Why Use AI For Surveying?

Aerial Surveying

Aerial surveying is a key tool for effective wildlife management. However the high costs associated with large-scale surveys means that this tool is often underutilised. We believe that computer vision can be used to dramatically decrease the costs associated with surveying, while at the same time improving the consistency of results.

The traditional approach to aerial surveying is to load rear-seat observers (RSOs), a recorder and a pilot in a light aircraft and fly multiple transects over a designated survey area. Rods mounted to the wing struts are used to delineate a known fixed viewing angle (given a supposedly fixed eye location for the observer). Using these angles and the recorded altitude of the carriage plane, allows one to calculate the realised width of the observation strip on the ground. Observers call out sightings of animals inside the observation strip and these sightings and their locations are recorded by the recorder. In some cases teams will use cameras mounted in fixed locations to photograph herds that are too large to accurately count in flight, although this refinement is not universally adopted. Afterwards the data can be processed with statistical techniques to yield a population estimate for the surveyed area and error bounds on this estimate.

While this approach has for years been the gold-standard for aerial surveying, the inherent dependence on human observers has some major drawbacks:


Spotting stationary animals that blend in well with the environment in the couple of seconds of observation time available as the plane passes is a highly challenging task. Performance will vary from one observer to the next and even the same observer’s performance might vary a lot due to fatigue levels, deteriorating eyesight etc. These extraneous sources of variation make it a lot more challenging to detect the true underlying trends in population when comparing one survey to another.


Accurately counting a large herd (that may only be partially inside the observation strip) in the limited time available is not possible, so observers need to use cameras and reconcile the counts afterwards or (commonly) just estimate the herd size. These estimates are never compared to other sources of information, so we do not know the magnitude of errors that a typical observer might commit here. Also, because there is no correction process, we can not assume that an experienced observer’s estimates will be any more accurate than those of a novice. For highly social animals like elephants that tend to gather in large herds this source of error could easily dominate.


Flying at low altitudes in light aircraft for long durations over remote and often inhospitable terrain is an inherently risky endeavour. By using a human crew we are placing human lives at risk. If the required crew could be reduced, the risk would reduce commensurately. Increasing the operating altitude would further reduce risk to remaining crew.


A significant proportion of the costs of elephant surveys (especially large-scale ones) relate to the logistics of transporting, feeding and housing the human crew members.

Due to these factors and others, large scale aerial surveying is often prohibitively expensive and consequently underutilised as evidenced by the fact that the Great Elephant Census was the first pan-African census in 40 years.


Our goal with the ESS is to develop a cheaper alternative that scales well. The system (hardware, software and data) is fully open source to encourage others to adopt it and iterate on it.

Ready to get started with our Elephant Survey System? Let's chat.