MapBiomas Colombia monitors land use and land cover changes at a national scale through annual multitemporal mapping, enabling the analysis of ecosystem pressures and human expansion across all Colombian biomes. To produce annual historical land use and land cover maps, the initiative uses Google Earth Engine and collaborates with a network of regional experts. By 2026, Colombia developed Collection 3, covering the entire national territory across all biomes including Amazon, Andes, Pacific, Caribbean, Orinoco, and Insular regions.
colombia-collection-3/
├── STEP-03/ # Training sample generation
│ ├── 031_stable-pixels.js # Stable pixel identification across the time series
│ ├── 032_pixels-proportion.js # Class area proportion calculation
│ ├── 033_samples-generation.js # Training sample generation for Random Forest
│ └── viz-mosaics-col3.js # Multitemporal mosaic visualization
├── STEP-04/ # Random Forest classification
│ ├── 041_rf_classification.js # Pixel-by-pixel Random Forest classification
│ └── viz-classification.js # Classification results visualization
└── STEP-05/ # Post-classification filters
├── join-collections.js # Join Collection 2 and Collection 3 time series
├── gapFill-filter.js # Temporal gap fill
├── temporalFilter.js # Temporal consistency filter
├── frequencyFilter.js # Frequency-based majority filter
├── frequency-adjustFilter.js # Adjusted frequency filter
├── spatialFilter.js # Spatial connectivity filter
├── generalMap-filter.js # General map integration filter
├── getAreas.js # Per-class area statistics export
└── viz-finalClassification.js # Final classification visualization
MapBiomas Colombia Collection 3 maps are generated using machine learning in Google Earth Engine through pixel-by-pixel classification of Landsat imagery at 30 m resolution. The time series covers 1985 to 2025, organized by classification regions across the national territory.
| Step | Description |
|---|---|
| STEP-01 | Annual Landsat mosaic generation (handled by the mosaics pipeline) |
| STEP-02 | Mosaic visualization and quality assessment |
| STEP-03 | Stable pixel identification and training sample generation |
| STEP-04 | Random Forest pixel-by-pixel classification |
| STEP-05 | Post-classification filters: gap fill, temporal, frequency, spatial, and integration |
For full methodological details, read the Colombia Collection 3 ATBD.
Specialized classification workflows run alongside the general LULC pipeline to map specific ecosystems and land uses at greater detail:
| Theme | Repository |
|---|---|
| Mangrove | colombia-mangrove |
| Flooded | colombia-flooded |
| Wetlands | colombia-wetlands |
| Mining | colombia-mining |
| Urban Area | colombia-urban |
| Glacier | colombia-glacier |
The processed outputs are integrated into the MapBiomas Colombia Collection 3 land use and land cover maps, enabling:
- Environmental monitoring and ecosystem accounting
- Public policy development and territorial planning
- Scientific research and biodiversity assessment
- Historical analysis from 1985 to 2024
This initiative is developed by Fundación Gaia Amazonas as part of the MapBiomas Network.
For more information, visit colombia.mapbiomas.org.
Last update: March 2026