Forest change
Forest change data measure tree cover loss, tree cover gain, or forest disturbance.
-
FORMA GEOGRAPHIC COVERAGE
This layer displays the geographic coverage of FORMA alerts, which corresponds to the coverage of tropical and subtropical moist broadleaf forests and tropical and subtropical dry broadleaf forests according to WWF ecoregions.
-
FORMA GEOGRAPHIC COVERAGE
This layer displays the geographic coverage of FORMA alerts, which corresponds to the coverage of tropical and subtropical moist broadleaf forests and tropical and subtropical dry broadleaf forests according to WWF ecoregions.
-
Tree Cover Canopy Density Settings
Drag the handle to adjust the minimum tree cover canopy (TCC) density for the visualization and analysis of Hansen/UMD/Google/USGS/NASA tree cover and tree cover loss. TCC density represents the estimated percent of a pixel that was covered by tree canopy in the year 2000, as determined from the analysis of satellite imagery. For the tree cover loss data, TCC density therefore corresponds to the density of tree cover before loss occurred. For example, if you select 25% as the minimum TCC density, you will only see tree cover loss pixels for which the original tree cover density was greater than 25%.
Adjustments to the minimum TCC density only affect the tree cover and tree cover loss data layers. This feature does not pertain to Hansen/UMD/Google/USGS/NASA tree cover gain or to other GFW data layers or statistics. Tree cover gain is displayed with a set minimum TCC density greater than 50%. The minimum TCC density cannot be changed independently for tree cover and tree cover loss. A change made to one data layer will immediately take effect in the other.
This feature is also available for statistics within the Country Profiles & Rankings. However, the adjustment made to the visualization and analysis through the map view will not be automatically reflected in other areas of the website. To adjust the minimum TCC density within the Country Profiles & Rankings pages, click on the settings icon.
-
Tree cover loss is not always deforestation
Loss of tree cover may occur for many reasons, including deforestation, fire, and logging within the course of sustainable forestry operations. In sustainably managed forests, the “loss” will eventually show up as “gain”, as young trees get large enough to achieve canopy closure.
-
Tree cover loss (annual, 30m, global, Hansen/UMD/Google/USGS/NASA)
- Function
- Identifies areas of gross tree cover loss
- RESOLUTION / SCALE
- 30 × 30 meters
- Geographic coverage
- Global land area (excluding Antarctica and other Arctic islands)
- Source data
- Landsat
- Frequency of updates
- Annual
- Date of content
- 2001–2014
- Tree cover canopy density
- Varies according to selection (use the legend on the map to change the minimum tree cover canopy density threshold)
- Cautions
-
This data layer was updated in January 2015 to extend the tree cover loss analysis to 2013, and in August 2015 to extend the tree cover loss analysis to 2014. The updates include new data for the target year and re-processed data for the previous two years (2011 and 2012 for the 2013 update, 2012 and 2013 for the 2014 update). The re-processing increased the amount of change that could be detected, resulting in some changes in calculated tree cover loss for 2011-2013 compared to the previous versions. Calculated tree cover loss for 2001-2010 remains unchanged. The integrated use of the original 2001-2012 (Version 1.0) data and the updated 2011–2014 (Version 1.1) data should be performed with caution.
For the purpose of this study, “tree cover” was defined as all vegetation taller than 5 meters in height. “Tree cover” is the biophysical presence of trees and may take the form of natural forests or plantations existing over a range of canopy densities. “Loss” indicates the removal or mortality of tree canopy cover and can be due to a variety of factors, including mechanical harvesting, fire, disease, or storm damage. As such, “loss” does not equate to deforestation.
When zoomed out (< zoom level 13), pixels of loss are shaded according to the density of loss at the 30 x 30 meter scale. Pixels with darker shading represent areas with a higher concentration of tree cover loss, whereas pixels with lighter shading indicate a lower concentration of tree cover loss. There is no variation in pixel shading when the data is at full resolution (≥ zoom level 13).
Overview
This data set measures areas of tree cover loss across all global land (except Antarctica and other Arctic islands) at approximately 30 × 30 meter resolution. The data were generated using multispectral satellite imagery from the Landsat 5 thematic mapper (TM), the Landsat 7 thematic mapper plus (ETM+), and the Landsat 8 Operational Land Imager (OLI) sensors. Over 1 million satellite images were processed and analyzed, including over 600,000 Landsat 7 images for the 2000-2012 interval, and approximately 400,000 Landsat 5, 7, and 8 images for updates for the 2011-2014 interval. The clear land surface observations in the satellite images were assembled and a supervised learning algorithm was applied to identify per pixel tree cover loss.
Tree cover loss is defined as “stand replacement disturbance,” or the complete removal of tree cover canopy at the Landsat pixel scale. Tree cover loss may be the result of human activities, including forestry practices such as timber harvesting or deforestation (the conversion of natural forest to other land uses), as well as natural causes such as disease or storm damage. Fire is another widespread cause of tree cover loss, and can be either natural or human-induced.
2015 Update (Version 1.1)
This data set has been updated twice since its creation, and now includes loss up to 2014. The analysis method has been modified in numerous ways, and the update should be seen as part of a transition to a future “version 2.0” of this data set that is more consistent over the entire 2001 and onward period. Key changes include:
- The use of Landsat 8 data for 2013-2014 and Landsat 5 data for 2011-2012
- The reprocessing of data from the previous two years in measuring loss (2011 and 2012 for the 2013 update, 2012 and 2013 for the 2014 update)
- Improved training data for calibrating the loss model
- Improved per sensor quality assessment models to filter input data
- Improved input spectral features for building and applying the loss model
These changes lead to a different and improved detection of global tree cover loss. However, the years preceding 2011 have not yet been reprocessed with the revised analysis methods, and users will notice inconsistencies between versions 1.0 (2001-2012) and 1.1 (2011-2014) as a result. It must also be noted that a full validation of the results incorporating Landsat 8 has not been undertaken. Such an analysis may reveal a more sensitive ability to detect and map forest disturbance using Landsat 8 data. If this is the case then there will be a more fundamental limitation to the consistency of this data set before and after the inclusion of Landsat 8 data. Validation of Landsat 8-incorporated loss detection is planned.
