ODC’s forest cover change maps show trends illustrating the extent and rate of reduction in Cambodia’s forest cover over the past 40 years. The maps’ primary focus is on changes in evergreen forest, referred to as “dense forest” in Cambodia. In English, Cambodia’s dense forest might also be termed old growth forest. ODC’s analysis of satellite images shows that in 1973 approximately 72 percent of Cambodia was covered by forest. More recent images suggest that today’s forest cover is closer to 46 percent, inclusive of tree plantations. Inclusion of tree plantations in this figure is consistent with the Cambodian government’s approach to assessing forest cover.
This page shows both animated and static maps – the animated maps highlight trends in forest cover, with changes in dense forest most clear. Seven videos animate the change in forest cover for the entire country, while six static maps show more detailed views of particular regions. Regional animations show the boundaries of Cambodia’s protected zones, comparing forest cover change inside the protected zones with that of unprotected areas. Most of Cambodia’s protected zones were designated in the 1990s. More information on protected zones, along with ODC’s static maps can be found on our interactive map page. Static maps can be used as a base layer for overlaying features and other kinds of development information that is available on ODC.
Other organizations, such as WWF and University of Maryland, have conducted similar studies to assess Cambodia’s forest cover change, with similar findings.
Update: On 22 January, the Cambodian Ministry of Agriculture, Forestry and Fisheries published an official response to the ODC Forest Cover Changes page. View their report here.
Who made these maps?
ODC mappers began work on this project in June 2013. The team included Masters graduates from the Royal University of Agriculture’s natural resource management program, and the GIS and remote sensing program at the Asian Institute of Technology in Thailand. They are among the first Cambodians to do this type of work and developed their skills in satellite image analysis throughout the project. Following continual training and support from mapping experts, they are now skilled in forest cover change analysis. ODC’s external consultant is a professor at Vietnam’s Can Tho University who completed his doctorate in the UK. He is a specialist in climate change modeling and reviewed work performed by ODC’s mappers at each stage of the project, from image selection to image analysis and classification.
Bands: Bands of color are used to perform image classification. The band designations used can be accessed at the USGS website.
Cloud: Visible cloud cover at the resolution of the satellite image. For 1973, the resolution of the satellite image is 60 meters. For all other years the resolution is 30 meters. The 2013 map has the highest amount of cloud cover at 6.47 percent – more information is included in the scope and limitations.
Dense forest: In Cambodia’s case, dense forest signifies evergreen forest, comprised of trees that do not lose their leaves through seasonal abscission. Areas classified as dense forest in these maps include “evergreen forest” and “semi-evergreen forest” as defined in the Forestry Administration’s Cambodia Forest Cover publication dated June 2008. Dense forest is mostly located at elevations higher than 500 meters, although Cambodia has also had large areas of lowland evergreen forests in the past. Dense forest may also be called old growth forest.
Mixed forest: Primarily regarded as dry mixed deciduous forest (deciduous trees drop and regrow their leaves seasonally.) Mixed forest may also include regrowth forest, stunted forest, mangroves, inundated or “flooded” forest, and bamboo, as well as forest plantations growing rubber, acacia, and eucalyptus or other tree crops. Areas classified as mixed forest in these maps include “deciduous forest” and “other forest” as defined in the Forestry Administration’s Cambodia Forest Cover publication dated June 2008.
Non forest: Non forest is not dense forest, mixed forest, water or cloud. Non forest includes urban areas, field crops, shrubland, fallow or barren land, and other human-impacted areas.
Satellite: An object placed in earth’s orbit that is able to photograph the planet’s surface at great distances. NASA’s Landsat satellites took the images used to generate the maps on this page.
Water: All bodies of water that are visible at the resolution of each satellite image. For 1973, the resolution of the satellite image is 60 meters. For all other years the resolution is 30 meters.
1. ODC produced forest cover maps for 1973, 1989, 2000, 2004, 2009, and 2013 from satellite images. The earliest images available for Cambodia date from the early 1970s. ODC selected satellite images for analysis based on their quality, particularly the absence of cloud cover.
