What I learnt at the National Academy of Sciences US-UK Forum on Biodiversity / Jun 2025
I spent a couple of days at the National Academy of Sciences in the USA at the invitation of the Royal Society, who held a forum on "Measuring Biodiversity for Addressing the Global Crisis". It was a packed program for those working in evidence-driven conservation:
Assessing biodiversity is fundamental to understanding the distribution of biodiversity, the changes that are occurring and, crucially, the effectiveness of actions to address the ongoing biodiversity crisis. Such assessments face multiple challenges, not least the great complexity of natural systems, but also a lack of standardized approaches to measurement, a plethora of measurement technologies with their own strengths and weaknesses, and different data needs depending on the purpose for which the information is being gathered.
Other sectors have faced similar challenges, and the forum will look to learn from these precedents with a view to building momentum toward standardized methods for using environmental monitoring technologies, including new technologies, for particular purposes. -- NAS/Royal Society US-UK Scientific Forum on Measuring Biodiversity
I was honoured to talk about our work on using AI to "connect the dots" between disparate data like the academic literature and remote observations at scale. But before that, here's some of the bigger picture stuff I learnt...
[…2343 words]Validating predictions with ranger insights to enhance anti-poaching patrol strategies in protected areas / Jun 2025
This is an idea proposed as a good starter project, and is currently being worked on by Hannah McLoone. It is co-supervised with Charles Emogor and Rob Fletcher.
Biodiversity is declining at an unprecedented rate, underscoring the critical role of protected areas (PAs) in conserving threatened species and ecosystems. Yet, many of these are increasingly dismissed as "paper parks" due to poor management. Park rangers play a vital role in PA effectiveness by detecting and potentially deterring illegal activities. However, limited funding for PA management has led to low patrol frequency and detection rates, reducing the overall deterrent effect of ranger efforts. This resource scarcity often results in non-systematic patrol strategies, which are sub-optimal given that illegal hunters tend to be selective in where and when they operate.
The situation is poised to become more challenging as countries expand PA coverage under the Kunming-Montreal Global Biodiversity Framework—aiming to increase global PA area from 123 million km2 to 153 million km2 by 2030. Without a substantial boost in enforcement capacity, both existing and newly designated PAs will remain vulnerable. Continued overexploitation of wildlife threatens not only species survival but also ecosystem integrity and the well-being of local communities who rely on wildlife for food and income.
This project aims to combine data from rangers in multiple African protected areas and hunters around a single protected area (Nigeria) to improve the deterrence effect of ranger patrols by optimising ranger efforts and provide information on the economic impacts of improved ranger patrols on community livelihoods and well-being. We plan to deploy our models to rangers in the field via SMART, which is used in > 1000 PAs globally to facilitate monitoring and data collection during patrols.
The two main aims are to:
- develop an accessibility layer using long-term ranger-collected data
- validate the results of this layer, as well as those from other models developed, using ranger insights.
Habitat mapping of the Cairngormes Connect restoration area / Jun 2025
This is an idea proposed as a good starter project, and is currently being worked on by Isabel Mansley. It is co-supervised with David Coomes and Aland Chan.
The Cairngorms Connect is the largest landscape restoration project in the UK. Four landowners (RSPB, Wildlands Ltd, FLS, and NatureScot) embarked on a 200-year vision to restore over 600 km2 of land in the Cairngorms National Park with an emphasis on natural processes.
In July, 2023, the Centre for Landscape Regeneration commissioned a flight over a 400 km2 stretch of land over the area, collecting both high resolution RGBI imagery (0.1m ground resolution) and LiDAR data. Various research projects were built on this dataset, including studies into carbon cycling, shrub ecology, tree regeneration, and deadwood detection.
Existing habitat maps of the area are based on Sentinel 2 satellite data at a ground resolution of 10m. While this dataset provides a good basis for some research objectives, a habitat map that could leverage the high resolution of the aerial imagery would potentially be able to capture fine-scale variations in habitat structure more accurately. This project involves applying new developments in geospatial machine learning (specifically the Tessera one developed locally in Cambridge) to achieve this.
Evaluating a human-in-the-loop AI framework to improve inclusion criteria for evidence synthesis / Jun 2025
This is an idea proposed as a good starter project, and is currently being worked on by Radhika Agrawal. It is co-supervised with Alec Christie and Sadiq Jaffer.
Whenever we do evidence synthesis (especially for conservation outcomes) to distil the world's scientific literature into actionable insights, we have to decide on what published studies we will include or exclude, and why they are categorised as such. This can be a challenging process, and sometimes inclusion criteria may not be very reproducible or clearly defined, leading to confusion between reviewers and more time-consuming reviews.
