{"id":824,"date":"2025-06-23T12:43:12","date_gmt":"2025-06-23T15:43:12","guid":{"rendered":"https:\/\/staging-brasil.mapbiomas.org\/?page_id=824"},"modified":"2026-05-21T11:29:24","modified_gmt":"2026-05-21T14:29:24","slug":"metodo-mapbiomas-fogo","status":"publish","type":"page","link":"https:\/\/brasil.mapbiomas.org\/en\/metodo-mapbiomas-fogo\/","title":{"rendered":"M\u00c9TODO MAPBIOMAS FOGO ANUAL COLE\u00c7\u00c3O 5"},"content":{"rendered":"

Para acesso aos Produtos do MapBiomas Fogo<\/strong> (incluindo Relat\u00f3rio Anual, assets e toolkit) acesse esse LINK<\/a><\/p>\n\n\n\n

Apresentamos um resumo do m\u00e9todo desenvolvido e aplicado na Cole\u00e7\u00e3o 5 do mapeamento das cicatrizes de fogo do Brasil (1985-2025). Para detalhes metodol\u00f3gicos completos, acesse o documento ATBD (Documento Base da Teoria do Algoritmo) neste LINK<\/a>.<\/p>\n\n\n\n

GENERAL CHARACTERISTICS<\/h6>\n\n\n\n

Todo o mapeamento de cicatrizes de fogo no Brasil foi baseado em mosaicos de imagens dos sat\u00e9lites Landsat com resolu\u00e7\u00e3o espacial de 30 metros. O per\u00edodo de mapeamento abrange de 1985 a 2025, com dados mensais e anuais de cicatrizes de queimada cobrindo todo o territ\u00f3rio brasileiro.<\/p>\n\n\n\n

O processo foi realizado de forma colaborativa entre as institui\u00e7\u00f5es do MapBiomas, utilizando intelig\u00eancia artificial com algoritmos de aprendizagem de m\u00e1quina (deep learning) na plataforma Google Earth Engine e Google Cloud Platforms, que oferecem imensa capacidade de processamento em nuvem, al\u00e9m de servidores locais para processamento adicional.<\/p>\n\n\n\n

The work was organized by biomes and regions, with sample collection in burned and unburned areas for training the algorithm by region. In addition, reference maps were used, such as the MODIS Burned Area product (MCD64A1 \u2013 https:\/\/lpdaac.usgs.gov\/products\/mcd64a1v006\/<\/a>) with a spatial resolution of 500 m, GABAM (Global Annual Burned Area Map - https:\/\/gee-community-catalog.org\/projects\/gabam\/<\/a>) with 30 m resolution, fire hotspots, and burn scars from INPE (https:\/\/terrabrasilis.dpi.inpe.br\/queimadas\/bdqueimadas\/<\/a>).<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n
2. VIS\u00c3O GERAL DO M\u00c9TODO<\/h6>\n\n\n\n

The image processing and classification routines used to map burned areas in Brazil followed six stages:<\/p>\n\n\n\n

    \n
  1. Definition of classification regions by biome: The biomes were divided into regions to increase the accuracy of the classification.<\/li>\n<\/ol>\n\n\n\n
      \n
    1. Creation of annual Landsat mosaics: An annual statistical approach was used to create quality annual mosaics by composing all 16-day images into a single mosaic, selecting the pixel with the lowest NBR (Normalized Burn Ratio) value.<\/li>\n<\/ol>\n\n\n\n
        \n
      1. Collection of training samples: Spectral signatures of burned and unburned areas were collected from the annual mosaics to make up the model's training sample set.<\/li>\n<\/ol>\n\n\n\n
          \n
        1. Classification with a Deep Learning model: The DNN (Deep Neural Network) model was trained using the collected samples and the RED, NIR, SWIR1 and SWIR2 spectral bands to classify burnt areas.<\/li>\n<\/ol>\n\n\n\n
            \n
          1. Post-classification: Masks and spatial filters were applied to improve accuracy and reduce noise in the classification results.<\/li>\n<\/ol>\n\n\n\n
              \n
            1. Validation with reference data and visual checks: The classification results were validated using reference data, along with visual checks.<\/li>\n<\/ol>\n\n\n\n
              \"\"<\/figure>\n\n\n\n
              2.1. DEFINITION OF REGIONS BY BIOME<\/h6>\n\n\n\n

