Data Sets

Scope

For the Amazon biome, the LUC-Brasil repository implements a comprehensive land cover classification system using Landsat satellite imagery from 2000 to 2024. The system produces annual land cover masks through a multi-stage pipeline that combines machine learning classification, temporal rule processing, and year-specific refinement using reference datasets from PRODES and TerraClass. The work processed 25 years of historical data to generate classified masks at 30-meter resolution, supporting deforestation monitoring and land use analysis in the Amazon region.

Image Data

The classification uses Landsat imagery from two distinct sources, each optimized for different temporal periods. This stage produces regularized datacubes with consistent 30m spatial resolution and regular temporal intervals.

GLAD Landsat Historical Data

Historical data for years 2000 to 2014 uses the GLAD (Global Land Analysis & Discovery) service accessed through the OGH (OpenGeoHub) interface. The regularization process aggregates observations into bi-monthly periods. The OGH service provides access to historical Landsat imagery through the GLAD processing system. This source is used for historical datacube generation.

Table 1: Technical specifications of the GLAD Landsat data
Property Value
Source identifier "OGH"
Collection name "LANDSAT-GLAD-2M"
Temporal coverage 2000-2014
Temporal resolution Bi-monthly (P2M)
Spatial resolution 30 meters
Coordinate system EPSG:4326 (WGS84)
Spectral bands BLUE, GREEN, RED, NIR, SWIR1, SWIR2

BDC Recent Landsat Data

The Brazil Data Cube (BDC) provides access to recent Landsat imagery through a pre-processed, analysis-ready data cube infrastructure maintained by INPE (Brazilian National Institute for Space Research). Recent data from 2015-2024 leverages the BDC infrastructure, which provides higher temporal density with monthly aggregation.

Table 2: Technical specifications of the BDC Landsat data
Property Value
Source identifier "BDC"
Collection name "LANDSAT-OLI-16D"
Temporal coverage 2015-2024
Temporal resolution Monthly (P1M)
Spatial resolution 30 meters
Coordinate system BDC_MD_V2 grid system
Spectral bands BLUE, GREEN, RED, NIR08, SWIR16, SWIR22, CLOUD

Map Data Sources

This section lists the maps used in the LUC-Brasil project as additional information, mostly as masks to improve the spatial and temporal accuracy of the result.

PRODES Deforestation Data

PRODES (Programa de Monitoramento do Desmatamento na Amazônia Legal) is Brazil’s official deforestation monitoring program, maintained by INPE. It provides annual deforestation polygons for the Legal Amazon region. The mapping uses Landsat satellite imagery or comparable data to identify and quantify deforested areas greater than or equal to 6.25 hectares. PRODES masks are referenced by multiple reclassification rules throughout the year-specific mask generation process.

Table 3: Technical specifications of the PRODES deforestation data
Property Value
Temporal coverage 2000-2024
Update frequency Annual
Spatial resolution Vector polygons and raster mask
Key attributes Deforestation year and forest/non-forest classification
Usage context Forest mask generation and deforestation reclassification rules

Terraclass Land Cover Surveys

Terraclass is a land use and land cover mapping project for the Brazilian Legal Amazon, produced by INPE and Embrapa. Unlike PRODES (annual), Terraclass provides detailed land cover classifications at specific years, which are used for various reclassification rules.

Table 4: Technical specifications of the Terraclass land cover surveys
Property Value
Survey years 2004, 2008, 2010, 2012, 2014, 2016, 2018, 2020, 2022, 2024
Spatial resolution Landsat - 30m (2004-2016) and Sentinel-2 - 10m (2018-2022)
Key classes (2022-2024 versions) Silviculture, Perennial, Semiperennial, Annual Agriculture, Herbaceous Pasture, Shrubby Pasture, Secondary Vegetation, Mining, Urban area, Water

Water Masks

Water masks are used in temporal consistency rules to identify persistent water bodies across multiple years. These are primarily utilized in temporal processing rules rather than base mask preparation.

