Predição de bandas espectrais e classificação de imagens de satélite para identificação e mapeamento de áreas úmidas: um estudo de caso na Bacia Hidrográfica do Rio dos Sinos

Abstract

Wetlands are ecosystems that have high biodiversity and productivity, in addition to promoting multiple ecosystem services of global importance. In Rio Grande do Sul, wetlands are known by the local term “banhados”. These ecosystems, although widely recognized, are still neglected and degraded. The Rio dos Sinos Hydrographic Basin (BHRS) has been suffering constant anthropogenic pressures that result in the degradation of wetlands, mainly due to the population increase that demands habitable areas. In this context, the need to develop conservation strategies for wetlands in the hydrographic basin is extremely urgent. As they are complex ecosystems due to their hydrology, soil, vegetation and spectral characteristics, Remote Sensing (SR) is an efficient tool in the identification and mapping of wetlands, generating subsidies for the public authorities to create protection and inspection strategies. Thus, this dissertation aimed to i) delimit the remaining wetlands of the BHRS using centimetric images GeoEye-1 and free GIS; ii) measure wetlands using an Unmanned Aerial Vehicle (UAV); iii) to propose a method to predict spectral bands from a centimeter-resolution RGB image by means of Artificial Neural Network (ANN), in order to generate an image with high spatial resolution and higher spectral resolution for subsequent classification of wetlands; and iv) prepare a guide for identifying BHRS wetlands, presenting a method for measuring wetland indicators in the field. The results showed that the use of centimetric images increases the accuracy of wetland mapping, with a total area of 93.11km² being mapped. The prediction of spectral bands using ANN was able to generate an image with spatial (2m) and spectral resolution (RGB+NIR+RE 1,2,3,4 + SWIR 1 and 2). The predicted image generated a better delimitation of wetlands when compared to the raw image (RGB+NIR). Finally, the Wetlands Identification Guide provided a scientific and technical basis for identifying wetlands in the field. It was concluded that the results of this work will serve as a subsidy to complement the environmental planning of the BHRS and the conservation of the remaining wetlands.