Forest AGE module
Contacts: Eben Broadbent ([email protected]) or Angelica Almeyda Zambrano with any questions, suggestions, or issues to be addressed.
Version release notes
v5.199 -- 01/08/26 -- Continued improvements in age estimation, interaction, and download capabilities.
v3.03 -- 12/22/25 -- FAROS Alpha release with greatly improved age fidelity across complex terrain. Currently available to generate and download on demand for specific study areas as defined using the web app's polygon tool. Many other improvements and additions.
v2.95 -- 12/17/25 -- Uses tiled multiple Geotiff downloads for larger areas (automatically created when selected polygon area exceeds size threshold for a single download.
v1.0 -- 05/01/25 -- Initial app development.
v5.199 -- 01/08/26 -- Continued improvements in age estimation, interaction, and download capabilities.
v3.03 -- 12/22/25 -- FAROS Alpha release with greatly improved age fidelity across complex terrain. Currently available to generate and download on demand for specific study areas as defined using the web app's polygon tool. Many other improvements and additions.
v2.95 -- 12/17/25 -- Uses tiled multiple Geotiff downloads for larger areas (automatically created when selected polygon area exceeds size threshold for a single download.
v1.0 -- 05/01/25 -- Initial app development.
Introduction user guide below
Quick Reference Guide: FAROS GeoTIFF Outputs
The downloaded output is a GeoTIFF (.tif) containing spatially explicit forest age estimates expressed as years since forest establishment. The raster uses a 16-bit integer data type, with each 10 m × 10 m pixel storing a single forest age value. The dataset is provided in the WGS84 coordinate reference system (EPSG:4326).
Pixel values are interpreted as follows. A value of 0 represents non-forest or very recently deforested land. Values from 1 to 10 indicate young regenerating forest. Values from 10 to 50 correspond to young to mid-age forest, 50 to 100 represent mature forest, and values from approximately 100 up to 300 indicate old-growth forest. These values represent elapsed time since forest establishment, not calendar year.
Common Tasks
Task -> How To
Interpreting Map Colors
Colors shown on the map are a visualization aid for forest age classes. Red tones represent young or recently disturbed forest (approximately 0–20 years). Yellow indicates mid-age forest (roughly 40–60 years). Green corresponds to mature forest (60–100 years), while darker green represents old-growth forest, typically exceeding 100 years and extending beyond 300 years in some regions.
Export File Characteristics
All exported files are GeoTIFFs at 10 m spatial resolution with pixel values representing years since forest establishment. Files are delivered in WGS84 coordinates. Individual downloads are limited to approximately 32 MB; larger areas are automatically split into multiple GeoTIFFs.
Troubleshooting
If the application does not load and the map remains light colored or blank, allow at least two full minutes for initialization, as loading can be slow. If the issue persists, refresh the page, try a different modern browser such as Chrome, Firefox, or Edge, and confirm that the internet connection is stable.
If clicking on the map does not return values, ensure that Point Mode is active and highlighted, zoom in to at least level 6, and confirm that the selected location is forested. The map background should be fully rendered (dark) before interacting.
If polygon drawing does not begin, verify that Polygon Mode is selected and highlighted. Clearing any existing polygon before drawing a new one often resolves the issue.
If a download does not start, ensure that the polygon is fully closed by double-clicking the final vertex. Very large polygons may require additional processing time or may need to be reduced in size. Browser pop-up blockers should also be checked.
“No data” messages are expected in some regions without baseline forest age coverage. In these areas, the “FAROS Expansion Age” layer may provide useful information for potential forest areas. Forest age data are not available everywhere, and coverage can vary spatially.
The downloaded output is a GeoTIFF (.tif) containing spatially explicit forest age estimates expressed as years since forest establishment. The raster uses a 16-bit integer data type, with each 10 m × 10 m pixel storing a single forest age value. The dataset is provided in the WGS84 coordinate reference system (EPSG:4326).
Pixel values are interpreted as follows. A value of 0 represents non-forest or very recently deforested land. Values from 1 to 10 indicate young regenerating forest. Values from 10 to 50 correspond to young to mid-age forest, 50 to 100 represent mature forest, and values from approximately 100 up to 300 indicate old-growth forest. These values represent elapsed time since forest establishment, not calendar year.
