New camera trap technology reveals new insight into animal movement.
Movement ecology is flourishing, fuelled by improvements in the technology used to track animal positions – ever smaller telemetry devices, better power management, high-frequency GPS data logging, satellite communication and so on.
Thanks to this technological bonanza, great strides are being made in our understanding of animal movement behaviour, but this approach cannot avoid one key problem – you have to catch and tag your animals before you can track them. This is invasive for the animals, and difficult to achieve in practice, so the range of species studied in this way is not as broad as we might like.
At the same time, a totally different kind of technology – the camera trap – is being used more than ever before to keep track of wildlife populations, remotely and unobtrusively recording animals going about their business in the wild.
In the past, camera traps used film, so wildlife records mostly consisted of single images. This was fine for estimating the abundance of some species, or just for getting an idea of the diversity of wildlife in an area. But camera traps these days are of course digital, and can easily generate video or near-video recordings of animals moving past cameras.
This footage lets us see animals moving, remotely and without having to get close to or catch them. We wondered whether we could use the vast amounts of camera trap imagery now emerging to estimate animal speeds of movement. A new paper published in the journal Remote Sensing in Ecology and Conservation shows that you can.
The basic idea is fairly simple. First, do your camera trap survey, find an animal in your footage, and trace its path on the ground in front of the camera to measure the total distance travelled.
Next, divide this distance by the time taken (from time stamps embedded in the image file metadata) to give a speed measure. Then average over a sample of speed observations and you have your estimate of how fast the animals travel, as well as how variable their speeds are.
This is only part of the story though, because inactive animals don’t (usually) trigger cameras. So our average speed only applies to active animals. If we want to know how far animals travel over the longer term – their day range – we need to correct for their level of activity.
Fortunately we have previously developed a method for estimating levels of animal activity from camera traps, so we can use this to adjust travel speeds to get day ranges.
The obvious benefit of these methods is that you can combine them with existing camera trap surveys to enhance your understanding of the animals you’re seeing. A less immediately obvious advantage of the methods is that they don’t suffer from the coastline paradox – the fact that the apparent length of a line can depend on how closely you look at it.
Tracking studies typically measure distances travelled using position fixes taken several times a day, rarely more than a few times an hour. But because a lot of hidden ground can be covered by animals between positions even just a few minutes apart, intermittent fixes like these underestimate travel distances by a substantial and unpredictable amount.
In contrast, camera traps using video or near video settings can capture movement at sub-second resolution, effectively revealing all the steps and turns that animals make and missing none of the action.
Camera-based speed estimates should therefore be accurate, repeatable and comparable across studies in a way that tracking-based estimates usually are not (although my impression is that this problem is somewhat glossed over in the tracking literature – something that needs more attention I think).
The camera trap approach has its challenges too, of course. The methods depend on animals acting “normally” around camera traps, so that the speeds we observe are not distorted, but while the devices are reasonably unobtrusive, they’re not completely non-invasive – we know that animals often see, hear or smell cameras and react by fleeing, or approaching for a good sniff.
In some cases, these reactions are sufficiently rare that we can just ignore the few records showing clear reactions and still get a good sample of unbiased speed observations. In other cases, it may be that reactions are too frequent to yield any useable data. Our experience so far is that we can get useable data for most species, but that might not always be the case.
Extracting speed observations from images is not a trivial process. In a study with thousands of animal records (not an uncommon occurrence), you will need to spend many days staring at images and trying to figure out where animals went on the ground to generate the necessary data. This is a pretty substantial additional time commitment. So to tackle this, we’re working on image analysis tools for converting digital positions to real world positions.
Looking ahead, I have a vision of tools for automatically tracking animals’ digital positions to reduce the workload further. Just a dream for now, but I don’t see why it couldn’t be a reality in due course.
Of course, you can get more than just speed of movement from tagged animals (home ranges, details of habitat associations and behavioural states, to name a few interesting phenomena), so camera traps are clearly not going to replace telemetry as the movement ecologist’s tool of choice.
But we do think that these new methods can add an exciting new dimension to camera trapping studies. For example, we hope that movement rate estimates will emerge for a wider range of species and environments than previously studied, helping to improve our understanding of the ecological determinants of these behaviours, and shedding light on how threatening processes such as loss of prey or habitat operate.
Measuring day range is also a fundamental component of the random encounter model (REM) for estimating animal densities using camera traps. This model has potential for cost-effectively monitoring populations of many elusive animals that have so far proven resistant to traditional survey methods. Currently, though, applying the REM requires independent estimates of day range, perhaps from tracking data, which are usually hard to come by, so the approach isn’t as widely used as it might be. These new methods for estimating movement should enable us to apply the REM to freestanding camera trap surveys, without the need for additional data on movement.