# A trio of papers I read on biodiversity and forests this week *2025-02-20 — note* This week I've been reading three really nice pieces of work by my colleagues, in the form of a [review paper](https://www.nature.com/articles/s44358-025-00022-3) on biodiversity and AI, a [benchmark](https://besjournals.onlinelibrary.wiley.com/doi/10.1111/2041-210X.14503) for 3D forest reconstruction using laser scanners and a [mobile app](https://github.com/MingyueX/GreenLens) for measuring the width of tree trunks. A real bonanza for forest lovers\! ## Review paper on mapping opportunities for AI in biodiversity A paper on '[Harnessing AI to fill global shortfalls in biodiversity knowledge](https://www.nature.com/articles/s44358-025-00022-3)' just came out in Nature Biodiversity today (via [Oisin Mac Aodha](http://oisin.info)). They start with the baseline present uses of AI (camera traps, acoustic monitoring and improved data analysis) which are pretty well known to anyone in the field, but then introduce [a lovely diagram of future uses](https://www.nature.com/articles/s44358-025-00022-3/figures/1) of AI for biodiversity which includes: - Rapid retrieval of existing information means both looking into existing literature, but also the digitisation of existing museum specimens. Coincidentally, I have just posted a new student project on the area of [insect digitisation at the Zoology museum](https://anil.recoil.org/ideas/digitisation-of-insects) from [Tiffany Ki]() on the latter topic, which I'd be very happy to hear from interested students about. I have also have been working on [LLM driven evidence retrieval](https://anil.recoil.org/papers/2024-ce-llm) recently, so I'm all in favour of lots more projects in this space. - Once the data is retrieved, they discuss how this could be used for richer hypothesis generation via detection of new patterns for humans to review, ranking high-value areas that need more observations, and generally doing more unsupervised learning over the vast space. This is a good zooming in from many of the general areas covered in the [Royal Society Science in the Age of AI](https://royalsociety.org/news-resources/projects/science-in-the-age-of-ai/) report as well, and very good to see given the sheer urgency of more action in the field of biodiversity conservation. - Finally, there's also the fascinating topic of ecological modelling where we move from individual species to whole communities, as well as knowledge-guided machine learning towards this. I'm planning on experimenting more with differentiable models (beyond ABMs, where both [differentiable](https://anil.recoil.org/ideas/differentiable-abm) and [reversible](https://anil.recoil.org/ideas/rev-abm) have worked very well). The recent paper on [NeuralGCM](https://www.nature.com/articles/s41586-024-07744-y) from the Google team underlined the huge potential of combining purely data-driven and purely-computational models into a combined system with much better predictive power than either by itself. Those interested in this may also want to look at our recent [horizon scan on AI and conservation](https://anil.recoil.org/papers/2024-ai-conhorizon) from a few months ago. The field is moving so quickly that I wouldn't be surprised if both of these were obsolete a year from now\!

If you like biodiversity, consider working with me on this project!

## Benchmark dataset for tree species identifications And then out in MEE is a comprehensive benchmark from a collection of forestry researchers on a [benchmark for tree species classifiction from proximal laser scanners](https://besjournals.onlinelibrary.wiley.com/doi/10.1111/2041-210X.14503). Their [FOR-species20k](https://zenodo.org/records/13255198) dataset is on Zenodo, and has tons of tree point clouds taken using a variety of laser scanning techniques ([TLS](https://www.earthscope.org/what-is/tls/), [MLS](https://www.sciencedirect.com/science/article/pii/S1618866723003710) and [ULS](https://www.gispro.pl/en/products/unmanned-laser-scanning-uls/)). As [Emily Lines](https://www.geog.cam.ac.uk/people/lines/) notes: > Most importantly, we demonstrate that community efforts and open science are the only way to make significant progress in this important task. With more researchers publishing and sharing high quality 3D forest datasets, I hope we see an end of single-site studies and that proper and broad benchmarking of all new 3D forest deep learning methods becomes the standard. > \-- [Emily Lines on LinkedIn](https://www.linkedin.com/posts/emily-lines-2b271a80_openscience-ai-deeplearning-activity-7292116486519676928-XfwF) I've been learning more about [tree species identification for tropical species](https://anil.recoil.org/papers/2024-hyper-tropical-mapping) last year, so I'm looking forward to delving more into laser scanning techniques soon from this work. ## A mobile app for measuring tree trunks And last but not least, I was delighted to see that my colleagues [Srinivasan Keshav](https://svr-sk818-web.cl.cam.ac.uk/keshav/wiki/index.php/Main_Page) and [Frank Feng](https://www.cst.cam.ac.uk/people/zf281) released the extremely cool mobile phone app they've been working on for some time along with a [paper in Ecological Informatics](https://www.sciencedirect.com/science/article/pii/S1574954124003169?via%3Dihub#s0125). Their app is a simple and elegant mobile phone app that can measure the diameter of a tree trunk (more specifically, the [DBH](https://en.wikipedia.org/wiki/Diameter_at_breast_height)) just using standard cameraphone hardware on most modern-ish Android phones. I was lucky enough to beta test this and try it out on my [recent trip to India](https://anil.recoil.org/notes/compass2024-ric-tripreport), and the [GreenLens](https://github.com/MingyueX/GreenLens) is also now open source as well. > Other apps for measuring forest plots are available \[...\] but those for Android phones tend not to perform as well as ours, while those designed for the iPhone require the purchase of a high-end phone that is not affordable for researchers in the Global South. > > We believe ours is the only app to sit in the 'sweet spot' of offering high quality for low cost. > \-- [Frank and Keshav on cam.ac.uk](https://www.cst.cam.ac.uk/using-ai-see-wood-trees)

I actually got quite distracted while trying to beta test GreenLens in India as I ran across these adorable stray street puppies, which seems important to post

Synopsis: Exploring biodiversity and forests through 3 papers on AI, 3D reconstruction, and a tree-measuring mobile app. Words: 793 DOI: 10.59350/t7ekw-e7y39 ## Related - [Phenological regularity, not functional traits, determines whether tropical tree species can be mapped from imaging spectroscopy](https://anil.recoil.org/papers/2024-hyper-tropical-mapping) (paper, 2026-05-01) - [Careful design of Large Language Model pipelines enables expert-level retrieval of evidence-based information from syntheses and databases](https://anil.recoil.org/papers/2024-ce-llm) (paper, 2025-05-01) - [Affordable digitisation of insect collections using photogrammetry](https://anil.recoil.org/ideas/digitisation-of-insects) (idea, 2025-02-01) - [The potential for AI to revolutionize conservation: a horizon scan](https://anil.recoil.org/papers/2024-ai-conhorizon) (paper, 2024-12-01) - [COMPASS 2024 report on the CoRE stack RIC meeting](https://anil.recoil.org/notes/compass2024-ric-tripreport) (note, 2024-07-08) - [Reverse emulating agent-based models for policy simulation](https://anil.recoil.org/ideas/rev-abm) (idea, 2023-01-01) - [Scalable agent-based models for optimized policy design](https://anil.recoil.org/ideas/differentiable-abm) (idea, 2022-01-01) --- Canonical: https://anil.recoil.org/notes/forest-apps-and-benchmarks Type: note License: CC BY 4.0 Tags: forests, biodiversity, conservation, sensing, ai, llms