The research, published in the Proceedings of the National Academy of Sciences and supported in part by funding from the U.S. National Science Foundation and the U.S. Department of Agriculture, demonstrates a proof of concept with real ecological payoff. By deploying large language models, the conversational AI systems behind tools like Claude and ChatGPT, the team identified comments containing pertinent ecological information, extracted the relevant details, and categorized them into structured datasets. In the process, they uncovered evidence of non-native plant species quietly creeping into ecosystems along America’s West Coast, signals that might have been missed entirely by conventional monitoring.

The core challenge the researchers set out to solve is one of the most stubborn in ecology: measuring species interactions. Modern technology has made it relatively easy to track individuals and groups of a single species in the wild, through camera traps, acoustic sensors, and citizen-science checklists. Interactions between different species are a far harder target. A butterfly pollinating a flower, a hawk eating a songbird, or two species of geese feeding side by side all require someone to be in the right place at the right moment, watching.

“In many cases, species interactions require direct observation and that’s really hard to come by simply because you can’t keep a lot of people in the field monitoring the same patch of an ecosystem for 24 hours a day, seven days a week,” said Hengxing Zou, the study’s lead author, who performed the work as a postdoctoral scholar in the University of Michigan Institute for Global Change Biology. Yet these interactions matter enormously, he noted, because they are tied to ecosystem function, ecosystem stability, and many of the things ecologists care about most, especially as the global environment changes.

The solution, the team realized, was already out there in abundance. Instead of a handful of trained ecologists observing a single location nonstop, they could tap observations already made by a community of hundreds of millions of nature enthusiasts. On platforms like eBird and iNaturalist, users upload observations along with crucial metadata, such as when and where the sighting occurred. Often, they also add free-text comments, describing a butterfly landing on a flower, a bird eating another bird, a flock of mixed species feeding cooperatively, or two birds squabbling over a spot at a feeder. These comments are exactly the kind of direct interaction records ecologists crave, but they are buried in millions of entries written in casual, unstructured human language.

That is where natural language processing comes in. The researchers suspected that AI tools designed to understand human language could do more than quickly flag which observations had comments; they could also determine which comments actually documented interactions, even when the observers never used formal ecological vocabulary. Working with undergraduate researchers from ecology and computer science, the team crafted prompts that led both Claude and ChatGPT to extract and classify species interactions with good accuracy and precision, Zou said.

To illustrate just how transformative the time savings can be, the team ran a direct comparison. First, ChatGPT was asked to isolate comments from eBird data that contained species interactions. Then Olivia Stein, an undergraduate research assistant, took a random sampling of nearly 500 of those comments and manually classified which type of interaction each one contained. The team then gave Claude the identical task. What took Stein eight hours of careful human work, the AI completed in six minutes. The result was not a replacement for human expertise but a dramatic extension of it, multiplying the reach of a trained eye by orders of magnitude.

The study included two case studies that showcase the versatility of the approach. The first focused on plant-pollinator interactions involving the West Coast lady butterfly, which pollinates plants along the west coast of the United States and Mexico, using observations recorded on iNaturalist. Here, the key interaction was often as visually straightforward as the butterfly landing on a flower. The second case study examined the wider array of interactions among Michigan birds found in the comments of eBird observations, ranging from predation and being preyed upon to cooperative behaviors like flocking and competitive behaviors like fighting over a perch at a bird feeder.

“This approach gives us something ecologists have long needed: a way to examine species interactions across broad spatial and temporal scales,” said Kai Zhu, a senior author and associate professor in the University of Michigan School for Environment and Sustainability. “AI can help turn scattered observations into ecological datasets that reveal how biological communities are responding to global change.” Brian Weeks, also a senior author and associate professor at the school, said he was surprised and delighted by the wide range of natural history observations that members of the public noted. Beyond their value as scientific data, he added, a highlight of the project was glimpsing the joy that observing birds clearly brings to people.

Zou is candid about the controversies swirling around artificial intelligence in research, but he frames this project as an example of AI being used to extend human effort rather than replace it. “I personally think it’s kind of a rustic way of using AI in this booming age where people are using it to solve Millennium Prize math problems,” he said. “It’d be almost impossible for humans to go through the millions of comments that are out there. We’re using the large language models to go through and generate data sets, instead of analyzing them or replacing any of the creative work that scientists are doing.” In other words, the machines handle the tedious mining; the scientists keep the interpretive and creative work.

