This is an audio long read from nature.
In this episode, I really get outside.
Scientists ditch fieldwork in the age of AI.
Written by Aisling Irwin and read by me, Benjamin Thompson.
Tadeo Ramirez Parada studied the timing of plant flowering for his PhD, but he didn't touch a single petal.
Instead, he developed a machine learning algorithm to analyse the digitised captions of one million herbarium species, which showed him how flowering times are changing with rising temperatures.
Ramirez-Parada's work has helped to solve an important mystery in ecology, showing that as temperatures change, plants shift their flowering times to cope with the heat rather than adapting through natural selection.
Yet his work so far has been almost entirely computer-based.
I had to do very little experimental or field work, says Ramirez Parada, who did his PhD at the University of California, Santa Barbara.
Ramirez Parada's work is typical of a change that is reaching into every part of ecology.
Whatever scientists are analysing digitised specimens, images of the natural world, DNA samples or data streaming in from sensors many are doing it indoors.
The technologies are creating a world that can be monitored at times, places and scales that were previously unimaginable.
We are moving towards the quote.
Fully automated monitoring of ecological communities, wrote Marc Besson, a marine scientist at the Sorbonne University Ocean Observatory in Banyuls-sur-Mer France, in a 2022 paper.
Many ecologists say this revolution offers huge potential for understanding the biodiversity crisis and discerning patterns of global change.
But some ecologists are dismayed.
They feel that the discipline is losing intimacy with its subject matter.
They argue that field experience is in decline and that this loss could lead to error, bias and oversimplification of results.
If it becomes a world where you don't actually have to go out in order to become an ecologist, we kind of lose sight of what the actual world is like, says Bill Sutherland, who studies conservation biology at the University of Cambridge, UK.
Like scientists everywhere, ecologists are grappling with how to make the most of a torrent of data.
Natural history museums and herbariums around the world have digitised more than one billion specimens over the past few decades, some with accompanying DNA records.
Meanwhile, citizen scientists and researchers alike have been feeding databases such as iNaturalist with hundreds of millions of observations, which are absorbed into the Global Biodiversity Information Facility, or GBIF, a central database for natural history.
There is also a stream of data from sensors such as camera traps which take pictures when activated by movement microphones, animal tracking devices drones, satellites and DNA samplers.
Such sensors can run for years without intervention.
Once, a remotely planted camera trap would eventually run out of power.
Now, the energy consumption of such a device is minimal, and it can rely on solar or wind energy.
Bandwidth is no longer an obstacle to data being transmitted 24 hours a day.
And computer science is more than keeping up.
Artificial intelligence systems are already identifying species from these data.
They are also being used for more complicated tasks, such as building species distribution models and ancestry trees.
Some ecologists predict that generative AI, which creates new content based on learning from huge datasets, will soon be able to make more complex models, leading the way to understanding ecological processes and forecasting how species will respond to environmental changes.
There are already at least 100 laboratories that would label their work as AI for Nature, according to Tanya Berger-Wolf, a computational ecologist at The Ohio State University in Columbus.
The approach is starting to bear fruit.
One European project called CamAlien is tracking invasive species using high-resolution cameras with machine learning processing power affixed to cars, boats and trains.
As they speed along, they rapidly photograph the sides of roads and tracks, analyse the images in situ and upload alerts about alien invasive plants to a Europe-wide online marketplace.
The system shows how, just in the past few years, new technologies combined with AI have quote gone from mostly demonstrating potential to actually beginning to deliver real implementations, says Toga Thomas Hoyer, an ecologist at Aarhus University in Denmark who co-developed CamAlien.
Some 16 European countries are trying out the technology to assess the distribution of invasive alien species.
Similarly, amid the steep declines in some insects, a consortium of researchers has finessed camera trap technology, originally designed to spot mammals, so that it can identify and monitor insect species which are much more numerous.
Automated insect monitoring didn't exist five years ago, says Hoyer.
Thanks to developments in AI, scientists can distinguish between thousands of species.
It's opening up a door to part of our natural world that is so much more diverse compared to what camera traps have been used for previously, says Hoyer.
He and his group think that making insect monitoring easier and less labour-intensive will shed light on the state of insect populations around the globe.
