APLICAÇÃO DO CLASSIFICADOR SVM E DADOS ALTIMÉTRICOS NA ESPACIALIZAÇÃO DE CLASSES DE VEGETAÇÃO NUMA PORÇÃO LITORÂNEA DO ESTADO DO PARANÁ

• Ano de defesa: 2014
• Autor(a) principal: Roza, Willian Samuel Santana da
• Orientador(a): Ribeiro, Selma Regina Aranha
• Banca de defesa: Ponzoni, Flávio Jorge , Moro, Rosemeri Segecin
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: UNIVERSIDADE ESTADUAL DE PONTA GROSSA
• Programa de Pós-Graduação: Programa de Pós Graduação Mestrado em Gestão do Território
• Departamento: Gestão do Território : Sociedade e Natureza
• País: BR
• Palavras-chave em Português: Mata Atlântica; classificação digital multifontes; modelo digital de elevação
• Palavras-chave em Inglês: Atlantic Forest; multisource digital classification; digital elevation model
• Área do conhecimento CNPq: CNPQ::CIENCIAS HUMANAS::GEOGRAFIA
• Link de acesso: http://tede2.uepg.br/jspui/handle/prefix/629

The Atlantic Forest displays many functions that ensure the quality of life for many Brazilians and corresponds to one of the 34 hotspots of biodiversity in the world. In the State of Paraná, areas with the largest remnants of Atlantic Forest are located in the eastern portion of the state, which correspond to the Dense Tropical Rainforest (DTR); it presents the following physiognomic units: ecological forest, alluvial, lowland, submontane, montane and upper montane and non-forest represented by pioneer formations: mangroves, salt fields, salt marshes and refuges vegetation. This study seeks to test strategies to spatialize forest and non-forest remnants of ecological physiognomic units of DTR in the southeastern portion of the State of Paraná. In order to spatialize the vegetation, digital classification through the algorithm Support Vector Machines (SVM) was used. Tests were conducted on Landsat 5 TM spectral bands and ancillary altitude data such as the DEM (Digital Elevation Model) and ASTER (Advanced Spaceborne Thermal Emission Reflection Radiometer) with its byproducts, namely slope and altimetric tracks. First tests were undertaken only with the spectral bands, followed by ones with the spectral bands and ancillary altitude data; and finally the last ones with different SVM settings. To calculate the accuracy of the classified images through Kappa Index (KI) and Confusion Matrix (CM), training samples were collected in images from sensors Spot 5 and P6LIS3, and altitude was verified by means of DEM SRTM (Shuttle Radar Topography Mission). After visual analysis, overall results and classes corresponding to the results from classified images, it was found that just with the spectral bands it was not possible to spatialize forest remnants from ecological physiognomic units of DTR. I was concluded that the separation between the classes of DTR (upper montane, montane, submontane and lowland) was not adequate. But, still observed through visual analysis, there was an accuracy improvement in digital classification when using spectral bands plus DEM ASTER. It seems that the most appropriate result from visual analysis and accuracy of the classified images were obtained through classifying spectral bands over altimetric tracks, enabling GIS to measure the values of the areas in the physiognomic units of DTR. It is noteworthy that all classifications were appropriate, however with the auxiliary altitude data, accuracy was increased in visual analysis, IK and MC by the aid of comparing them to altimetric tracks that define the position of the vegetation according to relief classes as addressed by Veloso, Rangel Filho and Lima (1991). This study serves as a resource for identifying, spatializing and mapping the distribution of forest and non-forest remnants in the southeastern portion of the Paraná DTR. This region encompasses several protected areas, located at both floodplains and slopes of the coastal mountain range.