In AI-assisted review methods, we are increasingly finding that LLMs may interpret inclusion criteria differently to human reviewers, potentially because human experts may implicitly assume certain things that are not obvious to those working outside the review team (or interpret things differently to fellow reviewers). We trialled an informal process earlier this year to iterate over the inclusion/exclusion criteria for an evidence synthesis using synthetic studies that represent "edge cases", whereby it is difficult to agree on whether they should be in or out. Through back-and-forth with an LLM, human reviewers were able to refine and improve their inclusion criteria.
This project will build on this work to develop a prototype, open-source tool that enables users to refine their inclusion criteria with the help of an LLM chatbot. This will be extremely useful for anyone conducting any type of evidence synthesis and so has great potential to be an impactful project beyond "just" the field of conservation.
Evaluating LLMs for providing evidence-based information on conservation actions / Jun 2025
This is an idea proposed as a good starter project, and is currently being worked on by Alex Wang. It is co-supervised with Alec Christie and Sadiq Jaffer.
We are building a Conservation Co-Pilot to improve worldwide conservation action through evidence-driven insights. Biodiversity loss is one of the biggest threats to our planet and to tackle it, we must improve the effectiveness of conservation action, which currently falls short of its full potential. This is because conservationists typically find it hard to access locally relevant evidence on what works to conserve biodiversity as research knowledge is not translated quickly or accessibly enough into policy and practice. We therefore need to accelerate the transfer of relevant, reliable evidence to decision-makers using more intuitive and interactive interfaces.
This project will use the comprehensive Conservation Evidence database (holding 8600 studies that have quantitatively tested 3600 actions) to evaluate the ability of a Mixture of Agents (MOA) approach, and/or individual LLMs, at providing rigorous evidence-based answers to priority questions from real conservationists.
This will extend our previous work that found that LLMs coupled with a hybrid retrieval strategy can answer multiple choice conservation questions as well as human experts. This will enable us to develop a "Conservation Co-Pilot" that can handle complex and nuanced questions from different users.
We become Junior Rangers at Shenandoah / May 2025
What might a Dame of the Realm, a Fellow of the Royal Society, the latest member of the UK Joint Nature Conservation Committee, and me all covet? That's right: a Junior Ranger badge from Shenandoah National Park! After an intense few days, Bill Sutherland, Julia P.G. Jones, EJ Milner-Gulland and I headed into nature to experience the spectacular landscapes of the Blue Ridge Mountains in Virginia and do some birding.
The National Park Service in the US runs a wonderful program for anyone aged 8+ (which we just about qualified for) to introduce people to nature, and Shenandoah is no exception. We visited the local ranger lodge in the park, and picked up a program booklet. They're full of activities for kids to do, but of course adults also pick up a lot of random knowledge (such as the endemic salamander species in the region).
Out-of-the-box LLMs are not ready for conservation decision making / May 2025
Our paper on how the careful design of LLMs is crucial for expert-level evidence retrieval has been published today in PLOS One and is available fully open access!
Our findings suggest that, with careful domain-specific design, LLMs could potentially be powerful tools for enabling expert-level use of evidence syntheses and databases. However, general LLMs used "out-of-the-box" are likely to perform poorly and misinform decision-makers. By establishing that LLMs exhibit comparable performance with human synthesis experts on providing restricted responses to queries of evidence syntheses and databases, future work can build on our approach to quantify LLM performance in providing open-ended responses.
In a nutshell, we tested 10 LLMs with six different retrieval strategies on their ability to answer questions related to conservation, benchmarked against the Conservation Evidence database that has been hand-assembled by experts over the last two decades. In some of the retrieval scenarios, models were only allowed to use their pretrained knowledge, whereas in others they had access to the relevant parts of the hand-curated database.
We found that language models had very varying results when relying only on their pretrained data, and were particularly bad at answering questions about reptile conservation. However, given some extra training with the CE database, their performance improved dramatically. When we put these models head to head with human experts (from the conservation evidence team), with a set of questions and with RAG access to the database, we found that the models were just as good as our experts, but answered the questions much much much faster (near instant).
Essentially, LLMs without extra training are likely to perform poorly and misinform decision-makers. This is crucial when considering how to build AI infrastructure for public policymaking.
[…377 words]Radhika Iyer, Alec Christie, Anil Madhavapeddy, Sam Reynolds, Bill Sutherland, and Sadiq Jaffer.
Journal paper in PLOS ONE (vol 20 issue 5).
Learnings from the Cambridge Environmental Sustainability Committee / May 2025
I joined Cambridge's loftily named Environment Sustainability Strategy Committee this academic year, and have attended a couple of meetings with the latest one being held today. While a lot of what goes on is intricately tied into the University's rather special governance structure and the complexity of the College system, there has been significant progress on making all of this more visible more widely.
Sally Pidgeon, our wonderful head of Enviromental Sustainaibility, has been redeveloping the public website and has put a lot of interesting data online. There is now a new Environmental Sustainability website that tracks the University committment structure more closely, with the areas broken up into Carbon & Energy, Travel & Transport, Waste & Circular Economy, Biodiversity, and Water usage.
