The browser you are using is not supported by this website. All versions of Internet Explorer are no longer supported, either by us or Microsoft (read more here:

Please use a modern browser to fully experience our website, such as the newest versions of Edge, Chrome, Firefox or Safari etc.

Mapping tree species using satellite observations

Hakim Abdi

Forests provide us with a large number of ecosystem services and function. These include taking up carbon dioxide from the atmosphere, wood for timber and pulp, stabilizing the soil to reduce landslides, maintaining biodiversity, and possibilities for recreational activities like hiking or camping. But, the quality and quantity of these services and functions depends on the types of trees that form the forest.

The current state of knowledge is that forests with more diversity of tree species provide more ecosystem services than forests with a fewer diversity of tree species. This means that knowledge about the distribution of tree species is necessary to better understand their contribution to the functioning of ecosystems and supply of ecosystem services. Sweden is a big country, so it is important to map tree species across large regions. Mapping tree species has usually been done with optical data from airborne or satellite remote sensing and simple algorithms such as maximum likelihood. The results have been good but these approaches are not suitable for very large areas. 

This project uses a state-of-the-art deep learning algorithm in combination with radar and optical data from the Sentinel-1 and Sentinel-2 satellites, respectively, to map tree species in southern Sweden. Microwave radiation emitted by the Sentinel-1 satellite are able to penetrate into the tree canopy and return information about the structure of the tree canopy. The sensor onboard the Sentinel-2 satellite can capture solar radiation reflected from the earth across a wide range of wavelengths. This includes wavelengths from the so-called “red-edge” portion of the electromagnetic spectrum that is sensitive to the chlorophyll content of vegetation and an thus help differentiate different tree species. In this way, these two satellites complement one another and help improve tree identification. Additionally, using the latest deep learning algorithms on this complementary combination of data sources has the potential to further enhance the detection of tree species from satellite imagery. All of this leads to an operational process that relies only on satellite data and deep learning to provide regular maps of tree species in Sweden.


Involved researchers

  • Henrik Smith
  • Hakim Abdi