Skip to main content


AI Provides Important Information on Africa’s Ecosystems

The results of a research project on the mapping of trees in West Africa can help to strengthen ecosystems. An important contribution was made by the Center for Computing Technologies with its expertise in AI. The results were published in Nature journal.

The development of forests across the world is a well-researched subject. However, it is just as important to identify individual trees and tree groups. They are of significance for the regional ecology and global climate change. An international team of researchers recently made the efficient mapping and counting of individual trees in large areas possible for the first time ever. The core was an artificial intelligence (AI) process, which was made to fit for this particular purpose and trained by scientists from the Center for Computing Technologies (TZI). The success of the project suggests that it may soon be possible to map all trees on Earth – with minimal restrictions.

More Trees than Assumed

As part of the research project, of which the results were recently published in the renowned Nature journal, researchers mapped every tree and bush with a crown that covered a minimum surface of three square meters. They did this across 1.3 million square kilometers in West Africa. The result: Around 1.8 billion individual trees can be found in the West African Sahara and the adjacent Sahel region – far more than assumed to date. The gained data could help to strengthen ecosystems, gain data for climate protection, and observe the deforestation processes, amongst other things.

Artificial Intelligence Combined with Geoinformatics

The identification of individual trees was made possible due to the fact that NASA and private space travel companies continually provide more and more high-definition photo material. In order to analyze the masses of data, the TZI scientists Professor Johannes Schöning and Ankit Kariryaa adapted an AI process from the field of Deep Learning – namely fully convolutional neural networks. “In the Human-Computer Interaction working group at TZI, we have both the expertise in artificial intelligence and in geoinformatics, explains Johannes Schöning, who realized the research work in the frame of his Lichtenberg Professorship from the Volkswagen Foundation. “We were thus able to solve this problem together with our friends at the universities in Copenhagen and Münster.”

The chosen AI process can recognize objects – for example treetops – based on their characteristic colors and shapes. The AI system was trained with the help of images in which the trees had been marked by hand. “Due to the special features of the region, we had to overcome many obstacles,” reports Ankit Kariryaa. “For example, the appearance of the vegetation and the ground is extremely different in the regions with low precipitation in comparison to those where it rains a lot. That is why we trained two separate systems.”

Long-Term Goal: Determination of Individual Tree Types with Satellite Images

The result of this work is a map showing all trees that have a diameter of two meters or above in the South of Mauretania, Senegal, and Mali. In the future, the results can not only be expanded geographically to include of regions of the world, but also be combined with additional data, such as radar sensors. This could help to determine different tree types, for example.

Further Information:

The Nature article with the title "An unexpectedly large count of trees in the West African Sahara and Sahel" can be found here:

The Nature editorial team also highlighted the project as one of ten especially remarkable scientific discoveries in 2020:

Nasa visualized the project:



Axel Kölling
Public Relations
Center for Computing Technologies (TZI)
University of Bremen
Phone: 0171-5305119
Email: axel.koellingprotect me ?!uni-bremenprotect me ?!.de

Landscape from above with free standing trees
The landscape near to Bandiagara (Mali) shows many freestanding trees that have not been included in tree counting methods to date.