classification
Type of resources
Available actions
Topics
Keywords
Contact for the resource
Provided by
Formats
Representation types
Update frequencies
status
Service types
-
This land cover classification refers to a study area in Mato Grosso state, in the Cerrado biome. For this map, the CBERS-4/AWFI monthly data cube was used, with a spatial resolution of 64 meters, using the best pixel composition function (Stack). This experiment uses the time series of an agricultural calendar year, from September 2018 to August 2019, extracted from the CBERS-4/AWFI data cube. The classification was made using 852 samples (Annual Crop: 257; Natural Vegetation: 245; Pasture: 216; Semi-Perennial Crop: 134) and the following data cube bands: bands red, green, blue, and near-infrared along with the EVI, NDVI, GEMI, GNDVI, NDWI2, PVR indices applying the random forest algorithm. The classification quality assessment using 5-fold cross-validation (Wiens et al., 2008) of the training samples showed an overall accuracy of 97.0% and a Kappa index of 0.96. For more information see the paper "CBERS DATA CUBE: A POWERFUL TECHNOLOGY FOR MAPPING AND MONITORING BRAZILIAN BIOMES". To access this resource in GeoTIFF format it is necessary to have an access key provided by the BDC-OAuth service.
-
This land cover classification refers to a study area in Mato Grosso state, in the Cerrado biome. For this map, the CBERS-4/AWFI monthly data cube was used, with a spatial resolution of 64 meters, using the best pixel composition function (Stack). This experiment uses the time series of an agricultural calendar year, from September 2018 to August 2019, extracted from the CBERS-4/AWFI data cube. The classification was made using 852 samples (Annual Crop: 257; Natural Vegetation: 245; Pasture: 216; Semi-Perennial Crop: 134) and the following data cube bands: bands red, green, blue, and near-infrared along with the EVI, NDVI, GEMI, GNDVI, NDWI2, PVR indices. We trained a multi-layer perceptron for a deep learning classification network to classify the data cube using sits R package. To access this resource in GeoTIFF format it is necessary to have an access key provided by the BDC-OAuth service.
-
This land cover classification refers to a study area in Goiás state, in the Cerrado biome. For this map, the CBERS-4/AWFI monthly data cube was used, with a spatial resolution of 64 meters, using the best pixel composition function (Stack). This experiment uses the time series of an agricultural calendar year, from September 2018 to August 2019, extracted from the CBERS-4/AWFI data cube. The classification was made using 701 samples (Annual Crop: 299; Natural: 202; Pasture: 200) and the following data cube bands: red, green, blue, and near-infrared (NIR) along with the NDVI and EVI indices. We trained a multi-layer perceptron for a deep learning classification network to classify the data cube using sits R package. To access this resource in GeoTIFF format it is necessary to have an access key provided by the BDC-OAuth service.
-
This land cover classification refers to a study area in Bahia state, in the Cerrado biome. For this map, the CBERS-4/AWFI monthly data cube was used, with a spatial resolution of 64 meters, using the best pixel composition function (Stack). This experiment uses the time series of an agricultural calendar year, from September 2018 to August 2019, extracted from the CBERS-4/AWFI data cube. The classification was made using 922 samples (Pasture: 258; Agriculture: 242; Natural Vegetation: 422) and the following data cube bands: red, green, blue, and near-infrared (NIR) along with the NDVI and EVI indices. We trained a multi-layer perceptron for a deep learning classification network to classify the data cube using sits R package. Validation was done using good practice guidelines by Olofsson. The validation was done independently for each map using the PRODES Cerrado data of 2019. This data obtained overall accuracy (OA) 0.74. For more information see the paper "Earth Observation Data Cubes for Brazil: Requirements, Methodology and Products". To access this resource in GeoTIFF format it is necessary to have an access key provided by the BDC-OAuth service.
-
This land cover classification refers to a study area in Bahia state, in the Cerrado biome. For this map, the Landsat-8 monthly data cube was used, with a spatial resolution of 30 meters, using the best pixel composition function (Stack). This experiment uses the time series of an agricultural calendar year, from September 2018 to August 2019, extracted from the Sentinel-2 data cube. The classification was made using 922 samples (Pasture: 258; Agriculture: 242; Natural Vegetation: 422) and the following data cube bands: red, green, blue, and near-infrared (NIR) along with the NDVI and EVI indices. We trained a multi-layer perceptron for a deep learning classification network to classify the data cube using sits R package. Validation was done using good practice guidelines by Olofsson. The validation was done independently for each map using the PRODES Cerrado data of 2019. This data obtained overall accuracy (OA) 0.90. For more information see the paper "Earth Observation Data Cubes for Brazil: Requirements, Methodology and Products". To access this resource in GeoTIFF format it is necessary to have an access key provided by the BDC-OAuth service.