Another group has deployed a system of microphones in search of a more detailed understanding of migration, as birds fly across Europe from Norway to the Mediterranean coast of Spain.
Known as the TABMON project, it is now streaming real-time soundscape data day and night.
An AI tool analyses the data and converts them into commonly used biodiversity indicators.
Having standardised ecological data on continental scales is extremely rare, says Sarab Sethi, who studies ecosystem sensing at Imperial College London and led the design of the microphones.
Quote especially when it's on the fine-scale temporal resolution that acoustics gives across a wide range of species and across multiple years.
End quote.
The project has yet to report its first results.
Fewer dispute the benefits of more data and detail, but there is an ominous side effect, says Kevin Gaston at the University of Exeter UK, who studies people's relationship with nature.
Field experience is on the wane.
Gaston and his co-author, Masashi Soga, who studies the loss of human-nature interactions at the University of Tokyo, argued in a March 2025 paper that there has been an extinction of experience, a widespread decline in fieldwork-based research and education, with knock-on effects on the depth of ecological understanding.
They also flagged other dangers, such as reduced engagement with local communities, a practice known to be crucial for successful conservation.
Others have expressed concern about AI colonialism, a practice in which data collected remotely in poorer countries are siphoned off for analysis in well-equipped labs elsewhere.
There are few quantitative data available to support or challenge Gaston and Segar's argument.
One analysis of ecological studies published between 1980 and 2014 found that fieldwork-based studies decreased by 20 as a proportion of the total, whereas modelling and data analyses increased by 600 and 800 respectively.
But these are relative changes rather than absolute numbers, and the dataset ends more than a decade ago.
Anecdotally, however, Gaston and Segar's paper struck a nerve.
Since publication, a number of groups have cited it, while warning that a lack of outdoor research is hindering studies on subjects ranging from solitary bees to dinosaur fossils.
There's also anecdotal evidence that more computer scientists have entered ecology, excited about what they can offer, but lacking field experience.
That was the case for Berger-Wolf, considered a founder of computational ecology.
She completed her PhD in theoretical computer science but being married to an ecologist says she would chat to others in the ecology community and walk away quote with a feeling like
Oh, there's got to be a different way of answering this question, end quote.
Berger-Wolfe changed tack in 2003 and by 2005 was developing algorithms for dynamic network analysis to depict the social interactions of zebras in the Kenyan Serengeti.
Field colleagues urged her to go and see her data, but she always refused.
Quote I'm a city girl and I don't like dust and bugs, and my answer was no.
My data looks beautiful on my screen.
End quote.
Sethi is another convert to ecology, having arrived in the field with an engineering background.
In 2016, he decided to apply acoustic monitoring to ecology for his PhD, but the self-confessed metrophile quickly found himself out of his depth in a rainforest in Malaysian Borneo.
I did what I now realise was the extremely dumb thing of trying to develop a new technology and for its first deployment to be in a tropical forest on the other side of the world.
Sethi grins.
On the first night he lay under a mosquito net in a pitch-black hut on stilts, wide awake, while his ecologist colleagues dozed comfortably amid the sounds of the rainforest.
He remembers thinking, quote, my God, is this just like a joke that's gone a bit too far, end quote.
Now, he values his field experiences, but works mostly from the lab.
Some ecologists have gone the other way, coming in from outdoors to embrace big data.
Laura Pollock at McGill University in Montreal Canada, began her career as a field ecologist first in the swamps of New Orleans Louisiana, and then in isolated mountain regions in Australia.
She saw a need for ecologists to do better data analysis, and now she uses machine learning to do predictive modelling of biodiversity across landscapes.
I rarely get outside, she says.
I'm trying, but it's really hard because there's so much technology creating so much data that we need people who have these data science skills to analyse this.
But Besson has embraced technology without diminishing his hours in the field.
He says that he is spending as much time outside as he did before automation arrived.
Quote.
Cameras and hydrophones can capture things in addition to my own eyes and ears, and they can stay out in the field when I need to go back to the lab and when I need to sleep.
End quote.
There are also many systemic forces driving ecologists indoors, argues Gaston.
There's a widespread perception that funding for field studies is in decline, although the data are not often differentiated into grants for fieldwork versus those for lab-based projects.
