Estimativa do índice de área foliar para cultura irrigada por meio de pivô central utilizando imagens de sensoriamento remoto e redes neurais artificiais

Abstract

Irrigation has become one of the main tools that made possible an increase in productivity in theworld. The center pivot irrigation system is one of the most used in the state of Goiás. They are3,326 equipment installed in the state, with a total irrigated area of 242,872.58 ha. The maincultures are grown on pivot soybean, corn, tomato and other industrial crops. For the developmentof this work was chosen a commercial area of 35 ha of industrial tomatoes, irrigated by centerpivot in the municipality of Vila Propício - GO. The choice of the industrial tomato crop tookplace in that it has dense leaf area and be of great commercial importance to the state, which dueto investments in the production of industrial tomatoes increased the volume of productionleading the state to occupy since 2008 1st place in the national tomato production ranking. Theobjective is to obtain the Leaf Area Index (LAI) of tomato, physiological characteristic that isdirectly related to production capacity. We compared two methods of obtaining the LAI: First,by setting the regression models between LAI collected in situ and Normalized DifferenceVegetation Index (NDVI) is obtained by means of remote sensing images. The second method,Artificial Neural Networks (ANN), which was used the same data as input variable regression intraining networks. For Samples were collected using a sampling grid of 60x60 m to 88georeferenced points by two means: In situ using a frame with an area of 1 m2 and remote sensingimages using Landsat 8 and the Sentinel-2. The data collected in situ were number of plants andleaf area, both per m2, which served to calculate the IAF. The data collected by remote sensingimages were georeferenced, which have been treated in order to extract values for calculating theNDVI. The best regression model was the Coefficient of Determination (r2) of 0.67 and RootMean Square Error (RMSE) of less than 11%. The best trained network had the general r2of 0.74and the overall RMSE less than 4%. It follows that the tomato LAI, grown in central pivot, canbe estimated by both methods: Regression Model and ANN, both with the main input variablederived NDVI remote sensing.