Satellite Count Recorded 15.7% Fewer Elephants Than Aircraft in Botswana

A satellite count of African bush elephants (Loxodonta africana) in northern Botswana recorded 15.7% fewer animals than a simultaneous aerial count. The comparison was part of a framework for testing whether high-resolution satellite imagery and artificial intelligence can contribute to large-scale wildlife surveys in vegetated landscapes.
The same area produced 816 aerial and 688 satellite detections
On 6 October 2022, researchers coordinated a satellite image and an aircraft survey over the same area near Ngwasha, south-east of Chobe National Park. The aerial observers counted 816 elephants, while 688 visible elephants were identified in the 30-centimetre-resolution satellite image — 128 fewer. The study site covered 140 km² and was dominated by open bushland and shrubs.
The AI detector struggled when moved to new terrain
The project trained an automated detector using 813 elephant examples from satellite images representing several vegetation types. Performance was substantially better during validation than on a held-out satellite image that the detector had not seen during training, illustrating how difficult it is to transfer automated detection to new terrain. Elephants in tight groups or among vegetation were especially difficult to detect, while rocks could be mistaken for small elephants.
A simple canopy correction overshot the aerial count
Vegetation created an additional problem before any automated detection was considered. Five land-cover classes described as open or sparse still had estimated canopy cover ranging from 11% to 47%. When the researchers applied a simple correction for the estimated 26% of the site hidden by canopy, the raw satellite count of 688 became an abundance estimate of 929 elephants. That was 13.8% above the aerial count of 816, showing that simply scaling up by hidden area can overcorrect when elephant density and use of shade vary across the landscape.
The result shows why correcting a satellite count for hidden terrain is not enough on its own. The authors conclude that satellite surveys can already add useful, rapid and permit-free information, but are not yet a direct replacement for Botswana’s aerial elephant surveys. For now, they see the strongest potential in combining satellite and aerial data while developing better corrections for canopy cover, species misclassification and detection uncertainty.
SPECIES IN THIS STORY
Species in this story
Independent research and conservation news archive
Did you find this information useful?
Wildlife Vagabond is independently built and maintained. Voluntary support helps cover source verification, hosting and continued work on research and conservation news.
Support Wildlife VagabondThe news archive will remain freely available.



