Modern computing makes it possible to better identify isolated trees and shrubs in arid and semi-arid areas -

© Martin Brandt / The Conversation

  • New technologies are enabling the mapping of billions of individual trees and shrubs in West Africa, according to our partner The Conversation.

  • The acquisition of very high resolution images and then the use of supercomputers reveal a much larger population than that which was previously accepted.

  • The analysis of this process was carried out by Martin Brandt, Lecturer in Geography, and Kjeld Rasmussen, Associate Professor Emeritus, both at the University of Copenhagen (Denmark).

Arid and semi-arid areas have long been studied to see if their vegetation cover is declining.

Indeed, the theory according to which the Sahara was expanding and woody vegetation was retreating was first put forward in the 1930s. Then, the "great drought" of the 1970s in the Sahel focused on the desertification caused. by overexploitation and by climate change.

In recent decades, it is the potential impact of climate change on vegetation that has been the main concern - and the retroactive effect of vegetation on climate, linked to the role of vegetation in the global carbon cycle. .

To better understand the state of the vegetation cover and its evolution in arid and semi-arid areas, we recently mapped billions of individual trees and shrubs in West Africa.

A challenge met by combining high-resolution satellite images and machine learning techniques, thanks to supercomputers.

Find a shrub in the desert - from space

Since the 1970s, vegetation in semi-arid areas around the world has been mapped using satellite data.

The images available are either of “high” spatial resolutions (with the satellites of NASA, Landsat MSS and TM, and of ESA, Spot and Sentinel), or of “medium or low” spatial resolutions (NOAA AVHRR and MODIS satellites ).

To accurately analyze land cover on a continental or global scale, use the highest resolution images available, with a resolution of one meter or less.

Until now, the costs of acquiring and analyzing this data have been prohibitive and most studies have relied on medium or low resolution data, which does not allow individual trees to be identified.

These studies therefore only give estimates of aggregate vegetation cover and productivity, mixing more herbaceous and woody vegetation.

A new study published in


in October 2020, covering much of the semi-arid zone of the Sahara, Sahel and Sudan in West Africa, overcomes these limits.

By combining an immense amount of high-resolution satellite data, advanced computational capabilities within a supercomputer, machine learning techniques, and extensive field data collected over decades, we were able to identify trees and shrubs. individuals with a crown surface greater than 3 m2 with great precision.

The result is a database of 1.8 billion trees in the study area, available to all interested.

Supercomputer, machine learning, satellite data and field assessments help map billions of individual trees in West Africa © Martin Brandt (via The Conversation)

Currently, this work is being extended to cover the semi-arid belt south of the Sahara across the African continent to the Red Sea.

The number of trees mapped to date is 13 billion, and the methodology is being improved.

The geographical coverage should be extended, first to the rest of the semi-arid zones of Africa, then to other continents.

To cover the entire Sahelian zone of Africa, from the Atlantic to the Red Sea, we used around 100,000 satellite images, for a total data volume of several hundred terabytes.

Using supercomputers from NASA and Blue Waters (University of Illinois at Urbana-Champaign), the images were stitched together to create a continuous mosaic.

The trees were then identified using deep learning, an artificial intelligence technique in which the computer is trained to recognize individual trees.

During his training, tens of thousands of trees were “shown” to the computer by an operator, who used his knowledge of the field in combination with his skills in image interpretation.

Then, the results of the machine identification were checked.

Overall, accuracy was found to correlate strongly with measurements in the field.

Unexpected information about individual trees

Our database of trees and shrubs contains information on each tree, its exact location (usually with an uncertainty of a few meters), the size of its crown, the date of acquisition of the satellite image on which it was identified, along with estimates of its above-ground woody mass and carbon content.

In the future, other information may be added, for example its height and phenology, i.e. periodic events such as leafing.

The people of the semi-arid Sahel safeguard and promote trees in settlements and agricultural land.

The relationship between humans and trees does not always result in tree cover losses © Martin Brandt (via The Conversation)

This is only the start of the research project, but important implications are already evident.

In the West African study, we found many more trees than we expected.

While other data sources indicate that trees are almost absent from the Sahara and the northern Sahelian zone, we have found hundreds of millions of trees.

The carbon stock associated with these trees would be greater and more stable than the carbon stocks in herbaceous vegetation.

In addition, trees in agricultural land are generally taller than those in pristine savannahs, and overall tree cover is high in populated or exploited areas.

This shows that a high human population density is not always linked to a loss of tree cover, as the inhabitants of the semi-arid Sahel protect and encourage trees in inhabited areas and agricultural land.

What will the database be used for?

This database serves different purposes.

It constitutes a reference base which will make it possible to study the temporal evolution of woody vegetation on a large scale, perhaps even on continental or global scales.

Our folder "Trees"

It will also make it possible to analyze the factors which control the presence of trees in arid zones, such as human occupation, precipitation, soils or geomorphology.

This information will feed into the modeling of ecosystems and the "Earth system", since trees play important roles in the interactions between the atmosphere and the earth's surface, controlling both carbon exchange, evapotranspiration and roughness. aerodynamic.

Finally, the information from the database could be used to inform and support environmental policies at national and international levels.


Climate change: Rethinking African agriculture to improve food security


Why welwitschia, one of the strangest trees on the planet, baffles botanists

This analysis was written by Martin Brandt, Lecturer in Geography, and Kjeld Rasmussen, Associate Professor Emeritus, both at the University of Copenhagen (Denmark).

The original article was translated and

published on The Conversation website.

Declaration of interests

  • Martin Brandt has received funding from the AXA Postdoctoral Research Fund (created in 2007 to accelerate scientific knowledge and its sharing, the Axa Research Fund supports around 650 projects worldwide led by researchers from 55 countries. To find out more, visit Axa Research Fund website or follow on Twitter @AXAResearchFund.)

  • Kjeld Rasmussen does not work, advise, own shares, receive funds from any organization that might benefit from this article, and has not declared any affiliation other than his research organization.

  • Nasa

  • Artificial intelligence

  • Environment

  • Africa

  • Forest

  • Tree

  • Podcast

  • Planet