Table 5: Technical specifications of the water masks
Property Value
Primary source Terraclass water class
Temporal window 2000-2024
Purpose Identify stable water bodies vs. temporary flooding

Non-forest Vegetation Mask

The Amazon biome also encompasses areas of Cerrado vegetation. Such areas are classified as “Non-forest vegetation” (NF) and comprise approximately 6.6% (280,000 km²) of the biome. NF includes open vegetation formations such as savannas and grasslands; seasonally flooded areas with sandy soils and sparse trees; ecotones; isolated forest patches with deciduous, semi-deciduous, and even broadleaf characteristics; and natural areas of bare lands. The PRODES system uses a mask covering these areas because it is designed only to measure the loss of forest cover. However, since 2022, PRODES has also mapped the clear-cut deforestation in NF vegetation within this mask, producing a historical dataset with biannual estimates from 2000 to 2018 and annual estimates thereafter. The LUC-Brasil team uses the non-forest vegetation mask and the associated deforestation map to distinguish between forest and non-forest areas and to periodically update and refine the mask throughout the study period.

Table 6: Technical specifications of the non-forest vegetation mask
Property Value
Primary source PRODES
Temporal window 2000-2024
Purpose Update the non-forest mask across the time series

Sample Datasets

This section lists the ground samples used by the LUC-Brasil project team as initial sources to select the training data used to build the modules for classification. We collected data from different sources. While all of the data sets had information on geographical location and associated class, their original time series have been collected from different data sources, such as Sentinel-2, HLS (Harmonized Landsat-Sentinel), Landsat-ETM and Landsat-OLI. For the LUC-Brasil maps, only the geographical locations and associated classes were used. All of the time series were collected from Landsat imagery.

These data sets are openly available in the Github repository lulcbrasil-samples.

Land Use and Land Cover in Baixo Tocantins (2021)

Table 7: Sample dataset: land use and land cover in Baixo Tocantins (2021)
Region Baixo Tocantins in Para state (Brazil)
Number of Time Series 533
Satellite-Sensor LANDSAT-OLI
Spatial Resolution 30 meters
Time Extent 2021-01-01 to 2021-12-31
Spectral Bands BLUE, GREEN, RED, NIR08, SWIR16, SWIR22
Spectral Indices NDVI, EVI, MNDWI, NBR
Land Cover Classes Small-Scale Agriculture, Water, Forest, Others, Clean Pasture, Dirty Pasture and Pasture with Regeneration, Urban Area, Advanced Secondary Vegetation, Secondary Vegetation, Large-Scale Agriculture
Note Original data collected using Sentinel-2 images. Geographical locations were re-used by LUC-Brasil to collect time series of Landsat images
Source Anielli Souza, Miguel Monteiro, Isabel Escada (INPE)

Land Cover in the Amazon Rainforest

Table 8: Sample dataset: land cover in the Amazon Rainforest
Region Amazon Rainforest
Number of Time Series 1489
Satellite-Sensor LANDSAT-OLI
Spatial Resolution 30 meters
Time Extent 2020-01-01 to 2020-12-31
Spectral Bands BLUE, GREEN, RED, NIR08, SWIR16, SWIR22
Spectral Indices NDVI, EVI, MNDWI, NBR
Land Cover Classes Forest
Source Luis Sadeck (INPE)

Land Use and Land Cover in Rondonia (2022)

Table 9: Sample dataset: land use and land cover in Rondonia (2022)
Region Rondonia state
Number of Time Series 6007
Satellite-Sensor LANDSAT-OLI
Spatial Resolution 30 meters
Time Extent 2022-01-01 to 2022-12-31
Spectral Bands BLUE, GREEN, RED, NIR08, SWIR16, SWIR2
Spectral Indices NDVI, EVI, , NBR
Land Cover Classes Clear_Cut_Bare_Soil, Clear_Cut_Burned_Area, Clear_Cut_Vegetation, Forest, Mountainside_Forest, Riparian_Forest, Seasonally_Flooded, Water, Wetland
Note Original data collected using Sentinel-2 images. Geographical locations were re-used by LUC-Brasil to collect time series of Landsat images
Source Anielli Souza, Ana Paula Del’Asta, Ana Rorato (INPE)