Técnicas de mineração de dados aplicadas a imagens MODIS para mapeamento de culturas de verão no estado do Paraná.
The objective of this work was to develop a methodology for mapping cultivated areas with summer crops (soybean and corn) in the state of Paraná using MODIS sensor images for the crop year of 2013/2014. For classification the Random Forest algorithm was used. It is hard to differ soybean from corn with MODIS. Due to the heterogeneous and high spectral-temporal dynamics of corn and soybean, including proximity or distinction in the sowing and initial development in mesoregions of the state the Random Forest algorithm was applied in order to present a clear differentiation between crops. For the evaluation of the spatial accuracy of the mapping, the Landsat8 / OLI satellite images were used. These images served as reference to generate the error matrix. The proposed method obtained a kappa index of 0.9678, which is considered excellent, and a global of 98.61%, which also represents an excellent index. However, the results obtained from the mapping underestimated the area destined to the culture of Corn in about 50% and overestimated the soybean area by about 24%. Besides that, the methodology was successful in mapping soybean and corn crops in the state of Paraná for the crop year 2013/2014, since the method is fast and inexpensive. Thus, the results indicate that the method is efficient for mapping the summer crops. Nevertheless, it is necessary to make some improvements to minimize the difference between the mapping result and the official data.