There is also a pleasing historical resonance in the method. While large language models are brand-new technology, the observations they harvest reflect the founding spirit of ecology itself, a tradition of curious people going out into nature, watching closely, and recording whatever they find interesting. Zou believes far more information could be recovered from these kinds of comments beyond species interactions, potentially opening new lines of inquiry into behavior, phenology, and community dynamics. The research was a broad collaboration, with contributions from the University of California Santa Cruz, the University of California Davis, the Georgia Institute of Technology, the American Bird Conservancy, Notre Dame University, the Santa Fe Institute, and Michigan State University. Additional funding came from the David and Lucile Packard Foundation, the Alfred P. Sloan Foundation, the Institute for Global Change Biology, the University of Michigan Undergraduate Research Opportunity Program, and Michigan State University’s Ecology, Evolution, and Behavior Program, with computational resources provided by Advanced Research Computing at the University of Michigan. For the growing community of birdwatchers, butterfly chasers, and backyard naturalists, the message is clear: that casual note you leave alongside your photo may soon become a data point in the science of a changing planet.

Subject of Research: Using large language models to extract species interaction data from citizen-science observation comments

Article Title: How nature lovers can aid ecology with an assist from AI

Article References: How nature lovers can aid ecology with an assist from AI. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: artificial intelligence, large language models, ecology, species interactions, citizen science, eBird, iNaturalist, biodiversity, ecosystem monitoring, natural language processing, University of Michigan, PNAS

The problem the researchers tackled is fundamentally physical. Eddy-covariance instruments measure the covariance between vertical wind speed and gas concentration at high frequency, typically 10 to 20 hertz, to capture the turbulent eddies that carry gases between the surface and the atmosphere. Yet no instrument is perfect. Limited sampling time, line averaging, the physical separation between the sonic anemometer and the gas analyser’s inlet, and the passage of air through tubes and filters all act like a low-pass filter, smoothing out the fastest fluctuations. Because small eddies contribute disproportionately to the total flux, this smoothing causes a systematic underestimation that can range from a few percent to tens of percent depending on the setup, the site and the weather. Correcting for it is one of the most stubborn technical challenges in micrometeorology.

Two experimental strategies dominate the field. The spectral approach computes the ratio of the attenuated gas concentration power spectrum to the unattenuated sonic temperature spectrum, deriving an empirical transfer function from which a correction factor follows. The co-spectral approach instead uses the co-spectrum of the gas flux itself relative to the sensible heat flux, folding every attenuation source, including sensor separation, into a single empirical characterisation. Each camp has its defenders: spectral methods, championed by widely used software such as EddyPro and embedded in the official ICOS processing pipeline, avoid the complications of time-lag and phase-shift estimation, while co-spectral methods are immune to high-frequency noise and need no separate sensor-separation term. Until now, no systematic, network-wide comparison of the two had ever been attempted.

Ariane Faurès of the University of Liege and her colleagues filled that gap using a full year of carbon dioxide and water vapour flux data from 38 ICOS Class 1 and 2 ecosystem stations, all equipped with an identical enclosed-path LI-7200 gas analyser and a Gill sonic anemometer. This standardisation was a scientific gift: because every site shares the same hardware, the same heated 0.73-metre sampling line, the same rain cap and inlet filter, and flow rates fixed between 10 and 16 litres per minute, differences between sites could be attributed to atmospheric conditions and ecosystem type rather than instrument quirks. The team classified stations into high-measurement-height sites, essentially forests, and low-measurement-height sites such as croplands, grasslands and non-forested peatlands, then computed cut-off frequencies and correction factors for both methods under carefully controlled wind speed, stability and humidity classes.

The first headline finding is reassuring: the standardised ICOS setup performs remarkably well. Cut-off frequencies for carbon dioxide generally exceeded 1 hertz and were strikingly consistent across the continent, with a mean of 2.1 hertz and a tight spread at a wind speed of 3 metres per second. Correction factors for carbon dioxide typically stayed below 1.2, meaning the fluxes needed boosting by less than 20 percent, and often by less than 10. For water vapour the picture was harsher: attenuation was stronger, with correction factors occasionally approaching 2 under humid, windy conditions, reflecting the well-known tendency of water molecules to adsorb onto and desorb from the surfaces of the sampling tube and filter in a manner that varies with relative humidity.

When the two correction methods were compared directly, the differences for carbon dioxide proved small. After the standard turbulence filtering that removes weakly turbulent periods, cumulative annual flux differences between the methods typically stayed below 2 percent, and at high forest towers they were often negligible. The dominant source of carbon dioxide attenuation turned out to be sensor separation, the spatial offset between the anemometer and the analyser inlet, which the spectral approach must add back through a theoretical formulation from Horst and Lenschow. Water vapour told a different story. Here the two methods diverged substantially, with cumulative flux discrepancies frequently reaching 5 to 10 percent, and the largest gaps occurring under stable atmospheric stratification, at low measurement heights, and under strong attenuation.