Some examples of improved change detection in the 2011–2014 update include the following:
- Improved detection of boreal forest loss due to fire
- Improved detection of smallholder rotation agricultural clearing in dry and humid tropical forests
- Improved detection of selective logging
These are examples of dynamics that may be differentially mapped over the 2011-2014 period in Version 1.1. A version 2.0 reprocessing of the 2001 and onward record is planned, but no delivery date is yet confirmed.
The original version 1.0 data remains available for download here.
Citation: Hansen, M. C., P. V. Potapov, R. Moore, M. Hancher, S. A. Turubanova, A. Tyukavina, D. Thau, S. V. Stehman, S. J. Goetz, T. R. Loveland, A. Kommareddy, A. Egorov, L. Chini, C. O. Justice, and J. R. G. Townshend. 2013. “High-Resolution Global Maps of 21st-Century Forest Cover Change.” Science 342 (15 November): 850–53. Data available online from: http://earthenginepartners.appspot.com/science-2013-global-forest.
Hansen, M. C., P. V. Potapov, R. Moore, M. Hancher, S. A. Turubanova, A. Tyukavina, D. Thau, S. V. Stehman, S. J. Goetz, T. R. Loveland, A. Kommareddy, A. Egorov, L. Chini, C. O. Justice, and J. R. G. Townshend. 2013. “Hansen/UMD/Google/USGS/NASA Tree Cover Loss and Gain Area.” University of Maryland, Google, USGS, and NASA. Accessed through Global Forest Watch on [date]. www.globalforestwatch.org.
-
PRODES deforestation (annual, 30m, Brazilian Amazon, INPE)
- Function
- Deforestation monitoring system for the Brazilian Amazon used by the Brazilian government to establish public policy
- RESOLUTION / SCALE
- 30 × 30 meters, minimum mapping unit of 6.25 hectares
- Geographic coverage
- Brazilian Amazon
- Source data
- Landsat, supplemented with CBERS, Resourcesat, and UK2-DMC
- Frequency of updates
- Annual
- Date of content
- 2001–2015
- Cautions
-
PRODES only identifies forest clearings of 6.25 hectares or larger, so forest degradation or smaller clearings from fire or selective logging are not detected.
Frequent cloud cover over areas of the Amazon may change the reported year of deforestation. The year reported is the first year deforestation is identified by analysts, but this does not necessarily correspond to the year of deforestation if the landscape has been covered by clouds in previous years.
- License
- Creative Commons BY SA 3.0
Overview
The PRODES project monitors clear cut deforestation in the Brazilian Legal Amazon, and has produced annual deforestation rates for the region since 1988. The Brazilian government uses these figures to establish public policy, including defining access to credit in the Amazon biome, establishing deforestation reduction goals, and soliciting funds to reduce deforestation. PRODES historically used Landsat 5 images, but now also incorporates imagery from Landsat 7 and 8, CBERS-2, CBERS-2B, Resourcesat-1, and UK2-DMC. PRODES is operated by the National Institute of Space Research (INPE) in collaboration with the Ministry of the Environment (MMA) and the Brazilian Institute of Environment and Renewable Natural Resources (IBAMA). Since 2002, all PRODES data is publicly available online.
Input images for each of the 220 Landsat footprints that cover the Brazilian Amazon are selected based on their lack of cloud cover and their capture date. The PRODES system uses the seasonal year, starting on August 1st, to calculate annual deforestation, so images are selected as near to this date as possible (generally from July, August, and September). From 2003 to 2005, analysts used image transformation to determine the components of vegetation, soil, and shadow using the program SPRING. These components were segmented and classified into the classes of forest, non-forest, deforestation in the target year, previous deforestation, clouds, and water, which are then manually corrected by experts. Starting in 2005, a new methodology was implemented which makes use of the open source TerraAmazon platform. The platform allows the PRODES analysis to be more uniform and can incorporate imagery from a variety of satellites. As before, images are selected to be as cloud free as possible. The images are then masked to exclude non-forest, previous deforestation, and water using the previous year’s analysis. Analysts then delineate deforested polygons in the intact forest of the previous year. More information on the methodology can be found on the PRODES website.
This data set shows annual deforestation between 2001 and 2015.
National Institute of Space Research (INPE). “PRODES.” 2015.
-
PRODES geographic coverage
This layer displays the geographic coverage of PRODES monitoring, which only covers the Brazilian Amazon.
-
Tree cover gain (12 years, 30m, global, Hansen/UMD/Google/USGS/NASA)
- Function
- Identifies areas of tree cover gain
- RESOLUTION / SCALE
- 30 × 30 meters
- Geographic coverage
- Global land area (excluding Antarctica and other Arctic islands)
- Source data
- Landsat 7 ETM+
- Frequency of updates
- Every three years
- Date of content
- 2001–2012
- Tree cover canopy density
- >50%
- Cautions
-
For the purpose of this study, “tree cover” was defined as all vegetation taller than 5 meters in height. “Tree cover” is the biophysical presence of trees and may take the form of natural forests or plantations existing over a range of canopy densities. “Loss” indicates the removal or mortality of tree canopy cover and can be due to a variety of factors, including mechanical harvesting, fire, disease, or storm damage. As such, “loss” does not equate to deforestation.
When zoomed out (< zoom level 13), pixels of gain are shaded according to the density of gain at the 30 x 30 meter scale. Pixels with darker shading represent areas with a higher concentration of tree cover gain, whereas pixels with lighter shading indicate a lower concentration of tree cover gain. There is no variation in pixel shading when the data is at full resolution (≥ zoom level 13).
