In a collaborative project between the University of Oxford, UK, and the University of Twente, in the Netherlands, scientists successfully used very high-resolution satellite imagery to count and detect wildlife species, including African elephants. The researchers detected elephants in South Africa from space using Artificial Intelligence with an accuracy that they have compared to human detection capabilities.
The team has described their work in the journal Remote Sensing in Ecology and Conservation.
The population of African elephants has plummeted over the last century due to poaching, retaliatory killing from crop-raiding and habitat fragmentation. Therefore, in order to conserve the species, it is important for scientists to track elephant populations. It is important that scientists know the exact number of elephants that exist in an area as inaccurate counts can lead to misallocation of conservation resources, which are already limited and have resulted in misunderstanding population trends.
The scientists employed the high-resolution images taken by satellite Worldview 3 to record African elephants moving through their range in the Addo Elephant National Park in South Africa, the country’s third-largest park and which has a high concentration of elephants. The team created a training dataset of 1,000 elephants and fed it to the Convolutional Neural Network (CNN) and compared the results to human performance. An automated system based on an algorithm developed by Olga Isupova, a computer scientist from the University of Bath, was able to locate elephants with the same accuracy as human observers on the ground. This research was particularly novel because it was able to detect the elephants across a number of different geographical backdrops, such as grasslands, forests and desert.
Isupova said: “Accurate monitoring is essential if we’re to save the species, we need to know where the animals are and how many there are.”
As Worldview 3 hurtles through the exosphere it can capture 5000 square kilometers of land imagery every few minutes. Using satellite images eliminates the risk of disturbing animals, the need for low flying planes, and any harm that may come to humans during the data-collection process.
The feasibility of future satellite studies will be influenced by the high cost of commercial satellite imagery and the high volume of data that needs processing after collection. But processing compares favorably to traditional species-detection efforts – a few hours instead of fieldwork that may take weeks, or longer; and new commercial satellites, which may serve to lower costs, are due for launch this year.
Isupova believes we’ll soon be capturing pictures of much smaller species and expanding the use of satellite imagery in conservation efforts.
Before researchers developed the new technique, one of the most common survey methods to keep a check on elephant populations in savannah environments involved aerial counts undertaken from manned aircraft.
However, this method does not deliver accurate results since observers on aircraft are prone to get exhausted, are sometimes hindered by poor visibility and may even succumb to bias. Further, aerial surveys are costly and logistically challenging.