Scientists who run long-term ecological studies in particular report that they struggle to find funding.
Other contributing factors include the fact that research institutes are increasingly in urban areas, that more scientists have childcare responsibilities that deter them from doing long or far-flung trips, that many feel the need to reduce their carbon footprint and that others want to avoid helicoptering in and out of a country to do fieldwork that local scientists could do.
Another major issue, says Sutherland, is that the fast track to career-boosting publications is to analyse, rather than physically collect data.
Supposing you do your PhD and you spend all your time doing fieldwork, he says, and the person sitting next to you has been extracting data from day one.
After three years, he says they might have published in increasingly highly ranked journals.
While quote you're still in the Amazon catching fish.
End quote.
Quote and then your papers, because they're considered case studies, probably won't get into prestigious journals.
End quote.
Other ecologists say they were advised against field-based study because it was not a reliable way to get tenure-track positions.
Ramirez-Parada agrees that the novelty of large-scale data analysis gets noticed.
There's definitely a shininess to it that gets rewarded by high-impact journals, and that is something that I think everybody rolls their eyes at a little bit, even the people doing it.
Some leading ecologists, both computational and field, think undervaluing fieldwork could send ecology as a whole badly off track.
The most obvious reason for this is that algorithms need training data, and the production of such data at high quality requires experts who know their species and can discover new ones.
In the first years of Sethi's acoustic monitoring work, he thought that it could provide a shortcut to assessing biodiversity.
The typical approach required him to feed vast quantities of data that linked sounds to particular species into an algorithm requiring time-consuming work in the field.
So he tried a top-down approach, inferring a landscape's biodiversity from its overall soundscape.
As more ecologists picked up on the idea, Sethi became uneasy about its reliability.
His group used some large datasets to compare the approaches and concluded that it's crucial to validate results using field data from the study site.
Since biodiversity sounds different everywhere, you can't really skip the stage of calibration and validation, he says.
If Sethi hadn't worked in the field, he says, he might have missed this insight.
I saw firsthand how much of nature was just undocumented, let alone at the level where you could start to build autonomous technologies to sense it.
He says.
Field data are both crucial and lacking.
Although reams of data have been gathered by citizen scientists, they tend to be from high-income countries, the fringes of urban areas and easily spotted species.
They are, quote, super biased, says Berger-Wolfe.
You need data on species quote that are rare, that are unobserved, that are at the edges of the ecosystems end quote.
Three years into her work on zebras, Berger-Wolfe finally visited the Serengeti and had an epiphany about fieldwork.
She was amazed that the field scientists were able to cut through the complexity of zebra social life and extract the hard data they were sending her.
How did they work out the different groups of zebras, or which females were partnered with, which males or which animals were friends?
All my assumptions were wrong, she says.
All the messiness of biology got in and I'm like we need to redefine how we're doing the analysis completely.
The experience also convinced her that the next generation of computational ecologists needed field experience.
An endeavour she began within two years of visiting the Serengeti by sending her own team on a tropical ecology course.
The need for immersion to gain understanding is something that David Lindenmayer, an ecologist at the Australian National University in Canberra, who is renowned for his long-term ecological studies, feels deeply about.
Lindenmayer spends days at a time in the forest.
Quote, End quote.
Gaston and Segar's warnings about the extinction of experience yielded a wave of correspondence from colleagues who shared their concerns.
But one reply took a different perspective.
Rafael Guariento, an ecologist at the Federal University of Mato Grosso do Sul in Campo Grande Brazil, is optimistic.
He sees a field that is maturing, one that, in common with other fields of science, is so diverse that it would be illogical to require everyone to be hands-on.
Specialisation often happens once a field gets to a certain level of complexity, he argues.
It's a promising picture for the future of ecology.
Marine scientist Besson, who is comfortable working with both algae and algorithms, thinks the future of ecology lies with versatile scientists who are adept at working both inside and outside.
He says that ecologists need to take advantage of technologies to collect diverse data more frequently.
But they also need to develop computer science expertise to help extract ecological meaning from those data.
Quote, I believe that ecologists simply need to adapt.
End quote.
To read more of nature's long form journalism, head over to nature.com slash news.