A validation assessment of the 2000 – 2012 Hansen/UMD/Google/USGS/NASA change data was carried out independently from the mapping exercise at the global and biome (tropical, subtropical, temperate, and boreal) scale. A stratified random sample (for no change, loss, and gain) of 1500 blocks, each 120 × 120 meters, was used as validation data. The amount of tree cover gain within each block was estimated using Landsat, MODIS, and Google Earth high-resolution imagery and compared to the map. Overall accuracies for gain were over 99.5% globally and for all biomes. However, since the overall accuracy calculations are positively skewed due to the high percentage of no change pixels, it is also important to assess the accuracy of the change predictions. The user’s accuracy (i.e. the percentage of pixels labelled as tree cover gain that actually gained tree cover) was 87.8% at the global level. At the biome level, user’s accuracies were 81.9%, 85.5%, 62.0%, and 76.7% for the tropical, subtropical, temperate, and boreal biomes, respectively.
Overview
This data set measures areas of tree cover gain across all global land (except Antarctica and other Arctic islands) at 30 × 30 meter resolution, displayed as a 12-year cumulative layer. The data were generated using multispectral satellite imagery from the Landsat 7 thematic mapper plus (ETM+) sensor. Over 600,000 Landsat 7 images were compiled and analyzed using Google Earth Engine, a cloud platform for earth observation and data analysis. The clear land surface observations (30 × 30 meter pixels) in the satellite images were assembled and a supervised learning algorithm was then applied to identify per pixel tree cover gain.
Tree cover gain was defined as the establishment of tree canopy at the Landsat pixel scale in an area that previously had no tree cover. Tree cover gain may indicate a number of potential activities, including natural forest growth or the crop rotation cycle of tree plantations.
Citation: Hansen, M. C., P. V. Potapov, R. Moore, M. Hancher, S. A. Turubanova, A. Tyukavina, D. Thau, S. V. Stehman, S. J. Goetz, T. R. Loveland, A. Kommareddy, A. Egorov, L. Chini, C. O. Justice, and J. R. G. Townshend. 2013. “High-Resolution Global Maps of 21st-Century Forest Cover Change.” Science 342 (15 November): 850–53. Data available on-line from: http://earthenginepartners.appspot.com/science-2013-global-forest.
Suggested citations for data as displayed on GFW: Hansen, M. C., P. V. Potapov, R. Moore, M. Hancher, S. A. Turubanova, A. Tyukavina, D. Thau, S. V. Stehman, S. J. Goetz, T. R. Loveland, A. Kommareddy, A. Egorov, L. Chini, C. O. Justice, and J. R. G. Townshend. 2013. “Hansen/UMD/Google/USGS/NASA Tree Cover Loss and Gain Area.” University of Maryland, Google, USGS, and NASA. Accessed through Global Forest Watch on [date]. www.globalforestwatch.org.
-
Tree cover loss (annual, 30m, global, Hansen/UMD/Google/USGS/NASA)
- Function
- Identifies areas of gross tree cover loss
- RESOLUTION / SCALE
- 30 × 30 meters
- Geographic coverage
- Global land area (excluding Antarctica and other Arctic islands)
- Source data
- Landsat
- Frequency of updates
- Annual
- Date of content
- 2001–2014
- Tree cover canopy density
- Varies according to selection (use the legend on the map to change the minimum tree cover canopy density threshold)
- Cautions
-
This data layer was updated in January 2015 to extend the tree cover loss analysis to 2013, and in August 2015 to extend the tree cover loss analysis to 2014. The updates include new data for the target year and re-processed data for the previous two years (2011 and 2012 for the 2013 update, 2012 and 2013 for the 2014 update). The re-processing increased the amount of change that could be detected, resulting in some changes in calculated tree cover loss for 2011-2013 compared to the previous versions. Calculated tree cover loss for 2001-2010 remains unchanged. The integrated use of the original 2001-2012 (Version 1.0) data and the updated 2011–2014 (Version 1.1) data should be performed with caution.
For the purpose of this study, “tree cover” was defined as all vegetation taller than 5 meters in height. “Tree cover” is the biophysical presence of trees and may take the form of natural forests or plantations existing over a range of canopy densities. “Loss” indicates the removal or mortality of tree canopy cover and can be due to a variety of factors, including mechanical harvesting, fire, disease, or storm damage. As such, “loss” does not equate to deforestation.
When zoomed out (< zoom level 13), pixels of loss are shaded according to the density of loss at the 30 x 30 meter scale. Pixels with darker shading represent areas with a higher concentration of tree cover loss, whereas pixels with lighter shading indicate a lower concentration of tree cover loss. There is no variation in pixel shading when the data is at full resolution (≥ zoom level 13).
Overview
This data set measures areas of tree cover loss across all global land (except Antarctica and other Arctic islands) at approximately 30 × 30 meter resolution. The data were generated using multispectral satellite imagery from the Landsat 5 thematic mapper (TM), the Landsat 7 thematic mapper plus (ETM+), and the Landsat 8 Operational Land Imager (OLI) sensors. Over 1 million satellite images were processed and analyzed, including over 600,000 Landsat 7 images for the 2000-2012 interval, and approximately 400,000 Landsat 5, 7, and 8 images for updates for the 2011-2014 interval. The clear land surface observations in the satellite images were assembled and a supervised learning algorithm was applied to identify per pixel tree cover loss.
Tree cover loss is defined as “stand replacement disturbance,” or the complete removal of tree cover canopy at the Landsat pixel scale. Tree cover loss may be the result of human activities, including forestry practices such as timber harvesting or deforestation (the conversion of natural forest to other land uses), as well as natural causes such as disease or storm damage. Fire is another widespread cause of tree cover loss, and can be either natural or human-induced.
2015 Update (Version 1.1)
This data set has been updated twice since its creation, and now includes loss up to 2014. The analysis method has been modified in numerous ways, and the update should be seen as part of a transition to a future “version 2.0” of this data set that is more consistent over the entire 2001 and onward period. Key changes include:
- The use of Landsat 8 data for 2013-2014 and Landsat 5 data for 2011-2012
- The reprocessing of data from the previous two years in measuring loss (2011 and 2012 for the 2013 update, 2012 and 2013 for the 2014 update)
- Improved training data for calibrating the loss model
- Improved per sensor quality assessment models to filter input data
- Improved input spectral features for building and applying the loss model
These changes lead to a different and improved detection of global tree cover loss. However, the years preceding 2011 have not yet been reprocessed with the revised analysis methods, and users will notice inconsistencies between versions 1.0 (2001-2012) and 1.1 (2011-2014) as a result. It must also be noted that a full validation of the results incorporating Landsat 8 has not been undertaken. Such an analysis may reveal a more sensitive ability to detect and map forest disturbance using Landsat 8 data. If this is the case then there will be a more fundamental limitation to the consistency of this data set before and after the inclusion of Landsat 8 data. Validation of Landsat 8-incorporated loss detection is planned.
