Sistema de Detecção Remota do Desmatamento Florestal no Estado do Espírito Santo

• Ano de defesa: 2024
• Autor(a) principal: Leite, Igor Vieira
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal do Espírito Santo; BR; Mestrado em Ciências Florestais; Centro de Ciências Agrárias e Engenharias; UFES; Programa de Pós-Graduação em Ciências Florestais
• Palavras-chave em Português: Sentinel-2; Mata Atlântica; degradação; desmatamento; subject.br-rjbn; Área(s) do conhecimento do documento (Tabela CNPq)
• Link de acesso: http://repositorio.ufes.br/handle/10/17345

Deforestation and forest degradation pose a significant threat to biodiversity and environmental balance. The state of Espírito Santo, although known for its rich biodiversity, faces challenges in monitoring these environmental impacts. Orbital remote sensing and the development of robust platforms for geospatial data processing, such as Google Earth Engine (GEE), emerge as alternatives to overcome these monitoring challenges. In this context, this study aimed to evaluate a system with high spatial and temporal resolution to monitor forest degradation and deforestation in the Atlantic Forest of Espírito Santo, using Sentinel-2 images. The native forest base map was derived from the annual cover map data of MapBiomas, combined with Sentinel-2 images in the Scene Classification Layer (SCL) band for soil, vegetation, and cloud classes. To validate the data obtained by FlorESat, a confusion matrix was calculated. A pixellevel concordance analysis between classes was performed using the alert database compiled by RAD MapBiomas, which contains 403 polygons from 2019 to 2022. FlorESat's deforestation mapping indicated that the total deforested area for the same years was 1,780.14 ha. In the validation, the proposed system showed mapping accuracy, precision, and specificity of 93.3%, 94.7%, and 93.8% respectively, for 52 randomly delineated polygons in forest areas in ES when compared to photointerpretation. Additionally, it was found that 59.85% of the pixels identified by the FlorESat tool matched directly with the alert database issued and validated in the field. Of the non-coincident pixels, 65.11% were covered by clouds and 34.89% were mapped as forests, highlighting a limitation of orbital data. However, when considering the scenario where cloud pixels are considered as deforested areas, the percentage of concordance was 85.99% between the two datasets.