Common Tasks
Task -> How To
- Check forest age at a point Activate Point Mode, click on the map, and read the value labeled “FAROS Forest Age”
- Analyze forest age over an area Activate Polygon Mode, draw a polygon, and review the returned statistics
- Export forest age data Draw a polygon and click “Download FAROS GeoTIFF”
- Export higher-accuracy estimates Draw a polygon and click “Download FAROS Alpha (RF)”
- View tree cover trends Use Point Mode and click the map to display the trend chart
- Change visible layers Open the Layers panel and toggle layers on or off
- Remove an active polygon Click the “Clear Polygon” button
Interpreting Map Colors
Colors shown on the map are a visualization aid for forest age classes. Red tones represent young or recently disturbed forest (approximately 0–20 years). Yellow indicates mid-age forest (roughly 40–60 years). Green corresponds to mature forest (60–100 years), while darker green represents old-growth forest, typically exceeding 100 years and extending beyond 300 years in some regions.
Export File Characteristics
All exported files are GeoTIFFs at 10 m spatial resolution with pixel values representing years since forest establishment. Files are delivered in WGS84 coordinates. Individual downloads are limited to approximately 32 MB; larger areas are automatically split into multiple GeoTIFFs.
Troubleshooting
If the application does not load and the map remains light colored or blank, allow at least two full minutes for initialization, as loading can be slow. If the issue persists, refresh the page, try a different modern browser such as Chrome, Firefox, or Edge, and confirm that the internet connection is stable.
If clicking on the map does not return values, ensure that Point Mode is active and highlighted, zoom in to at least level 6, and confirm that the selected location is forested. The map background should be fully rendered (dark) before interacting.
If polygon drawing does not begin, verify that Polygon Mode is selected and highlighted. Clearing any existing polygon before drawing a new one often resolves the issue.
If a download does not start, ensure that the polygon is fully closed by double-clicking the final vertex. Very large polygons may require additional processing time or may need to be reduced in size. Browser pop-up blockers should also be checked.
“No data” messages are expected in some regions without baseline forest age coverage. In these areas, the “FAROS Expansion Age” layer may provide useful information for potential forest areas. Forest age data are not available everywhere, and coverage can vary spatially.
FAROS Methods Documentation (Detailed)
Updated to FAROS v3.3; 12/22/25
FAROS (Forest Age Remote Observation System) is a Google Earth Engine–based platform for monitoring global forest age at 10-meter spatial resolution from 2015 to the present. The system integrates multiple forest age baseline datasets with annual satellite observations to track forest persistence, disturbance, and regeneration over time. Processing is automatically updated through the most recent complete year, using current-year data when executed in Q4 (October–December) and the previous year otherwise.
1. Baseline Forest Age Construction (2015)
1.1 Multi-Source Integration
FAROS constructs a global 2015 forest age baseline by integrating multiple independent forest age products using a conservative maximum-value composite approach. For each 10-meter pixel, the system assigns the highest available age estimate across all baseline sources:
Age₍₂₀₁₅₎ = max(GAMI-IPCC Mean, GAMI Interpolation, LandTrendr, Canada NTEMS, Siberian Larch, China Forest Age)
This approach prioritizes the most conservative age estimate where datasets overlap, reducing systematic underestimation of forest age in well-studied regions.
1.2 Baseline Data Sources
Global baseline datasets include GAMI v2.0 (100 m, reference year 2020, globally rescaled to 2015 by subtracting five years), the NASA/ORNL IPCC Forest classification (30 m, reference year 2020, using discrete age assignments of 300 years for primary forest, 30 years for old secondary forest, and 10 years for young secondary forest with a −5 year adjustment), and a GAMI-IPCC composite that computes the mean where both products are available and uses individual values otherwise. LandTrendr (Wang et al. 2025) provides disturbance-based age estimates derived from Landsat trajectories spanning 1985–2024 and back-calculated to 2015.
Regional high-resolution datasets supplement global coverage and include Canada NTEMS (30 m, reference year 2019), Siberian Larch stand age products (~2010), and a 30 m China forest age map (reference year 2020; Cheng et al. 2024).
1.3 GAMI Interpolation: Gap-Filling with Random Forest
To achieve complete spatial coverage, FAROS applies a five-round iterative Random Forest (RF) regression to predict forest age in regions lacking baseline data. The model is trained on GAMI v2.0 forest age pixels using a stratified random sampling scheme across global forest biomes, with accuracy assessed through out-of-bag error estimation.