Some examples of improved change detection in the 2011–2014 update include the following:
- Improved detection of boreal forest loss due to fire
- Improved detection of smallholder rotation agricultural clearing in dry and humid tropical forests
- Improved detection of selective logging
These are examples of dynamics that may be differentially mapped over the 2011-2014 period in Version 1.1. A version 2.0 reprocessing of the 2001 and onward record is planned, but no delivery date is yet confirmed.
The original version 1.0 data remains available for download here.
Citation: Hansen, M. C., P. V. Potapov, R. Moore, M. Hancher, S. A. Turubanova, A. Tyukavina, D. Thau, S. V. Stehman, S. J. Goetz, T. R. Loveland, A. Kommareddy, A. Egorov, L. Chini, C. O. Justice, and J. R. G. Townshend. 2013. “High-Resolution Global Maps of 21st-Century Forest Cover Change.” Science 342 (15 November): 850–53. Data available online from: http://earthenginepartners.appspot.com/science-2013-global-forest.
Hansen, M. C., P. V. Potapov, R. Moore, M. Hancher, S. A. Turubanova, A. Tyukavina, D. Thau, S. V. Stehman, S. J. Goetz, T. R. Loveland, A. Kommareddy, A. Egorov, L. Chini, C. O. Justice, and J. R. G. Townshend. 2013. “Hansen/UMD/Google/USGS/NASA Tree Cover Loss and Gain Area.” University of Maryland, Google, USGS, and NASA. Accessed through Global Forest Watch on [date]. www.globalforestwatch.org.
-
GLAD Tree Cover Loss Alerts (weekly, 30m, select countries, UMD/ GLAD)
- Function
- Identifies areas of likely tree cover loss in near-real time
- RESOLUTION / SCALE
- 30 × 30 meters
- Geographic coverage
- Peru, Republic of Congo, and Kalimantan (Indonesia). In the future will expand to cover the humid tropics
- Source
- Hansen, M.C., Krylov, A., Tyukavina, A., Potapov, P.V., Turubanova, S., Zutta, B., Ifo, S., Margono, B., and Stolle, F. Near-real time forest disturbance alerts using Landsat data. Environmental Research Letters, in review.
- Frequency of updates
- Updated weekly
- Date of content
- January 2015 - present
- Cautions
While Landsat 7 and 8 satellites together have a revisit period of 8 days, cloud cover can majorly limit the availability of imagery, particularly in the wet season. Alert dates represent the instance of detection, though tree cover loss could have taken place weeks earlier due to persistent cloud cover.
In this data set, “tree cover” is defined as all vegetation greater than 5 meters in height with greater than 60% canopy cover, and may take the form of natural forests or plantations. “Loss” indicates the removal or mortality of >=50% of tree cover and can be due to a variety of factors, including mechanical harvesting, fire, disease, or storm damage. As such, “loss” does not equate to deforestation.
In Peru, where the alert system was first prototyped, the authors evaluated the data to have 13.5% commission error (false positives), though the value drops to 1% when excluding pixels on the boundaries of other loss and unconfirmed alerts. The data has 33% omission errors, though most of these occur in secondary forests where tree cover and other vegetation are more difficult to distinguish. Omission errors in primary forests fall to 17% when excluding pixels on the boundaries of other loss, suggesting that most of the omissions happen on the edges of loss patches.
This data set, created by the GLAD (Global Land Analysis & Discovery) lab at the University of Maryland and supported by Global Forest Watch, is the first Landsat-based alert system for tree cover loss. While most existing loss alert products use 250-meter resolution MODIS imagery, these alerts have a 30-meter resolution and thus can detect loss at a much finer spatial scale. The alerts are currently operational for Peru, the Republic of Congo, and Kalimantan in Indonesia, and will eventually be expanded to the rest of the humid tropics.
New Landsat 7 and 8 images are downloaded as they are posted online at USGS EROS, assessed for cloud cover or poor data quality, and compared to the three previous years of Landsat-derived metrics (including ranks, means, and regressions of red, infrared and shortwave bands, and ranks of NDVI, NBR, and NDWI). The metrics and the latest Landsat image are run through seven decision trees to calculate a median probability of forest disturbance. Pixels with probability >50% are reported as tree cover loss alerts. Alerts remain unconfirmed until two or more out of four consecutive observations are labelled as tree cover loss. For more information on methodology, see the forthcoming Environmental Research Letters paper.
Citation for displaying data: GLAD/UMD, accessed through Global Forest Watch
Citation: Hansen, M.C., Krylov, A., Tyukavina, A., Potapov, P.V., Turubanova, S., Zutta, B., Ifo, S., Margono, B., and Stolle, F., Near-real time forest disturbance alerts using Landsat data. Environmental Research Letters, in review. Accessed through Global Forest Watch on [date]. www.globalforestwatch.org
-
GLAD Tree Cover Loss Alerts (weekly, 30m, select countries, UMD/ GLAD)
- Function
- Identifies areas of likely tree cover loss in near-real time
- RESOLUTION / SCALE
- 30 × 30 meters
- Geographic coverage
- Peru, Republic of Congo, and Kalimantan (Indonesia). In the future will expand to cover the humid tropics
- Source
- Hansen, M.C., Krylov, A., Tyukavina, A., Potapov, P.V., Turubanova, S., Zutta, B., Ifo, S., Margono, B., and Stolle, F. Near-real time forest disturbance alerts using Landsat data. Environmental Research Letters, in review.