[…842 words]Humans are the ones that will save nature, helped by AI / May 2025
In my earlier note about how AI should unite conservation, I talked about the robust debate ongoing within Cambridge about whether or not we're too "AI obsessed" and are losing track of our goals in the rush to adopt learning algorithms. Jacqueline Garget has written a brilliant roundup about how colleages like Sam Reynolds, Chris Sandbrook and Sadiq Jaffer in the CCI are leading conversations to make sure we advance with eyes wide open.
[…537 words]Technology needs to unite conservation, not divide it / Apr 2025
I had a tremendous time participating in last year's horizon scan of AI and Conservation, which laid out the opportunities that technological progress from AI (a catchall phrase here) could bring to hard-working conservation practitioners. Since then, there's been a lot of corridor conversations about future projects (and even dinner with the Wildlife Trusts). However, there has also been discussion about the potential harms of our work, most notably in a response letter to our paper written by Katie Murray and colleagues.
Murray et al make two really important points:
[…1481 words]
- [...] importance of ecological expertise must be recognised as much more than just the expert annotation of training data
- [...] effort should be made to build capacity for AI development in the Global South, so that the rewards of successful research can be shared -- The potential for AI to divide conservation
2nd Programming for the Planet workshop CFP out / Apr 2025
Dominic Orchard and I had a blast running the first PROPL workshop a couple of years ago, with a full room and engaged audience in POPL in London. Last year, our sister conference LOCO took over, and it's our turn again this year! PROPL will return for a second outing in October, co-located with ICFP/SPLASH in Singapore in October. Read the call for papers here (deadline 3rd July 2025).
[…565 words]Using graph theory to define data-driven ecoregion and bioregion maps / Apr 2025
This is an idea proposed as a good starter project, and is available for being worked on. It may be co-supervised with Daniele Baisero and Michael Dales.
Maps of biologically driven regionalization (e.g. ecoregions and bioregions) are useful in conservation science and policy as they help identify areas with similar ecological characteristics, allowing for more targeted, efficient, and ecosystem-specific management strategies. These regions provide a framework for prioritizing conservation efforts, monitoring biodiversity, and aligning policies across political boundaries based on ecological realities rather than arbitrary lines. However these products have historically been "hand drawn" by experts and are mostly based on plant distribution data only.
[…270 words]Battery-free wildlife monitoring with Riotee / Apr 2025
This is an idea proposed as a good starter project, and is currently being worked on by Dominico Parish. It is co-supervised with Josh Millar.
Monitoring wildlife in the field today relies heavily on battery-powered devices, like GPS collars or acoustic recorders. However, such devices are often deployed in remote environments, where battery replacement and data retrieval can be labour-intensive and time-consuming. Moving away from battery-powered field devices could radically reduce the environmental footprint and labour cost of wildlife monitoring. The rise of batteryless energy-harvesting platforms could enable ultra-low-power, long-term, maintenance-free deployments. However, existing battery-less devices are severely constrained, often unable to perform meaningful on-device computation such as ML inference or high-frequency audio capture.
This project explores the development of next-generation, battery-less wildlife monitoring platforms using Riotee, an open-source platform purpose-built for intermittent computing. Riotee integrates energy harvesting with a powerful Cortex-M4 MCU and full SDK for managing state-saving, redundancy, and graceful resume from power failures.
[…273 words]An access library for the world crop, food production and consumption datasets / Apr 2025
This is an idea proposed as a good starter project, and is available for being worked on. It may be co-supervised with Alison Eyres and Thomas Ball.
Agricultural habitat degradation is a leading threat to global biodiversity. To make informed decisions, it's crucial to understand the biodiversity impacts of various foods, their origins, and potential mitigation strategies. Insights can drive actions from national policies to individual dietary choices. Key factors include knowing where crops are grown, their yields, and food sourcing by country.
The FAOSTAT trade data offers comprehensive import and export records since 1986, but its raw form is complex, including double counting, hindering the link between production and consumption.
[…372 words]3D printing the planet (or bits of it) / Apr 2025
This is an idea proposed as a good starter project, and is currently being worked on by Finley Stirk. It is co-supervised with Michael Dales.
Thanks to a combination of satellite information, remote sensors and data-science, we now are able to reason about places all over the globe from the comfort of our desks and offices. But sometimes, you just want to be able to see or touch an area to understand it properly: the flat 2D-projection on a screen doesnt necessarily reveal the subtle geography of a landscape, and data locked into a computer feels less immediate than even a physical model of the same area.
In recent work, Michael Dales has experimented with making 3D-printed models of surface terrain to make some areas of study more relatable. By combining high resolution Digital Elevation Maps (DEMs), and CAD software we were able to scale and print this section of a Swedish forest used to observe Moose migrations.
Displaying the 15 most recent news items out of 59 in total (see all the items)