2. Analysis of Cambodia’s prewar forest, compared to forest cover in 1989, showed only minor changes, while significant changes were observed between 2000 and 2013. Mappers were able to determine where dense forest was lost – the changes are shown using contrasting colors. Dark green represents dense forest. Red on a map denotes areas of lost dense forest that were first detected on that year’s map. As the animation progresses to the next frame, red areas disappear to reveal the new status of the land as either mixed forest or non forest. Mixed forest may be seen to fluctuate, sometimes increasing over the time series. This is because as when dense forest is lost it is often replaced by mixed forest. Mixed forest is sometimes difficult to distinguish from shrubland, classified as non forest. In the last frame of the animation, red is used to show the total amount of dense forest lost between 1973 and 2013.
3. Each “forest cover map” is composed of 16 to 19 NASA satellite images, which were obtained from the US Geological Survey. The images are free, publicly available and can be downloaded at the USGS website.
4. The available map resolutions required the use of image classification methodology, rather than other available methods to detect forest cover changes. This process involved converting multiband raster images into a single composite band raster image. The composite band raster image allowed the mappers to identify five forest cover classifications to be mapped: dense forest, mixed forest, non forest, water and cloud. The band wavelengths used to classify the images can be found at the USGS website.
5. Cloud shadow, hill shadow or water reflections appearing on a map presented classification challenges and verge areas can be subject to seasonal variances. Where these ambiguities were present on a map, the mappers analyzed later maps to mitigate the risk of incorrect classification.
6. The 1973 forest cover map consists of 16 Landsat MSS images with 60 meter resolution. A mosaic was created from these images, to which color bands 3, 2 and 1 were applied to identify the Region of Interest. Landsat TM images were used at 30 meter resolution to create the 1989, 2000, 2004, and 2009 forest cover maps, with color bands 4, 3 and 2 applied. Landsat 8’s Operational Land Imager satellite images were used at 30 meter resolution to create the 2013 forest cover map. The map was analyzed by applying color bands 5, 4 and 3 to identify the ROI.
7. A forest change detection model was used with the forest cover maps to depict the loss of forest from one map to the next, shown on five additional maps known as “forest cover change maps.”
8. Each of the seven animations, one national and six regional, is composed of 11 static maps: the six forest cover maps and the five forest cover change maps.
Although the forest cover change maps are a product of rigorous analysis, a number of limitations, and potential mapping imperfections persist.
1. Scope: The ODC forest cover maps show trends in forest cover change focusing on dense forest. Classification is limited to dense forest, mixed forest, and non forest. Since the maps are based on satellite images with a resolution of either 60 meters or 30 meters, it was not possible to differentiate the specific tree species that comprise the forests. For this reason, the mixed forest classification also includes tree plantations. At this resolution it was not possible to identify small pockets of deforestation.
2. Cloud cover: All of the satellite images used to produce the maps contained a small amount of cloud cover. This presents a challenge for image classification, as information underneath clouds cannot be detected. To minimize the impact of this, ODC sought the clearest images with the least cloud cover. The 1973 image contained less than 1 percent cloud cover at 60 meter resolution. The 1989, 2000, 2004 and 2009 images also contained less than 1 percent cloud cover, but at 30 meter resolution. The 2013 satellite images contained 6.47 percent cloud coverage, also at 30 meter resolution. In ODC’s forest cover maps, clouds are depicted as white areas. Areas underneath clouds could not be analyzed. Where clouds were present on a map, the mappers analyzed later maps to mitigate the risk of incorrect classification.
3. Season: Cambodia has two seasons: rainy and dry. The landscape of the country changes during each season. For example, vegetation is greener during the rainy season than during the dry season. There are also variations in the size of water-covered areas between the seasons. Since image classification involved combining multiple satellite images taken during different seasons, this may have led to slight variances in forest cover, mostly in verge areas where shrubland can be mistaken for mixed forest and vice versa at this resolution. Where these ambiguities were present on a map, the mappers analyzed later maps to mitigate the risk of incorrect classification.
4. Scan and stroke lines: There were three instances of small stroke lines appearing on maps – one in 1989 over Western Cambodia, and two in 2009 over Mondulkiri. The scan lines resulted in negligible information gaps.
5. Ground truthing: Since these maps apply to the entire country, ground truthing was not feasible. The image classification method used to create these trend maps does not typically include ground truthing.