- Frequency of updates
- Updated weekly
- Date of content
- January 2015 - present
- Cautions
While Landsat 7 and 8 satellites together have a revisit period of 8 days, cloud cover can majorly limit the availability of imagery, particularly in the wet season. Alert dates represent the instance of detection, though tree cover loss could have taken place weeks earlier due to persistent cloud cover.
In this data set, “tree cover” is defined as all vegetation greater than 5 meters in height with greater than 60% canopy cover, and may take the form of natural forests or plantations. “Loss” indicates the removal or mortality of >=50% of tree cover and can be due to a variety of factors, including mechanical harvesting, fire, disease, or storm damage. As such, “loss” does not equate to deforestation.
In Peru, where the alert system was first prototyped, the authors evaluated the data to have 13.5% commission error (false positives), though the value drops to 1% when excluding pixels on the boundaries of other loss and unconfirmed alerts. The data has 33% omission errors, though most of these occur in secondary forests where tree cover and other vegetation are more difficult to distinguish. Omission errors in primary forests fall to 17% when excluding pixels on the boundaries of other loss, suggesting that most of the omissions happen on the edges of loss patches.
This data set, created by the GLAD (Global Land Analysis & Discovery) lab at the University of Maryland and supported by Global Forest Watch, is the first Landsat-based alert system for tree cover loss. While most existing loss alert products use 250-meter resolution MODIS imagery, these alerts have a 30-meter resolution and thus can detect loss at a much finer spatial scale. The alerts are currently operational for Peru, the Republic of Congo, and Kalimantan in Indonesia, and will eventually be expanded to the rest of the humid tropics.
New Landsat 7 and 8 images are downloaded as they are posted online at USGS EROS, assessed for cloud cover or poor data quality, and compared to the three previous years of Landsat-derived metrics (including ranks, means, and regressions of red, infrared and shortwave bands, and ranks of NDVI, NBR, and NDWI). The metrics and the latest Landsat image are run through seven decision trees to calculate a median probability of forest disturbance. Pixels with probability >50% are reported as tree cover loss alerts. Alerts remain unconfirmed until two or more out of four consecutive observations are labelled as tree cover loss. For more information on methodology, see the forthcoming Environmental Research Letters paper.
Citation for displaying data: GLAD/UMD, accessed through Global Forest Watch
Citation: Hansen, M.C., Krylov, A., Tyukavina, A., Potapov, P.V., Turubanova, S., Zutta, B., Ifo, S., Margono, B., and Stolle, F., Near-real time forest disturbance alerts using Landsat data. Environmental Research Letters, in review. Accessed through Global Forest Watch on [date]. www.globalforestwatch.org
-
GLAD ALERTS GEOGRAPHIC COVERAGE
This layer displays the geographic coverage of GLAD alerts.
-
SHOW ONLY CONFIRMED ALERTS
Alerts become confirmed when a second satellite pass has also identified the pixel as an alert. Most of the alerts that are not confirmed have not had another satellite pass, due to the 8-day revisit time or cloud cover.
-
2015 DATA COVERAGE
This layer shows the areas of the world that have updated tree cover loss data for 2015, including Latin America, Africa, southeast Asia, and Oceania. Other areas of the world currently only have data available until 2014, but will be updated to include 2015 tree cover loss in the coming months.
-
Gran Chaco deforestation (monthly, 30m, Gran Chaco, Guyra)
- Function
- Identifies monthly changes in forest cover in the Gran Chaco region of Paraguay, Argentina, and Bolivia.
Identifica cambios de cobertura forestal mensuales en el Gran Chaco Americano de Paraguay, Argentina, y Bolivia
- RESOLUTION / SCALE
- 30 × 30 meters
- Geographic coverage
- Gran Chaco (see “Geographic coverage” layer in the legend for more information)
- Source data
- Guyra Paraguay
- Frequency of updates
- Monthly
Mensual
- Date of content
- 2011 – present
- Cautions
The data does not distinguish between primary and regenerated forests. Regenerated areas that appear as forests in the previous two years and experience change will be marked as deforestation.
Plots are identified as deforestation in the month when they are first identified. If imagery for previous months were poor quality or covered by clouds, the actual deforestation event may have happened before the reported month.
Los datos no distinguen entre bosque primario y bosque regenerado. Áreas regeneradas que parecen como bosques en los últimos dos años y cambian son marcadas como deforestación.
Parcelas son identificadas como deforestación en el mes cuando están identificadas por primera vez. Si las imágenes de los meses anteriores eran de pobre calidad o tenían cobertura de nubes, la deforestación real pudiera ser antes del mes declarado.
- License
- CC BY 4.0
Overview
The tropical dry forests of the Gran Chaco region in Paraguay, Argentina, and Bolivia have become a hotspot of deforestation as cattle ranching and soy expand into the area. Deforestation monitoring in the Gran Chaco has been carried out by the non-profit Guyra Paraguay since 2011, using 30-meter resolution Landsat images for the 55 scenes that cover the Gran Chaco. The interpretation of forest change areas is done through multi-temporal analysis, which uses a base image from the last two years and a current image from the study month. Analysts use visual interpretation techniques to identify deforestation, including elements of tone, shape, size, texture, pattern, shadow, and context.
How to cite this data: Guyra Paraguay. “Gran Chaco deforestation.” Accessed through Global Forest Watch on [date]. www.globalforestwatch.org
Resumen
Los bosques secos del Gran Chaco Americano en Paraguay, Argentina, y Bolivia se han vuelto un punto focal de la deforestación, mientras la ganadería y la soya se expanden en el área. El monitoreo de deforestación en el Gran Chaco Americano es realizado por Guyra Paraguay desde el año 2011, utilizando imágenes del satélite Landsat, de 30 metros de resolución espacial, para las 55 escenas que cubren el Gran Chaco. La interpretación de las áreas de cambios de bosque es lograda por medio de un análisis multi-temporal, en la que son analizadas dos imágenes, una base de los últimos dos años y una actual correspondiente al mes de estudio. Los analistas usan técnicas de interpretación visual para identificar deforestación, incluyendo elementos como tono, forma, tamaño, textura, patrón, sombra, y asociación.
