Self-resetting (automatic) traps have been on the market for a while now. Speaking to trappers far and wide (as we have a habit of doing) we hear mixed reviews. Most people seem to get some success when first deploying an auto-trap but the results tend to dip pretty swiftly. At Cacophony, we do value such anecdotal evidence (it really helps us understand the problems of using devices in the field) but we value hard, physical evidence even more.
Importantly, when we hear of a device not working as well as we all hoped it might, our endlessly intrigued minds feel the need to dig a bit deeper into the problem. Finding out why a device doesn’t meet expectations is interesting in itself but understanding how predators interact (or don’t) with a device is important input into our own design processes.
As part of our testing we often put our cameras in front of existing traps to see if we can work out the pros and cons of different tools. In this case we had one of our thermal cameras set up in front of an AT220 NZ Autotrap and a Good Nature possum trap. Below is a photo of the test set up on a tree that we know possums regularly go up and down. Ideally we would test the traps separately but we thought it may be interesting to see how animals behave close to both of them.
The experiment took place between 8/8/2019 and 18/11/2019. Over that time, our camera made around 4000 recordings. About 380 of those were automatically tagged as Possums by our Artificial Intelligence system. This actually represented only 60 possum unique visits (each possum visit averaged about 6 videos). Looking through 4000 videos manually would obviously be very time consuming but the combination of our deep learning model and recording analysis software worked out the 60 visits for us. We were keen to ensure there was no errors in that data before we published this, so we verified that each of those was indeed a possum interaction with the traps (it’s ok, we don’t really mind watching possum videos).
So what did the traps catch?
We were rather surprised to see that the end result over 3 months was zero catches of possums but one of a bird (an unfortunate finch) in the AT220 trap.
The zero catches is obviously disappointing but is actually pretty much in line with the typical behaviour we see around most traps. All the evidence we’ve gathered so far suggests that a common predator behaviour is to simply walk past a trap. We don’t like making statements like that without providing some good evidence, so below is a video we’ve pulled together that shows a compilation of the typical possum behaviour. It’s worth taking a minute to watch.
As you’ll see, even when the possums do find the traps interesting, they don’t find the traps interesting enough to actually interact with in a way that might trigger the trap. Of course this is just one test so we can’t draw statistically conclusive results from it. But here at Cacophony, we have now watched tens of thousands of videos of predators just walking past existing traps. If that conclusion feels unlikely to you, note that we know our camera is more sensitive than normal trail cameras so if you are using normal trail cameras you may not be seeing many of the predators wandering around (and not being caught). Normal trail camera placement best practice is 1.5 meters from device - this risks missing a lot of what is happening close to the trap.
Our goal at The Cacophony Project is not to rate the effectiveness of traps and lures but to create tools everyone can use to improve predator monitoring and trapping. We are very reluctant to show results like this as we do take our hat off to everyone trying to do innovative things in the trapping space. Both these traps are likely to be as effective as other possum traps without the need for as much labour so they are definitely a step in the right direction but it seems that this approach is only going to be useful for predator suppression. Our goal is total elimination.
We do quite like the possibility of using the elimination method of the AT220 when combined with a more open ground based trap combined with some artificial intelligence to ensure the decisions the trap makes are well-informed.
We are constantly gathering more evidence to suggest that the biggest problem with trapping is very low trap interaction rates and we are keen to get more people focusing their ideas and skills on this part of the problem.
Much more on this soon.