Dinâmica da cobertura do solo para a Bacia Hidrográfica do Alto Rio Paraná

Abstract

The main objective of this work was to develop land cover products using remote sensing techniques for the Upper Paraná River Basin (UPRB), as well as evaluate changes between the years 1985 and 2015. The products were developed with the SVM (Support Vector Machine) supervised classifier and Landsat TM (Thematic Mapper) and OLI (Operational Land Imager) sensor data. The product classification was based on more than 50 scenes from 1985 and 2015, covering an entire area of the UPRB, and totalizing about 300 training samples for each Landsat scene. Pixel-based and object-based methods were used for the scene classification. The accuracy of the generated products was evaluated through statistical criteria adopted in the literature - Overall Accuracy and Kappa Index. The product was also compared with global products traditionally in use. The results did not show significant variation between the pixel-based and object-based classifications - the object-based classification was more Representative for the year 1985 (Overall Accuracy of 63%) and the pixel-based classification for the year 2015 (Overall Accuracy of 80%). It was observed that the lower values of accuracy are associated to the sub-basins with greater roughness of the terrain, as well as greater spatial complexity of the vegetation - mosaics of cerrado, agriculture and pasture. The comparison with global products showed reasonable agreement between the product developed for 2015 and the global products GlobCover-2009 and Globeland30-2010 - agreement in 50% of the area. On the other hand, the greatest disagreement was found for the MODIS-2013 product, which presents lower spatial resolution, among the evaluated products. Spatially the comparison between the products showed that the sub-basins of greater agreement were those with intense agricultural activities. The most significant change in land cover between 1985 and 2015 refers to the increase in the Agriculture class and the decrease in the areas of the Cerrado/Shrub vegetation class. About 63% of the original Cerrado/shrub vegetation area was classified as Agriculture in the year 2015. For the Cerrado/Shrub vegetation class, the sub-basins with the sharpest reductions were those located on the right bank of the BHAPR, where the class occupied more than 50% of the area in 1985 and decreased to values of a few percentages in 2015. Overall, the results show the advance of the agricultural areas on pasture areas and from these on the Cerrado. It was also evident that the forest areas underwent an intense change, due both to the reduction of native areas and to the emergence of new afforestation and reforestation areas.