Cita: Guyra Paraguay. “Deforestación del Gran Chaco.” Accedido a través de Global Forest Watch. www.globalforestwatch.org
-
GRAN CHACO DEFORESTATION GEOGRAPHIC COVERAGE
This layer displays the geographic coverage of the Gran Chaco deforestation data.
-
Forma alerts (quarterly, 500m, humid tropics, WRI/CGD)
- Function
- Detects areas where tree cover loss is likely to have recently occurred
- RESOLUTION / SCALE
- 500 × 500 meters
- Geographic coverage
- Humid tropical forest biome
- Source data
- MODIS
- Frequency of updates
- Quarterly
- Date of content
- January 2006-present
- Cautions
-
Viewing data for the first period available via the time slider may include alerts that reflect detections from the February 2000 to December 2005 training period as the algorithm "catches up" post-training.
The GFW team has clipped out some data in regions where data quality is suspected to be low because of persistent cloud cover over identified ecoregions. This covers parts of Liberia, Venezuela, Guyana, Vietnam, Laos, and Burma/Myanmar. The current extent of the FORMA alerts is available for viewing here—FORMA geographic extent.
The algorithm behind the FORMA alerts is constantly evolving to fix bugs and improve accuracy. As a result, what appears on the site and the results of analyses conducted on the site may change over time. It is important to always include the access date when citing FORMA. Analysis of FORMA alerts on the map should be considered definitive, as analysis is based on raw FORMA data, whereas the data on the map are optimized for visual display. For questions, please join the GFW discussion forum or email us.
For the purpose of this study, “tree cover” was defined as areas with greater than 25% canopy cover (as determined by the Vegetation Continuous Fields data set), and change was measured without regard to forest land use. Tree cover assemblages that meet the 25% threshold include intact forests, plantations, and forest regrowth.
When zoomed out, this data layer displays some degree of inaccuracy because the data points must be collapsed to be visible on a larger scale. Zoom in for greater detail.
Overview
FORMA is a near real-time tree cover loss alert system. It uses a cloud computing algorithm to analyze frequently updated satellite imagery along with complementary information on factors that affect tree cover loss, such as fires and precipitation. The system generates twice-monthly “alerts” for the world’s humid tropical forests that identify 500 × 500 meter areas where new, large-scale loss is likely to have occurred.
FORMA is designed for quick identification of new areas of tree cover loss. The system analyzes data gathered daily by the MODIS sensor, which operates on NASA’s Terra and Aqua satellites. The FORMA alerts system then detects pronounced changes in vegetation cover over time, as measured by the Normalized Difference Vegetation Index (NDVI), a measure of vegetation greenness. These pronounced changes in vegetation cover are likely to indicate forest being cleared, burned, or defoliated.
FORMA alerts only appear in areas where the probability of tree cover loss is greater than or equal to 50%.
Upcoming upgrades to FORMA include improving the resolution to 250 × 250 meters, and expanding coverage to tropical dry forest and eventually to other biomes across the global scale.
Citation: Hammer, Dan, Robin Kraft, and David Wheeler. 2013. “FORMA Alerts.” World Resources Institute and Center for Global Development. Accessed through Global Forest Watch on [date]. www.globalforestwatch.org.
Methodology
This data set uses freely available satellite imagery, collected by the Moderate Resolution Imaging Spectroradiometer (MODIS), which operates on NASA's Terra and Aqua (EOS PM) satellite platforms and views the entire Earth’s surface every 1 to 2 days. The images help to reveal notable changes in vegetation cover over time using an indicator that measures vegetation intensity called the Normalized Difference Vegetation Index (NDVI). By applying an automated algorithm to this input and in combination with complementary data inputs for fire hotspots and global precipitation averages, areas where tree cover loss is likely to have occurred are therefore identified. The algorithm also employs parallel processing in a remote server system the cloud that enables the rapid analysis of these very large data sets.
Explanation
Tree cover loss alerts appear when and where new, large-scale loss is likely to have occurred after 2005. Thus the alerts should not be interpreted as an analysis of total tree cover loss area but rather as an indication of an area that has a high probability of having experienced tree cover loss or disturbance over time. The system employs advanced statistical techniques to achieve the best fit to scientifically validate information on loss, measured as a probability. On the GFW website, alerts appear only for areas where there is a 50% or higher probability of tree cover loss. That is, alerts appear for a particular period when there has been significant loss in the area during or before that period.
Temporal and spatial resolution
FORMA alerts are displayed on the GFW site as monthly data, and the site is updated each month. Users who download data for analysis will find that the underlying data set is actually available at 16-day intervals, intervals that do not line up perfectly with calendar months. Data is displayed at a monthly resolution on GFW due to the monthly availability of precipitation data input. Users can manipulate the GFW time slider to view trends in loss alerts from December 2005 to present.
The alert system currently identifies 500 × 500 meter areas where loss is statistically likely to have occurred. Alerts for a 250-meter resolution version are currently being developed on Google’s Earth Engine platform and will be available in 2014.
Geographic extent
FORMA alerts are currently available only for humid tropical forests (as defined by Hansen et al. (2008), based on WWF’s terrestrial ecoregions) spanning portions of 89 countries. The development team is working to incorporate additional data to extend the geographic coverage beyond the current extent. To visualize the geographic extent of the alerts on the map, switch on the “Humid Tropical Forest Biome” layer.
Data applications
The alerts have been designed for quick identification of tree cover loss as it happens. This allows for rapid response and prioritization of scarce financial and human resources dedicated to forest conservation or sustainable forest management. Armed with this information, stakeholders can use preemptive methods such as on-the-ground visits or aerial inspection with high-resolution satellite imagery (less than 5-meter pixel resolution) to investigate suspected tree cover loss areas.
In addition, the alerts may be of value to a variety of researchers who study both temporal and spatial patterns related to tree cover loss areas.
