Artificial Intelligence in conservation

Since the advent of Chat GPT 4, the culture has been obsessing over Artificial Intelligence. What might change? How could AI help us? Could AI turn against us?

In science and technology circles, however, AI has been a central talking point for decades. From robotics to machine learning, AI assistants to genetic algorithms, the possibilities are endless. In the last few years, Bush Heritage Spatial Data Officer, Eva You, has also noticed a marked increase in how many of her colleagues are talking about using AI as a tool to protect our landscapes. 

“Maybe it's the circles that I'm in, but everyone has an opinion on it. Whether it works, or whether it's overused, or whether it's underutilised... and questions over its efficiency,” says Eva. “It's like the invention of the wheel.” 

Eva herself is using AI to test whether satellite imagery or drone imagery is better at mapping invasive buffel grass distribution. The research will help determine the accuracy of the mapping and ultimately contribute to innovative ways of controlling the colonising grass species.

“Buffel grass is an introduced species. It's an ecological transformer. It results in monoculturalisation of the environment, and it increases fire risk. It can be attributed to the same economic and environmental cost as things like pests, such as rabbits, foxes, and cats.” 

“My aim is to compare satellite imagery with UAV (drone) imagery and quantify the actual extent to which uncertainty is reduced, and whether that weighs up the cost and benefit for a land manager,” she says. (see The Conversation for more on this) 

Colin Broughton, a Conservation Systems Developer for Bush Heritage, is solving a different problem, which is already having tangible time-saving benefits for field staff. 

“Remote sensor cameras, or ‘camera traps’ are used fairly intensively across Bush Heritage Reserves,” says Colin. “One of the main uses is the detection of feral animals, but also as a monitoring tool for native animals. With all that camera trapping, one of the big issues is how to process all the data.” 

These camera sensors are triggered by movement. Since the birth of the technology, a common vexation for land managers and ecologists is the hours spent trawling through tens of thousands of images of a meagre strand of swaying grass or wildflower.  

“It's painful and it's not a good use of people's time,” says Colin. “But that problem's pretty much been solved. Microsoft has made a machine learning model called Mega Detector. It can tell if there's an animal present in the image, so it means we can filter out the rest.” 

Colin and his team figured out how to take this technology to the next level, training models to detect common animals picked up by cameras. When the model has done its job, the land manager or ecologist is left with a spread sheet detailing what the camera picked up i.e. three red foxes, four feral cats, one Western Quoll.

So, what could the future look like in this space? Just how much could land managers automate?  

“Say if camera trapping was integrated with other remote monitoring methods like acoustics and even weather data, you could combine factors together and make predictions like, ‘based on XYZ, we predict that XYZ will happen.’ For example, we predict feral pigs are going to be in that wetland in a couple of weeks, so you might want to go out there and keep an eye on it. That could be where it’s headed,” says Colin.  

What we do know is that innovative uses of AI in conservation are already having positive impacts on how much land managers can achieve, and considering the mounting challenges faced by the environment in 2024, that can only be a good thing.