The model incorporates 81 predictor variables grouped into biomass, climate, topography, soils, human influence, and deep-learning spatial features. Core predictors include ESA CCI Above-Ground Biomass (2015 and 2020) and OpenLandMap precipitation. Topographic variables (elevation, slope, northness, eastness) are derived from SRTM 30 m data. Climate variables include temperature and precipitation metrics from WorldClim, along with evapotranspiration, climatic water deficit, and vapor pressure deficit from TerraClimate. Soil predictors include organic carbon, clay fraction, and pH from OpenLandMap. Spatial context is provided by RESOLVE ecoregions and the Global Human Modification Index. The model further incorporates 64 AlphaEarth satellite embeddings (Bands A00–A63), which capture complex spatial patterns at 10 m resolution using deep learning.
All coarse-resolution predictors are upscaled to 10 m using inverse distance weighting based on the four nearest pixel centers, preserving spatial gradients while avoiding artificial smoothing. Source resolutions range from 1 km (WorldClim) to 4 km (TerraClimate), 250 m–1 km (OpenLandMap), and 30 m (SRTM).
The iterative RF procedure reduces systematic bias. The first round predicts forest age directly from GAMI training data (out-of-bag RMSE ≈ 32 years). Subsequent rounds (2–5) iteratively model residuals from the previous round, progressively reducing bias. Final out-of-bag error is approximately 9 years RMSE, following the GAP residual-learning methodology, with improved performance in heterogeneous landscapes and complex terrain.
2. Annual Forest State Tracking (2016–Present)
2.1 Dynamic World Tree Percentage Analysis
FAROS uses Google Dynamic World Version 1 (10 m) to track annual forest state from 2016 through the current year. Dynamic World provides near-daily land-cover classifications with associated probability layers. For each year, all available observations are aggregated, and the tree probability layer (0–100%) is averaged to produce an annual tree-cover composite.
2.2 Forest State Transition Logic
Forest persistence, disturbance, and regeneration are detected using a conservative consecutive-year threshold approach designed to minimize false positives from clouds, seasonality, or phenological variation. Five thresholds govern state transitions: a persistence threshold of 15% tree cover, a regrowth threshold of 25%, and a requirement of three consecutive years below or above these thresholds to trigger deforestation or reforestation, respectively.
Forest age increments by one year when a pixel remains forested and does not meet loss conditions. Deforestation resets age to zero following three consecutive low-tree years. Reforestation assigns an initial age of three years following three consecutive high-tree years.
2.3 Cloud and Missing Data Handling
To prevent cloud contamination from inducing spurious change detection, FAROS applies asymmetric missing-data logic. For disturbance detection, missing observations are treated as 100% tree cover, breaking loss streaks. For regrowth detection, missing observations are treated as 0% tree cover, preventing false reforestation signals.
3. CCI Land Cover Integration (1992–2015)
3.1 Historical Forest Classification
To extend forest age estimates prior to Dynamic World availability, FAROS incorporates ESA CCI Land Cover annual maps (300 m) for 1992–2015. Forest classes include CCI categories 50, 60, 61, 62, 70, 80, 81, 82, 90, 100, 110, 160, and 170.
3.2 Temporal Bridging Methodology
Forest age tracking from 1992 to present is achieved by combining CCI-based forest masks with the FAROS 2015 baseline and Dynamic World updates. Pixels classified as forest in both CCI and FAROS in 2015 retain their baseline age. Earlier years are back-calculated by subtracting elapsed years (e.g., Age₁₉₉₂ = Age₂₀₁₅ − 23), assuming persistence during periods without annual monitoring. Consistency is validated against the 2015 baseline.
4. Interactive Features
4.1 Point Inspector Mode
The point inspector retrieves pixel-level data at user-selected locations, including current FAROS age with year reference, Dynamic World forest status, and a 10-year time series of age progression. Additional outputs include Dynamic World tree percentage and class labels, ESA WorldCover class, ESA WorldCereal crop classification, and total forest area within a 500 m radius. All baseline age layers are also displayed.