The reasons for this divergence are methodological as much as physical. Both approaches rest on the assumption of spectral similarity, the idea that gas and temperature fluctuations share the same spectral shape at low frequencies. The study found that this assumption holds well for carbon dioxide but breaks down for water vapour, particularly in humid air, where the gas spectra departed from the ideal reference far more strongly than the co-spectra did. On top of that, the spectral approach was repeatedly plagued by high-frequency noise, requiring a dedicated denoising procedure at 58 percent of sites for carbon dioxide and 84 percent for water vapour, while the theoretical sensor-separation term it depends on produced unrealistically large corrections under stable conditions. The co-spectral approach, by contrast, delivered stable, physically consistent results across all sites and conditions, leading the authors to recommend it for this specific experimental configuration.

Perhaps the most provocative result emerged when the team compared their independently computed corrections with the official fluxes produced by the ICOS Ecosystem Thematic Centre. Since the network pipeline also uses a spectral method, one might expect close agreement, but the opposite was true: differences between the study’s spectral corrections and the official products were often larger than the differences between the spectral and co-spectral approaches themselves. For water vapour, cumulative flux differences against the ICOS pipeline averaged 10.7 percent at low-measurement-height sites, with some stations showing gaps exceeding 20 percent, and even carbon dioxide showed average differences of 1.6 percent and station-level extremes above 10 percent. The causes are a cascade of small choices: different low-pass and anemometer correction equations, a different sensor-separation formulation, different filtering thresholds and fitting ranges, and a contested use of the transfer function when computing correction factors, a choice that earlier work suggests can bias fluxes by up to 10 percent.

These numbers matter far beyond the technical community. Eddy-covariance fluxes feed the global carbon budgets that inform climate policy, underpin assessments of whether forests and peatlands are net sinks or sources of greenhouse gases, and enter the long-running debate over the surface energy balance closure problem. A systematic 10 percent uncertainty in water vapour fluxes propagates directly into estimates of evapotranspiration and ecosystem water use, while even a few percent in carbon dioxide can shift the calculated carbon balance of a young forest or a drained peatland. The study also carries a practical design lesson: because attenuation grows as the measurement height approaches the canopy, mounting sensors as high as feasible, within footprint constraints, dramatically reduces the correction burden, and at many forest towers the carbon dioxide correction is now so small it could arguably be skipped altogether.

The authors have released their analysis as an open-source Python tool, FreqCor, designed to be configurable across setups and trace gases, and they argue that the high-frequency correction step deserves a broader reassessment within the ICOS processing pipeline. They are candid that a fully automated, network-scale procedure remains elusive: thresholds, fitting ranges and denoising decisions still involve user judgement, and future work should replace uniform flux thresholds with signal-to-noise criteria and validate denoising across many sites. The study’s scope is limited to enclosed-path analysers, and the authors note that closed-path systems with long tubes, common for methane and nitrous oxide measurements, could show even larger method differences and merit a similar network-wide test. For now, one thing is clear: the quiet correction applied inside every flux tower’s software is not a formality, and choosing the right way to do it can change what Europe’s ecosystems appear to breathe.

Subject of Research: Comparison of spectral and co-spectral high-frequency attenuation corrections for eddy-covariance flux measurements across the ICOS ecosystem network

Article Title: Spectral versus co-spectral correction of eddy-covariance fluxes: a multi-site assessment across the ICOS network

Article References: Faurès, A., Fratini, G., Nicolini, G., Papale, D., Sabbatini, S., & Heinesch, B. (2026). Spectral versus co-spectral correction of eddy-covariance fluxes: a multi-site assessment across the ICOS network. Atmospheric Measurement Techniques, 19(18), 6037-6074. https://doi.org/10.5194/amt-19-6037-2026

Image Credits: AI Generated

DOI: 10.5194/amt-19-6037-2026

Keywords: eddy covariance, ICOS, spectral correction, co-spectral correction, high-frequency attenuation, carbon dioxide flux, water vapour flux, gas analyser, sensor separation, micrometeorology, ecosystem monitoring, Atmospheric Measurement Techniques

The submission took an unusual and deliberately democratic form. Rather than speaking with a single voice, AMI combined a survey reflecting the majority view of participating members with a detailed supporting document that sets out the scientific case for greater recognition of marine microorganisms in European ocean policy. The society’s central message is unambiguous: marine microbes must be explicitly recognised as fundamental components of ocean ecosystems, and their needs and contributions must be considered across biodiversity conservation, ecosystem management, ocean-climate policy and the assessment of overall marine health. In other words, an ocean act that only protects what can be seen with the naked eye is, in the view of these scientists, an act that protects only a fraction of what matters.

