Fusão de imagens Landsat e MODIS para remoção de nuvens no contexto do mapeamento de áreas de soja

Abstract

Soybean is one of the main agricultural crops in Brazil, which is the world's largest soybeans producer and so it is very important to correctly survey cropped areas, especially within new farming regions. Remote Sensing techniques can assist the identification and mapping of cropped areas mainly by two different methods. The automatic ones are faster whereas the visual ones are time-consuming but more accurate. The use of automatic methodologies may has limited application, especially regarding the use of images that present noisy data. Image fusion is an alternative when there is a need for noise-free images. Thus, this work aimed to develop a rapid and semi-automatic methodology to identificate and map soybean cropped areas in southern Rio Grande do Sul State, Brazil, through the Crop Enhancement Index methodology by using fused OLI and MODIS images. First, we use both set of images to replace noisy pixels within OLI images. Second, we applied three filters to improve mapping accuracy, which were a water mask, a Shortwave infrared (SWIR) band and a temperature image. Finally, to assess the fusion performance we applied the CEI methodology excluding four images from de analysis and also replaced these ones by fully-fused images. The most suitable minimum and maximum IV periods for soybean mapping are from mid-November to early January and from late January to late March, respectively. Filters aproach increased soybean mapping performanc by 4.34% and the increase of each filter was 0.95%, 2.84% and 0.55% for water, SWIR and temperature, respectively. Best soybean mapping accuracy (71.31%) was provided by using images only fused over noisy areas in comparison with the analises that excluded (60.78%) or used four fully fused (69.48%) images.