4.2 Polygon Mode
Polygon mode enables custom area-of-interest analysis. Outputs include total area, forested area and percentage, and counts of old-growth forest plots (GAP). Summary statistics include mean FAROS age, mean baseline ages (GAMI, LandTrendr), mean canopy height, and mean ESA CCI biomass. Export options include standard FAROS and Alpha RF outputs.
4.3 Export Capabilities
Standard FAROS exports produce single-band Int16 GeoTIFFs at 10 m resolution in WGS84 coordinates, with pixel values representing years since forest establishment. Files are limited to approximately 32 MB and are automatically tiled for larger areas.
The FAROS Alpha (RF) export trains a localized Random Forest model using 1,000 random points sampled from the selected polygon plus a 10 km buffer. The model uses 85 predictors and a five-round iterative bias-corrected RF. Output specifications match standard FAROS exports, and predicted layers are displayed in-app following model execution.
5. Data Sources
Primary satellite data include Dynamic World V1, AlphaEarth satellite embeddings, ESA WorldCover, ESA WorldCereal, and Sentinel-2 Level-2A imagery. Environmental predictors include WorldClim, TerraClimate, OpenLandMap soils and precipitation, and SRTM topography. Context and validation layers include RESOLVE ecoregions, the Global Human Modification Index, ESA CCI biomass, Meta/WRI canopy height, NatureTrace natural forest probability, forest persistence metrics, and regional forest age datasets.
6. Quality Control and Limitations
FAROS integrates multiple independent forest age products, applies conservative change detection, and incorporates cloud-aware processing. The 10 m spatial resolution and annual temporal updates provide detailed and consistent monitoring.
Limitations include regional variability in baseline quality, an effective maximum age cap near 300 years, approximately 9-year RMSE in gap-filled regions, potential omission of sub-annual disturbances, and occasional misclassification of natural vegetation. The system estimates forest age only and does not provide species-level information.
Validation relies on Random Forest out-of-bag error metrics, cross-validation against held-out baseline datasets, comparison with regional products, visual time-series inspection, and alignment with known disturbance events. Uncertainty quantification is under active development.
7. Technical Implementation
FAROS is implemented entirely within Google Earth Engine. Baseline construction processes approximately 1 TB of global raster data, while annual updates incorporate roughly 10 TB of Dynamic World observations per year. Random Forest models use 81–85 predictors and thousands of training samples. Exports are generated on demand with automatic spatial tiling.
Citation, License, and Disclaimer
Please cite FAROS and associated datasets when using these products. FAROS is distributed under the Creative Commons Attribution 4.0 license (CC BY 4.0). The system is provided as-is for research and educational use. Forest age estimates are derived from multiple sources and contain inherent uncertainty. Users are encouraged to validate outputs for their specific applications.
Acknowledgments
FAROS integrates data and tools from Google Earth Engine, ESA, NASA/ORNL, regional forest age product teams, WorldClim, TerraClimate, OpenLandMap, WRI, and Meta, among others.
Updated to FAROS v3.3; 12/22/25
FAROS (Forest Age Remote Observation System) is a Google Earth Engine–based platform for monitoring global forest age at 10-meter spatial resolution from 2015 to the present. The system integrates multiple forest age baseline datasets with annual satellite observations to track forest persistence, disturbance, and regeneration over time. Processing is automatically updated through the most recent complete year, using current-year data when executed in Q4 (October–December) and the previous year otherwise.
1. Baseline Forest Age Construction (2015)
1.1 Multi-Source Integration
FAROS constructs a global 2015 forest age baseline by integrating multiple independent forest age products using a conservative maximum-value composite approach. For each 10-meter pixel, the system assigns the highest available age estimate across all baseline sources:
Age₍₂₀₁₅₎ = max(GAMI-IPCC Mean, GAMI Interpolation, LandTrendr, Canada NTEMS, Siberian Larch, China Forest Age)
This approach prioritizes the most conservative age estimate where datasets overlap, reducing systematic underestimation of forest age in well-studied regions.
1.2 Baseline Data Sources
Global baseline datasets include GAMI v2.0 (100 m, reference year 2020, globally rescaled to 2015 by subtracting five years), the NASA/ORNL IPCC Forest classification (30 m, reference year 2020, using discrete age assignments of 300 years for primary forest, 30 years for old secondary forest, and 10 years for young secondary forest with a −5 year adjustment), and a GAMI-IPCC composite that computes the mean where both products are available and uses individual values otherwise. LandTrendr (Wang et al. 2025) provides disturbance-based age estimates derived from Landsat trajectories spanning 1985–2024 and back-calculated to 2015.
