Rede neural artificial na predição de atributos físicos e químicos do solo

Abstract

The tomato is considered to be one of the most demanding nutrients in the soil and the study of the physical and chemical soil properties is a relatively high cost and time procedure. In the search for alternatives to predict these attributes from a smaller number of soil samples, the use of Artificial Neural Networks (RNA) has been pointed as a computational technique with great capacity to solve problems through experience, since they have the capacity of knowledge acquisition and maintenance. The objective of this work was to evaluate the efficiency of soil physical and chemical attributes estimates obtained through an artificial neural network. The data were collected in an area of 23 ha in an industrial tomato crop irrigated by central pivot in the municipality of Morrinhos (GO). The deformed soil samples were collected at a depth of 0.0 to 0.2 m using a sampling grid of 50x50 m, totaling 120 sampling points to determine the physical and chemical attributes of the soil. We randomly selected 4 known sample points from the 120 points obtained in the soil analysis, to compose the input variables of the RNA, after obtaining the estimates of the physical and chemical attributes using the RNA together with the data determined by the soil analysis, these were submitted to descriptive analysis, geostatistical analysis (ordinary kriging), Student's t-test, soil fertility analysis by Fuzzy logic, fertilizer need and corrective and accuracy analysis of spatial variability maps by Kappa index and Global accuracy. The use of RNA technique was promising to obtain estimates of soil attributes using a smaller number of soil samples. RNA acquired the knowledge necessary to estimate mean values of soil attributes efficiently, but this was not done to estimate attributes on time. Soil fertility was classified as good for the cultivation of industrial tomato both by the data determined in soil analysis and those estimated by RNA. The experimental area, for presenting areas with higher and lower requirements of phosphate fertilizer can be divided into areas of management facilitating the application of variable form.