-
This is a land cover classification map of Brazilian Cerrado, from August 2017 to August 2018. This classification was made on top of Landsat-8 monthly cubes with spatial resolution of 30 meters, using the best pixel composition function named as Least Cloud Cover First (LCF), which was previously named Stack in BDC older versions. The input datacube was Landsat-8 - OLI - Cube Stack 16 days - v001, which was deprecated. The classification model was trained using 67359 sample points spread across the whole Cerrado biome (Annual Crop: 9390, Dune Beach: 35, Forest: 5439, Pasture: 19697, Savanna: 30014, Semi-Perennial Crop: 1161, Silviculture: 1268, Water: 355). The spectral band used were B1, B2, B3, B4, B5, B6, and B7 along with the vegetation indices EVI and NDVI; the clouded observation were identified using the Fmask4 algorithm and estimated using linear interpolation. The classification algorithm was Multi-Layer Perceptron (Deep Learning). This product was funded by the Brazilian Development Bank (BNDES).
-
This is a land cover classification map of Brazilian Amazon, from January to December of 2018. This classification was made on top of Landsat-8 biweekly cubes with spatial resolution of 30 meters, using the best pixel composition function named as Least Cloud Cover First (LCF), which was previously named Stack in BDC older versions. The input datacube was Landsat-8 - OLI - Cube Stack 16 days - v001, which was deprecated. The classification model was trained using 2303 sample points (Agriculture: 405, Forest: 1284, Pasture: 482). The spectral band used were B1, B2, B3, B4, B5, B6, and B7 along with the vegetation indices EVI and NDVI; the clouded observation were identified using the Fmask4 algorithm and estimated using linear interpolation. The classification algorithm was Random Forest. The post-processing included masking water (using Pekel et al 2016). This product was funded by the Brazilian Development Bank (BNDES).
-
This is a land cover classification map of Brazilian Caatinga, from January to December of 2017. This classification was made on top of Landsat-8 16 days data cubes with spatial resolution of 30 meters, using the best pixel composition function named as Least Cloud Cover First (LCF), which was previously named Stack in BDC older versions. The input datacube was Landsat-8 - OLI - Cube Stack 16 days - v001, which was deprecated. The classification model was trained using 9324 sample points of the classes Agriculture 172, Country formation 577, Forest (Formação florestal) 222, Savanna 4819, Pasture 3538. The spectral band used were B1, B2, B3, B4, B5, B6, B7, along with the vegetation indices EVI and NDVI; the clouded observation were identified using the Fmask algorithm and estimated using linear interpolation. The classification algorithm was Random Forest. The post-processing consisted on cropping the images to the biome's boundary. This product was funded by the Brazilian Development Bank (BNDES).
-
This land cover classification refers to a study area in Mato Grosso state, in the Cerrado biome. For this map, the CBERS-4/WFI monthly data cube was used, with a spatial resolution of 64 meters, using the best pixel composition function named as Least Cloud Cover First (LCF), which was previously named Stack in BDC older versions. This experiment uses the time series of an agricultural calendar year, from September 2018 to August 2019, extracted from the CBERS-4/WFI data cube. The input datacube was CBERS-4 (WFI) Cube Identity - v001, which was deprecated. The classification was made using 852 samples (Annual Crop: 257; Natural Vegetation: 245; Pasture: 216; Semi-Perennial Crop: 134) and the following data cube bands: bands red, green, blue, and near-infrared along with the EVI, NDVI, GEMI, GNDVI, NDWI2, PVR indices. We trained a multi-layer perceptron for a deep learning classification network to classify the data cube using sits R package. This product was funded by the Brazilian Development Bank (BNDES).
-
This is a land cover classification map of Brazilian Pantanal, from January to December of 2018. This classification was made on top of Landsat-8 days data cubes with spatial resolution of 30 meters, using the best pixel composition function named as Least Cloud Cover First (LCF), which was previously named Stack in BDC older versions. The input datacube was Landsat-8 (OLI) Cube Identity - v001, which was deprecated. The classification model was trained using 6601 sample points (Agriculture: 695, Forest: 1586, Pasture: 2930, Savanna and grassland: 1181, Wetland: 209). The spectral band used were B1, B2, B3, B4, B5, B6, B7 along with the vegetation indices EVI and NDVI; the clouded observation were identified using the Fmask4 algorithm and estimated using linear interpolation. The classification algorithm was Random Forest. The post-processing included masking the label Water (Using Pekel et al 2016) and improving the labels Savanna and grassland and Pasture (using PRODES data). This product was funded by the Brazilian Development Bank (BNDES).