A research and development practice for conservation data.
Operational Ecology sits between two kinds of work that rarely meet: methods research on how machines perceive animals, and the engineering discipline that puts those methods into a biologist's hands and keeps them working there.
Archive scale is the whole problem.
A method that works on a curated benchmark is a different thing from a method that holds up across four decades of recordings, dozens of observers, changing equipment, and catalogues built by hand long before anyone thought about machine learning. Most of the difficulty in this field lives in that gap.
The research side is aimed squarely at it: low-resource learning where labels are scarce, representations that survive noise, and evaluation that reports what a model does not know. The engineering side makes the results usable, with numbers that trace back to the code and data that produced them.
Three lines of work, one substrate.
Computer vision
Reading what is in the photographs: individual identity across years and observers, scarring and injury events, body condition, and the social and temporal context an expert would bring to the same image.
Bioacoustics
Finding animal voices in ocean noise, denoising them, and learning representations that hold dialect and variation rather than flattening it into fixed classes.
Data systems
Turning independent, multi-decade archives into one queryable system: the step that makes cross-population analysis possible at all.
Alexander Barnhill
I have worked on machine learning for killer whales since 2017, at the Pattern Recognition Lab at Friedrich-Alexander-Universität Erlangen-Nürnberg, where my doctoral research covers low-resource deep learning for repository-scale photo-identification. The recurring problem is not model capacity. It is noisy recordings, incomplete archives, and datasets assembled over decades by many different hands.
The methods work stays close to the field. I have collaborated for five years with researchers in British Columbia, visiting the study area annually for extended periods, which is where finwave came from and where the published results were tested. Methods developed alongside the people using them are the ones that hold up across a field season.
Before conservation work took over I built production software and models in industry: rail infrastructure monitoring, clinical decision support, disaster-response damage assessment. That is why the systems here are designed to be operated by your team, with the resources a research programme actually has, for longer than a single grant cycle.
- FAU Erlangen-Nürnberg
- Researcher, Pattern Recognition Lab
- Project Fathom
- Principal investigator, a Raincoast Research Society initiative
- Alert Bay Cetological Society
- Board member
-
finwave
Open platform for cetacean photo-ID data curation. 1M+ images, in use by international research groups.
-
ANIMAL-CLEAN
Species-independent deep denoising for bioacoustic signal enhancement.
-
ORCA-WHISPER
Automatic generation and analysis of killer whale sound types.
Publications
- 2026 Marine Mammal Science Machine learning detection of scarring events in killer whales
- 2026 Ecology and Evolution Leveraging co-occurrence to improve deep learning photo-identification in social animals
- 2025 Ecological Informatics Advances in deep learning-driven photo identification and meta analysis of cetaceans in large data repositories
- 2024 Biological Reviews Automatic detection for bioacoustic research: a practical guide from and for biologists and computer scientists