Not long ago, an ecologist doing a badger survey was a boots-and-patience job, walking field margins at dawn looking for setts, latrines, hair snags and tracks, while making tally marks on a paper form. Data lived in binders, destined for the office where it would be transcribed into mapping software. It was slow and tedious, a process that Andrew Speer, the chief technical officer with ecological consultancy Scott Cawley Ltd, describes now as a “waste of time”.
That labour-intensive world is fading fast. In its place, a technological and AI revolution is under way, involving thermal cameras, soil microphones, drones, eDNA, acoustic recorders, remote-sensing satellites, infrared sensors and machine-learning algorithms. It’s changing what we can learn about the world around us and what role humans will play in an age when a drone can survey a clifftop seabird colony in minutes and AI can identify thousands of species in the soil in seconds.
Perhaps the most striking frontier is the acoustic one. Microphones inserted into the soil can pick up the movements and feeding sounds of invertebrates such as earthworms shifting through the earth and larvae chewing. Researchers at Baker Consultants, working with Warwick University, are developing standardised hardware that non-specialists can use, with 2,000 recordings now gathered from farms across the UK. Underground, it turns out, has a sound, and healthy soil sounds different from degraded soil.
For bird researchers, measuring biodiversity used to involve standing outside for hours, noting any birds seen or heard, and repeating it over 24 hours at various locations. All of this can now be automated in real time. In Borneo, researchers at Imperial College, London, set up an acoustic monitoring network that records the sounds of the forest – including bird song, monkeys moving through the tree canopy, rainfall – producing thousands of hours of data.
READ MORE
This was transmitted wirelessly and fed into Google’s AudioSet machine-learning tool, which can identify “audio fingerprints” for different types of sound. They trained it on the forest recordings to build the sound of a healthy ecosystem. From the soundscape alone, the AI could predict indicators such as habitat quality and biodiversity.
For Dr Aisling Moffat, a soil ecologist in UCD, technology has been transformative. Soil biology – the teeming world of bacteria, fungi and invertebrates, from nematodes and collembola to earthworms, living in vast numbers in every teaspoon of earth – has long been under-studied compared to the more charismatic life above ground. But image annotation software is changing that.
Identifying soil life once meant hours of Dr Moffat’s time at a microscope, clicker in hand, as she painstakingly sorted through samples. She can now scan samples with a high-resolution scanner, upload the images to an AI platform and get species identification and abundance counts in a fraction of the time. Across 200 field sites in Ireland, she mapped the relationship between soil geochemistry and biodiversity. It’s research at a scale that simply wasn’t possible before.
[ Making the invisible visible through the use of technologyOpens in new window ]
Camera traps with AI classifiers can identify foxes, badgers, and deer from hours of overnight footage. Drones with thermal imaging can auto-classify habitats, and bat detectors record calls and images that a software programme can identify by species and number.
During a recent bat survey, Speer recorded 47 hours of footage and used AI to extract the relevant clips. Speer’s team audited the results by hand anyway, because, for all the promise, Speer doesn’t yet believe the tools are trustworthy enough to stand alone, especially given that planning applications and judicial reviews require an ecologist to be able to stand over the integrity of every data point. Treat AI like an intern, Speer says: always double-check it.
For now, the automation is on data collection and processing. The interpretation, the argument, the ecological judgment – writing impact assessments, weighing up competing evidence in a planning dispute – these are still human acts. Whether machines will eventually do these jobs as well is an open question, as is whether anyone will still be checking when they do.
There is an irony here that ecologists are not blind to: the AI systems now mapping the natural world at unprecedented scale are themselves voracious consumers of energy and water, with data centres identified as one of the few sectors where emissions are set to grow rather than fall.
[ We need to talk about AI’s staggering ecological impactOpens in new window ]
The deeper problem that has yet to be solved is a material one: decades of publicly funded ecological survey data – from infrastructure projects, national surveys, consultancy reports – sit in filing cabinets and pdf documents on office shelves, undigitised, invisible to any algorithm. In Ireland, there is no single national digital repository where this information accumulates into a picture of biodiversity change over time, something that could and should be a condition of all publicly funded ecological work.
Planning decisions – for example, whether an industrial wind farm should be built on a bog surrounded by legally protected habitats – are being made on the basis of incomplete scientific knowledge, not because the historical data doesn’t exist, but because no one has yet dusted off this rich trove of documents and made them available online.
Perhaps the greater challenge – the one that could help nature the most, and soonest – is learning from what we already saw.
