Satellite imaging and deep learning provide a clearer picture of the Sahara.
Learn about how satellite technology is used to understand the ecosystem and land and water management in the Sahara. This article is suitable for Year 7 and 9 Biology and Earth and Space students learning about how water affects the ecosystem.
Word Count: 527
A surprising new study has revealed that there are more than 1.8 billion trees and shrubs in the Sahara and Sahel regions of West Africa.
“One would not expect that many trees in the Sahara Desert, which people image as only sand dunes and stones,” says Martin Brandt from Denmark’s University of Copenhagen, lead author of a paper in the journal Nature.
“But there are actually areas with a relatively high tree density, although there is almost no rainfall, and even between the sand dunes there are a couple of trees.”
While the overall tree cover is low – 13.4 per hectare – the researchers found that the number of individual trees is relatively high, with an average crown size of 12 square metres.
Until now, satellite images haven’t been able to detect single trees outside forests, so maps just show white spots.
With advanced deep learning methods and high-resolution satellite images, the team of 24 scientists from across Europe, the US and Africa was able to map and measure every single tree more than three square metres in size over 1.3 million square kilometres, spanning a rainfall gradient from zero to 1000 millimetres per year.
Over the course of a year, Brandt labelled many of the trees by hand to teach the computer what they look like, and the resulting deep learning algorithm scanned more than 11,000 satellite images. If any trees were misclassified, he added further training.
As well as assessing tree density and crown size of each tree, the analysis linked these with rainfall and land use, finding the greatest density and cover at about 400 millimetres of annual rainfall.
The results upend previous assessments, most of which mapped 0% tree cover due to lack of closed canopies and exclusion of non-forest trees, the authors note. Even the current study’s numbers are underestimates, they say, because the analysis excluded trees or shrubs with small canopies.
This “challenges prevailing narratives about dryland desertification”, they write.
The study also shows the potential for mapping individual trees worldwide, write Niall Hanan and Julius Anchang from New Mexico State University, US, in an accompanying News and Views
article.
They note that this is an important advance, as remote sensing of vegetation with coarse resolutions has been limited to gross landscape descriptions, with a dearth of detail on tree density, canopy size or individual tree locations.
Over the past couple of decades, many commercial satellites have started to capture images at higher resolutions and can detect objects as small as one square metre or less.
“This transformation in how we observe terrestrial ecosystems – savannas, shrublands, woodlands and forests – will change how we monitor, model and manage them,” says Hanan.
It could inform their role in biodiversity, shelter for humans and animals and food sources, according to Brandt and co-authors. The trees could also help prevent erosion, provide carbon reservoirs, support pollination and improve soil and water and nutrient recycling.
While Hanan says the deep learning approach will be challenging at global scales because of the sheer scale of data, training and computation required, necessitating further advances in efficient deep learning methods, “this study is the first to really demonstrate the possibility”.
Years: 7, 9
Login or Sign up for FREE to download a copy of the full teacher resource