              Para cada bioma, foram definidas regi\u00f5es espec\u00edficas para a coleta de amostras de treinamento e a classifica\u00e7\u00e3o de \u00e1reas queimadas, com o objetivo de obter uma classifica\u00e7\u00e3o mais precisa, baseada em fatores edafoclim\u00e1ticos e na vegeta\u00e7\u00e3o regional. As seguintes regi\u00f5es foram definidas para cada bioma:<\/p>\n\n\n\n

              \"\"<\/figure>\n\n\n\n
              2.2. ANNUAL MOSAICS<\/h6>\n\n\n\n

              A classifica\u00e7\u00e3o foi realizada usando mosaicos de reflet\u00e2ncia de superf\u00edcie (SR) da USGS Landsat Collection 2 (Tier 1) (30m \u00d7 30m), constru\u00eddos para cada ano de 1985 a 2025. Foram avaliadas todas as cenas dispon\u00edveis dos sat\u00e9lites Landsat 5 (de 1985 a 2011), Landsat 7 (de 1999 a 2021), Landsat 8 (de 2013 a 2025) e Landsat 9 (de 2022 a 2025), com um intervalo de retorno de 16 dias.<\/p>\n\n\n\n

              2.3. SAMPLE COLLECTION<\/h6>\n\n\n\n

              We created a spectral library based on the manual delineation of burned and unburned areas to be used as training samples. These samples were stratified by Landsat sensors (collected in different years) and by each biome. The collection of training samples was carried out across all 28 classification regions, ensuring the representation of the distinct spectral characteristics present in each region.<\/p>\n\n\n\n

              Thus, we obtained a sample set per sensor and for the 28 regions of Brazil, to be used for training the classification model.<\/p>\n\n\n\n

              2.4. CLASSIFICATION<\/h6>\n\n\n\n

              The classification model used was the Deep Neural Network (DNN), which consists of computational models based on mathematical calculations capable of performing machine learning and visual pattern recognition.<\/p>\n\n\n\n

              The burned area mapping algorithm consisted of two phases: training and prediction. Based on the training samples of burned and unburned areas, the following spectral bands were used as input for the burned area classification model: red (RED \u2013 0.65 \u00b5m), near-infrared (NIR \u2013 0.86 \u00b5m), and short-wave infrared (SWIR 1 \u2013 1.6 \u00b5m and SWIR 2 \u2013 2.2 \u00b5m). These Landsat spectral bands were chosen due to their sensitivity to fire events.<\/p>\n\n\n\n

              The training data were divided into two sets: 70% of the samples were used for training and 30% for testing.<\/p>\n\n\n\n

              2.5. POST-CLASSIFICATION<\/h6>\n\n\n\n

              Ap\u00f3s treinar e testar o modelo, a classifica\u00e7\u00e3o foi aplicada com imagens Landsat quinzenais para todo o per\u00edodo de an\u00e1lise (1985 a 2025). Um filtro espacial foi aplicado para remo\u00e7\u00e3o de ru\u00eddo e preenchimento de pequenas lacunas vazias: \u00e1reas menores ou iguais a 1,4 hectares (16 pixels) foram removidas, e lacunas vazias menores ou iguais a 5,8 hectares (64 pixels) foram preenchidas como \u00e1reas queimadas.<\/p>\n\n\n\n

              Ap\u00f3s a avalia\u00e7\u00e3o dos resultados da classifica\u00e7\u00e3o, foram aplicados filtros de p\u00f3s-classifica\u00e7\u00e3o para remover pixels que estavam nas seguintes classes de cobertura e uso da terra da Cole\u00e7\u00e3o 10.1 do MapBiomas nos biomas:<\/p>\n\n\n\n