Regional high-resolution datasets supplement global coverage and include Canada NTEMS (30 m, reference year 2019), Siberian Larch stand age products (~2010), and a 30 m China forest age map (reference year 2020; Cheng et al. 2024).
1.3 GAMI Interpolation: Gap-Filling with Random Forest
To achieve complete spatial coverage, FAROS applies a five-round iterative Random Forest (RF) regression to predict forest age in regions lacking baseline data. The model is trained on GAMI v2.0 forest age pixels using a stratified random sampling scheme across global forest biomes, with accuracy assessed through out-of-bag error estimation.
The model incorporates 81 predictor variables grouped into biomass, climate, topography, soils, human influence, and deep-learning spatial features. Core predictors include ESA CCI Above-Ground Biomass (2015 and 2020) and OpenLandMap precipitation. Topographic variables (elevation, slope, northness, eastness) are derived from SRTM 30 m data. Climate variables include temperature and precipitation metrics from WorldClim, along with evapotranspiration, climatic water deficit, and vapor pressure deficit from TerraClimate. Soil predictors include organic carbon, clay fraction, and pH from OpenLandMap. Spatial context is provided by RESOLVE ecoregions and the Global Human Modification Index. The model further incorporates 64 AlphaEarth satellite embeddings (Bands A00–A63), which capture complex spatial patterns at 10 m resolution using deep learning.
All coarse-resolution predictors are upscaled to 10 m using inverse distance weighting based on the four nearest pixel centers, preserving spatial gradients while avoiding artificial smoothing. Source resolutions range from 1 km (WorldClim) to 4 km (TerraClimate), 250 m–1 km (OpenLandMap), and 30 m (SRTM).
The iterative RF procedure reduces systematic bias. The first round predicts forest age directly from GAMI training data (out-of-bag RMSE ≈ 32 years). Subsequent rounds (2–5) iteratively model residuals from the previous round, progressively reducing bias. Final out-of-bag error is approximately 9 years RMSE, following the GAP residual-learning methodology, with improved performance in heterogeneous landscapes and complex terrain.
2. Annual Forest State Tracking (2016–Present)
2.1 Dynamic World Tree Percentage Analysis
FAROS uses Google Dynamic World Version 1 (10 m) to track annual forest state from 2016 through the current year. Dynamic World provides near-daily land-cover classifications with associated probability layers. For each year, all available observations are aggregated, and the tree probability layer (0–100%) is averaged to produce an annual tree-cover composite.
2.2 Forest State Transition Logic
Forest persistence, disturbance, and regeneration are detected using a conservative consecutive-year threshold approach designed to minimize false positives from clouds, seasonality, or phenological variation. Five thresholds govern state transitions: a persistence threshold of 15% tree cover, a regrowth threshold of 25%, and a requirement of three consecutive years below or above these thresholds to trigger deforestation or reforestation, respectively.
Forest age increments by one year when a pixel remains forested and does not meet loss conditions. Deforestation resets age to zero following three consecutive low-tree years. Reforestation assigns an initial age of three years following three consecutive high-tree years.
2.3 Cloud and Missing Data Handling
To prevent cloud contamination from inducing spurious change detection, FAROS applies asymmetric missing-data logic. For disturbance detection, missing observations are treated as 100% tree cover, breaking loss streaks. For regrowth detection, missing observations are treated as 0% tree cover, preventing false reforestation signals.
3. CCI Land Cover Integration (1992–2015)
3.1 Historical Forest Classification
To extend forest age estimates prior to Dynamic World availability, FAROS incorporates ESA CCI Land Cover annual maps (300 m) for 1992–2015. Forest classes include CCI categories 50, 60, 61, 62, 70, 80, 81, 82, 90, 100, 110, 160, and 170.
3.2 Temporal Bridging Methodology
Forest age tracking from 1992 to present is achieved by combining CCI-based forest masks with the FAROS 2015 baseline and Dynamic World updates. Pixels classified as forest in both CCI and FAROS in 2015 retain their baseline age. Earlier years are back-calculated by subtracting elapsed years (e.g., Age₁₉₉₂ = Age₂₀₁₅ − 23), assuming persistence during periods without annual monitoring. Consistency is validated against the 2015 baseline.