Using the GFW platform, the alerts can be compared against other relevant data layers, such as protected areas and concessions boundaries, to evaluate the effectiveness of forest management practices across time and spatial extent.
Accuracy and validation
Inaccuracies are an inherent part of remote sensing analysis. FORMA alerts appear in areas with a greater than 50% probability of tree cover loss, based on the algorithm described under Methodology. However, persistent cloud cover is a continuous issue in the tropics, and extreme flooding can also produce unreliable remotely sensed data that will result in tree cover loss “false positives” (alerts where no actual tree cover loss has occurred). Furthermore, the alerting system cannot detect all forest cover loss, whether due to the small size of the loss area, persistent cloud cover, or other explanations still being identified through GFW validation efforts.
The major instances of false positives may occur as the following:
- A random, "speckled" distribution of alerts across an ecoregion, or complete filling of a small ecoregion. Caused by limited or sparse training data, particularly in small ecoregions, which makes it difficult to tune the model there. As a result, alerts cannot be reliably detected. In a normal ecoregion, alerts are usually clustered.
- A rapid explosion of alerts over 1-3 months covering a relatively large area. Caused by a significant, persistent drop in detected vegetation levels due to persistent cloud cover along coastlines, in mountains, or elsewhere.
- Alerts in water. Caused by shifting water bodies. These alerts should be considered not necessarily as false positives but rather as ambiguous alerts requiring additional data for corroboration.
The GFW team is working aggressively to address potential inaccuracies in the data through rigorous validation methods. Specifically, the GFW team is comparing the growing data set of historical alerts to other validated data sets, which are being used for similar applications.
This issue brief demonstrates the spatial correlation of the alerts with the PRODES and DETER data sets, produced by the Brazilian Space Agency for the Amazon. The conclusions from the working paper help to illustrate the potential pitfalls of the algorithm, along with its strengths. Through future refinement and proposed crowdsourcing efforts, the GFW team expects the data quality of the FORMA Alerts will continue to improve.
Additional resources on FORMA
Working Papers:
- FORMA: Forest Monitoring for Action – Rapid Identification of Pan-tropical Deforestation Using Moderate-Resolution Remotely Sensed Data (CGD Working Paper 192, November 2009) [pdf]
- From REDD to Green: A Global Incentive System to Stop Tropical Forest Clearing (CGD Working Paper 282, December 2011) [pdf]
- Forest Clearing in the Pantropics: December 2005-2011 (CGD Working Paper 283, December 2011) [pdf] Data Set for Working Paper 283
- FORMA and fCPR: Accelerating a Performance-Based Payment System for REDD+ (CGD Policy Paper 006, June 2012) [pdf]
- Satellite-Based Forest Clearing Detection in the Brazilian Amazon: FORMA, DETER, and PRODES (WRI Working Paper, February 2014)
GitHub Repositories:
Geographic coverage of FORMA alerts
Download- Function
- Displays the geographic coverage of FORMA alerts
- RESOLUTION / SCALE
- Regional
- Geographic coverage
- This data layer shows the geographic coverage of FORMA alerts, which largely corresponds to the extent of the humid tropical forest biome, as defined by Hansen et al. (2008), and based on WWF’s terrestrial ecoregions. The biome illustrated by this layer includes a number of smaller forest ecoregions, which span portions of 89 countries.
- Date of content
- 2001
Citation: Olson, D. M., E. Dinerstein, E. D. Wikramanayake, N. D. Burgess, G. V. N. Powell, E. C. Underwood, J. A. D’Amico, I. Itoua, H. E. Strand, J. C. Morrison, C. J. Loucks, T. F. Allnutt, T. H. Ricketts, Y. Kura, J. F. Lamoreux, W.W. Wettengel, P. Hedao, and K. R. Kassem. 2001. “Terrestrial Ecoregions of the World: A New Map of Life on Earth.” BioScience 51, no. 11 (November): 933–38.
-
forma geographic coverage
- Function
- Displays the geographic coverage of FORMA alerts
- RESOLUTION / SCALE
- Regional
- Geographic coverage
- This data layer shows the geographic coverage of FORMA alerts, which largely corresponds to the extent of the humid tropical forest biome, as defined by Hansen et al. (2008), and based on WWF’s terrestrial ecoregions. The biome illustrated by this layer includes a number of smaller forest ecoregions, which span portions of 89 countries.
- Date of content
- 2001
Citation: Olson, D. M., E. Dinerstein, E. D. Wikramanayake, N. D. Burgess, G. V. N. Powell, E. C. Underwood, J. A. D’Amico, I. Itoua, H. E. Strand, J. C. Morrison, C. J. Loucks, T. F. Allnutt, T. H. Ricketts, Y. Kura, J. F. Lamoreux, W.W. Wettengel, P. Hedao, and K. R. Kassem. 2001. “Terrestrial Ecoregions of the World: A New Map of Life on Earth.” BioScience 51, no. 11 (November): 933–38.
-
SAD alerts (monthly, 250m, Brazilian Amazon, Imazon)
- Function
- Deforestation alert system that monitors forest cover loss and forest degradation
- RESOLUTION / SCALE
- 250 × 250 meters
- Geographic coverage
- Brazilian Amazon
- Frequency of updates
- Monthly
- Date of content
- January 2007–present
- Cautions
- When zoomed out, this data layer displays some degree of inaccuracy because the data points must be collapsed to be visible on a larger scale. Zoom in for greater detail.
Overview
The Deforestation Alert System (Sistema de Alerta de Desmatamento—SAD) is a monthly alert that monitors forest cover loss and forest degradation in the Brazilian Amazon. The system generates information that is published monthly by Imazon, a Brazilian NGO, through its Forest Transparency Bulletin. The monthly alerts are derived from a temporal mosaic of MODIS daily images that are scaled down from 500 × 500 meter to 250 × 250 meter resolution. The monthly results are then validated using medium resolution images from the China-Brazil Earth Resources Satellite (CBERS) and NASA Landsat data in order to “ground-truth” the results being reported.
