Dados do Trabalho

Título

PREDICTION OF INFILTRATION RATE FOR SOILS UNDER RECOVERY CONSIDERATING TREATMENTS, TIME AND SOIL RESISTANCE

Resumo

Recovered soil has a faster, uniform water infiltration rate and without puddles and runoff when compared to degraded soil. In this work, an artificial neural network was applied to predict the water infiltration rate in a dystrophic Red Latosol located in the municipality of Selvíria/MS being managed since 1992 according to data obtained from the doctoral thesis of Bonini (2012). For infiltration rate prediction, the Multi-layer Perceptron Neural Network (MLP) through Backpropagation was applied based on 400 data originated from the following input information: nine soil conditions or treatments, analysis time and resistance of the soil. Briefly, soil conditions were limestone+pigeon pea (c+g), limestone+gypsum+pigeon pea (c+ge+g), limestone+gypsum+velvet bean (c+ge+mp), limestone+velvet bean (c+mp), pigeon pea (g), native Cerrado vegetation/ Brazilian Savanna (ma), velvet bean (mp), exposed soil (se) and mobilized soil (sm). The sampled soil surface corresponded to 0-10 cm. The network showed efficient and presented similar results compared to experiment based on an error of 0.0001 for training. In order to validate the presented model, 20% of the data (80) were separated and did not go through the training phase. In this case, an error of 0.0017 for the infiltration rate was obtained during the network operation showing it effectiveness to be applied as an additional tool in the prediction of infiltration data under the presented conditions and parameters. Graphical results of infiltration data as a function of time and soil resistance, as well as treatments, can also be obtained separately by the offered model.

Palavras-chave

Keywords: Artificial Neural Networks, Estimation, Soil Management, Mathematical Modeling.

Instituição financiadora

São Paulo State University - UNESP

Agradecimentos

São Paulo State University - UNESP