4. Interactive Features
4.1 Point Inspector Mode
The point inspector retrieves pixel-level data at user-selected locations, including current FAROS age with year reference, Dynamic World forest status, and a 10-year time series of age progression. Additional outputs include Dynamic World tree percentage and class labels, ESA WorldCover class, ESA WorldCereal crop classification, and total forest area within a 500 m radius. All baseline age layers are also displayed.
4.2 Polygon Mode
Polygon mode enables custom area-of-interest analysis. Outputs include total area, forested area and percentage, and counts of old-growth forest plots (GAP). Summary statistics include mean FAROS age, mean baseline ages (GAMI, LandTrendr), mean canopy height, and mean ESA CCI biomass. Export options include standard FAROS and Alpha RF outputs.
4.3 Export Capabilities
Standard FAROS exports produce single-band Int16 GeoTIFFs at 10 m resolution in WGS84 coordinates, with pixel values representing years since forest establishment. Files are limited to approximately 32 MB and are automatically tiled for larger areas.
The FAROS Alpha (RF) export trains a localized Random Forest model using 1,000 random points sampled from the selected polygon plus a 10 km buffer. The model uses 85 predictors and a five-round iterative bias-corrected RF. Output specifications match standard FAROS exports, and predicted layers are displayed in-app following model execution.
5. Data Sources
Primary satellite data include Dynamic World V1, AlphaEarth satellite embeddings, ESA WorldCover, ESA WorldCereal, and Sentinel-2 Level-2A imagery. Environmental predictors include WorldClim, TerraClimate, OpenLandMap soils and precipitation, and SRTM topography. Context and validation layers include RESOLVE ecoregions, the Global Human Modification Index, ESA CCI biomass, Meta/WRI canopy height, NatureTrace natural forest probability, forest persistence metrics, and regional forest age datasets.
6. Quality Control and Limitations
FAROS integrates multiple independent forest age products, applies conservative change detection, and incorporates cloud-aware processing. The 10 m spatial resolution and annual temporal updates provide detailed and consistent monitoring.
Limitations include regional variability in baseline quality, an effective maximum age cap near 300 years, approximately 9-year RMSE in gap-filled regions, potential omission of sub-annual disturbances, and occasional misclassification of natural vegetation. The system estimates forest age only and does not provide species-level information.
Validation relies on Random Forest out-of-bag error metrics, cross-validation against held-out baseline datasets, comparison with regional products, visual time-series inspection, and alignment with known disturbance events. Uncertainty quantification is under active development.
7. Technical Implementation
FAROS is implemented entirely within Google Earth Engine. Baseline construction processes approximately 1 TB of global raster data, while annual updates incorporate roughly 10 TB of Dynamic World observations per year. Random Forest models use 81–85 predictors and thousands of training samples. Exports are generated on demand with automatic spatial tiling.
Citation, License, and Disclaimer
Please cite FAROS and associated datasets when using these products. FAROS is distributed under the Creative Commons Attribution 4.0 license (CC BY 4.0). The system is provided as-is for research and educational use. Forest age estimates are derived from multiple sources and contain inherent uncertainty. Users are encouraged to validate outputs for their specific applications.
Acknowledgments
FAROS integrates data and tools from Google Earth Engine, ESA, NASA/ORNL, regional forest age product teams, WorldClim, TerraClimate, OpenLandMap, WRI, and Meta, among others.
Supporting tools
Alpha Earth Explorer
In our ongoing efforts to use Google Alpha Earth to improve the accuracy and coverage of our FAROS forest age map, we developed the Alpha Earth Explorer web app. Perhaps you will find it useful to understand this amazing dataset. If you do find it useful, please cite as: Broadbent, EN; Almeyda Zambrano AM; SPECLab Alpha Earth Explorer - accessed on date.
Web app link
Alpha Earth Explorer
In our ongoing efforts to use Google Alpha Earth to improve the accuracy and coverage of our FAROS forest age map, we developed the Alpha Earth Explorer web app. Perhaps you will find it useful to understand this amazing dataset. If you do find it useful, please cite as: Broadbent, EN; Almeyda Zambrano AM; SPECLab Alpha Earth Explorer - accessed on date.
Web app link