The scientific rationale for this position is compelling. Microorganisms, including phytoplankton, bacteria, archaea and viruses, drive many of the biological and chemical processes that regulate the ocean. They are responsible for recycling around half of global primary production, a staggering figure that underscores just how much of the planet’s carbon and nutrient throughput depends on organisms too small to see. Through the biological carbon pump, phytoplankton fix carbon dioxide at the surface and export it to the deep sea, while the microbial carbon pump converts dissolved organic matter into long-lived reservoirs that can keep carbon out of the atmosphere for centuries. Microbes also orchestrate the cycling of nitrogen, phosphorus and iron, elements that determine how productive marine food webs can be and how much carbon the ocean can ultimately sequester.

The urgency of this agenda is growing as the ocean changes. As marine waters warm, acidify and lose oxygen, microbial communities are being pushed into conditions they have not experienced on human timescales. Because microbes reproduce quickly and respond rapidly to shifting temperatures, pH levels and nutrient supplies, changes in their community composition and metabolic function often precede visible changes in larger organisms and habitats. Understanding how these communities respond to ocean warming, acidification and deoxygenation will therefore be increasingly important for predicting future shifts in ecosystem function, for anticipating how climate regulation services may weaken, and for designing interventions that support microbial processes capable of buffering environmental stress rather than amplifying it.

Beyond free-living microbes, AMI’s response draws attention to the intimate partnerships between microorganisms and larger marine life. Corals, sponges, seagrasses and fish all exist in close association with complex microbial communities that can influence nutrition, disease resistance and resilience to environmental stress. The coral microbiome, for example, can determine whether a reef survives a bleaching event, while the microbial partners of seagrasses are essential for nutrient uptake in the sediments that anchor these meadows. AMI argues that recognising these host-associated microbiomes could materially strengthen conservation and restoration efforts, by ensuring that the microscopic processes underpinning ecosystem health are not overlooked when protected areas are designated, when species are translocated, or when damaged habitats are actively restored.

There is also a powerful economic argument embedded in the society’s submission. Marine microbial communities represent a largely unexplored reservoir of genetic and biochemical diversity, a living library of molecules shaped by billions of years of evolution in some of the most extreme environments on Earth. Potential applications range from novel medicines and industrial enzymes to new biomaterials and biotechnological processes, with microbial enzymes from cold-adapted and pressure-tolerant species already finding uses in industry and research. AMI stresses that protecting marine microbial biodiversity is therefore not only an environmental priority but also an investment in future innovation, feeding directly into the ambitions of the European blue bioeconomy and the continent’s broader strategy for sustainable growth grounded in living resources.

Translating this science into law, AMI offers a series of concrete recommendations. First, the society calls for strengthening the definition and assessment of Good Environmental Status, the benchmark against which European seas are measured, so that it explicitly considers microbial diversity, community composition, functional diversity and the ecosystem services that microbes provide. Second, the submission urges that host-associated microbiomes be formally recognised wherever they are important to the health and resilience of protected species, habitats and ecosystems. These changes would mean that monitoring programmes, conservation targets and restoration criteria under the Ocean Act would no longer treat microbes as an afterthought but as measurable, reportable components of environmental quality in their own right.

The response also argues that an ecosystem-based approach to ocean management must move beyond a focus on visible organisms and iconic habitats. Microorganisms drive the processes that make marine ecosystems possible, including nutrient cycling, carbon sequestration, primary productivity and decomposition. Their rapid responses to environmental change could provide early-warning indicators of ecosystem condition and recovery, offering policymakers near-real-time signals of stress long before fish stocks decline or habitats collapse. Incorporating microbial indicators into routine monitoring would give regulators a finer-grained, more responsive picture of ocean health, and could reveal whether management measures are actually working at the functional level where the ocean’s essential chemistry is performed.

By bringing together the expertise and perspectives of its global membership, AMI is advocating for an Ocean Act that recognises the full biological complexity of marine ecosystems and puts microbiology firmly within Europe’s vision for a healthy, resilient and sustainable ocean. The society, the oldest microbiology society in the United Kingdom with more than half of its members based outside the country, publishes The Microbiologist magazine and three internationally recognised journals in partnership with Oxford University Press, and it frames this intervention as part of its wider mission to amplify the collective influence of applied microbiologists in evidence-based decision making. The consultation response makes clear that the microbes drifting in every litre of seawater are not a footnote to ocean policy; they are its scientific foundation, and the legislation that will govern European seas for decades should say so explicitly.

Subject of Research: The role of marine microorganisms in European ocean policy and the proposed EU Ocean Act

Article Title: AMI calls for marine microbes to be at the heart of Europe’s future ocean policy

Article References: AMI calls for marine microbes to be at the heart of Europe’s future ocean policy. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: marine microbes, EU Ocean Act, Applied Microbiology International, ocean policy, microbial carbon pump, Good Environmental Status, blue bioeconomy, host-associated microbiomes, ocean acidification, biodiversity conservation, ecosystem monitoring, phytoplankton