Citation: Souza, C. M., S. Hayashi, and A. Veríssimo. 2009. “Near Real-Time Deforestation Detection for Enforcement of Forest Reserves in Mato Grosso.” FIG—Land Governance in Support of the MDGS: Responding to New Challenges. www.fig.net/pub/fig_wb_2009/papers/trn/trn_2_souza.pdf.
Suggested citation for data as displayed on GFW: “SAD Alerts.” Imazon. Accessed through Global Forest Watch on [date].www.globalforestwatch.org.
Geographic coverage of Imazon SAD alerts
DownloadThe geographic coverage of Imazon SAD alerts is the Legal Amazon (political) excluding the Brazilian state of Maranhão.
-
Imazon geographic coverage
This layer displays the geographic coverage of SAD alerts, which only cover the Brazilian Amazon.
-
FIRMS active fires (daily, 1km, global, NASA)
- Function
- Displays fire alert data for the past 7 days
- RESOLUTION / SCALE
- 1 × 1 kilometer
- Geographic coverage
- Global
- Source data
- MODIS
- Frequency of updates
- Daily
- Date of content
- Past 7 days
- Cautions
-
Not all fires are detected. There are several reasons why MODIS may not have detected a certain fire. The fire may have started and ended between satellite overpasses. The fire may have been too small or too cool to be detected in the (approximately) 1 km2 pixel. Cloud cover, heavy smoke, or tree canopy may completely obscure a fire.
It is not recommended to use active fire locations to estimate burned area due to spatial and temporal sampling issues.
When zoomed out, this data layer displays some degree of inaccuracy because the data points must be collapsed to be visible on a larger scale. Zoom in for greater detail.
Overview
The Fire Information for Resource Management System (FIRMS) delivers global MODIS-derived hotspots and fire locations. The active fire locations represent the center of a 1-kilometer pixel that is flagged by the MOD14/MYD14 Fire and Thermal Anomalies Algorithm as containing one or more fires within the pixel.
The near real-time active fire locations are processed by the NASA Land and Atmosphere Near Real-Time Capability for EOS (LANCE) using the standard MODIS Fire and Thermal Anomalies product (MOD14/MYD14). Data older than the past 7 days can be obtained from the Archive Download Tool. The tool provides near real-time data and, as it becomes available and is replaced with the standard NASA (MCD14ML) fire product.
More information on active fire data is available from the NASA FIRMS website.
Citation:NASA FIRMS. “NASA Fire Information for Resource Management System (FIRMS).” Accessed on [date]. earthdata.nasa.gov/data/near-real-time-data/firms.
Suggested citation for data as displayed on GFW: “NASA Active Fires.” NASA FIRMS. Accessed through Global Forest Watch on [date]. www.globalforestwatch.org.
-
Terra-i Alerts (monthly, 250m, Latin America, CIAT)
- Function
- Detects areas in Latin American where tree cover loss is likely to have recently occurred
- Resolution/scale
- 250 × 250 meters
- Geographic coverage
- Latin America
- Source data
- Moderate Resolution Imaging Spectroradiometer (MODIS) vegetation indices and water body’ presence data; Tropical Rainfall Measuring Mission (TRMM) precipitation data
- Frequency of updates
- Monthly
- Date of content
- 2004 – present
- Cautions
-
Given the lack of ground-based data, the methodology was validated using data from other forest monitoring systems such as PRODES (http://www.obt.inpe.br/prodes/index.php) which have been validated separately.
All clouds, water, and mist were masked based on MODIS Quality Assessment and MOD35 products and their values changed to “No Data”.
The Terra-i algorithm for change detection does not automatically identify events that occurred because of wild fires or within secondary forests or oil palm plantations. Furthermore, the moderate resolution of the MODIS sensor does not detect small scale events (<5ha). Terra-i is intended to be used to quickly identify deforestation hotspots which should then be more thoroughly investigated with higher resolution imagery or field validation.
Overview
Terra-i is a near real-time monitoring system that detects land cover changes in Latin America. It uses satellite data from MODIS vegetation indices (MOD13Q1 and NDVI) and products related to presence of water bodies (MOD35) as well as Tropical Rainfall Measuring Mission (TRMM) precipitation data to detect anthropogenic changes in vegetation cover every 16 days. Terra-i is a collaboration between the International Center for Tropical Agriculture (CIAT - DAPA), CGIAR’s Research Program on Forestry, Trees and Agroforestry (FTA), The Nature Conservancy (TNC), the University of Applied Sciences Western Switzerland (HEIG-VD), and King’s College London (KCL).
The system, which uses a computational algorithm similar to FORMA (http://www.globalforestwatch.org/sources/forest_change#forma), is based on the premise that natural vegetation follows a predictable pattern of change in greenness from one date to the next, brought about by site-specific land and climatic conditions over the same period. The model is trained to understand the normal pattern of changes in vegetation greenness in relation to terrain and rainfall for a site, which allows for prediction of what the next vegetation response should be based on the historical data. If the prediction is significantly different from the historical responses in relation to pattern of rainfall and lasts for two 16-day periods in a row, the pixel is marked as potentially having changed by anthropogenic means.
Citation:Reymondin, Louis, Andrew Jarvis, Andres Perez-Uribe, Jerry Touval, Karolina Argote, Julien Rebetez, Edward Guevara, and Mark Mulligan. 2012. “Terra-i: A methodology for near real-time monitoring of habitat change at continental scales using MODIS-NDVI and TRMM.” CIAT-Terra-i. http://terra-i.org/dms/docs/reports/Terra-i-Method/Terra-i%20Method.pdf.
Suggested citation for data as displayed on GFW: Reymondin, Louis, Andrew Jarvis, Andres Perez-Uribe, Jerry Touval, Karolina Argote, Julien Rebetez, Edward Guevara, and Mark Mulligan. 2012. “Terra-i alerts." CIAT-Terra-i. Accessed through Global Forest Watch on [date].www.globalforestwatch.org.
-
terra-i geographic coverage
This layer displays the geographic coverage of Terra-i alerts.
) located at the bottom of the window with analysis results and enter your email address.
) on the right-hand side of the map and choose “Country or region."
