Identificação e mapeamento de áreas de soja na região sul do Rio Grande do Sul por meio de imagens MODIS

Abstract

Due to the importance of soybean for the Brazilian economy, the accurate identification and mapping of soybean areas are highly relevant for agricultural planning e trading. In this sense, the goal of this work was to develop a rapid and operational methodology to identify and map soybean fields in municipalities of southern Rio Grande do Sul State, Brazil, by adapting the Crop Enhancement Index (CEI) methodology over MODIS images. First, a soybean reference map for the 2013/14 crop year performed by a multitemporal visual interpretation over Landsatimages was obtained for three municipalities. To adapt the CEI methodology to the soybean crop season (periods of maximum and minimum Vegetation Index (VI)), the CEI methodology used 16-days composited MODIS images for red (RED) and Near Infrared (NIR) spectral bands, through the Enhanced VI 2 (EVI2). Then, a spectral adjustment was performed by adding a Short Wave Infrared band (SWIR). During the spatial adapting, MODIS imagery was resampled by using three resampling algorithms (nearest neighbor, bilinear and cubic convolution) and four pixel sizes (250, 125, 62.5 e 30m). Finally, the ability of the CEI/MODIS methodology in identifying soybean considering the field size was assessed. Best result for maximum and minimum VI periods goes from November 17th to January 1st and January 17th to March 22th, respectively. The SWIR band increased 2.5% the performance of CEI for mapping soybean fields. Best result for spatial adapting was provided by cubic convolution and pixel size of 125m, which increased 6.77% the soybean mapping accuracy. Overall, producer’s accuracy was 49.07 and 98.72% and user’s accuracy was 84.31 and 93.97% for soybean and nonsoybean classes, respectively, with global accuracy of 92.63%. Best soybean identification was observed for fields between 50 and 100 (54%) and above 100 ha (83.16%).