The study aims to propose a methodology for estimating the water retention curve for soils of the State of Rio Grande do Sul, by using artificial neural networks. For the development of the research it was assembled a database with information available in the literature, texture and structure of soils of Rio Grande do Sul. The modeling was developed using the software Matlab, where the networks were trained with different architectures, varying the numbers of neurons in the input layer and the hidden layer. The efficiency of the network was analyzed graphically by the ratio 1:1 between the estimated versus the observed data by means of statistical indicators. It was observed from the results that the architecture with best predictive performance was the 4-24-7, with index classification of "great" performance. Thus it can be inferred that the use of neural networks to estimate the water retention curve of the soil is a tool with high predictive ability which will bring great contribution to the agricultural sector.
pedofunctions; artificial intelligence; soil moisture; matric potential
