AI identifies marine species from your dive photos in seconds. Practical guide to iNaturalist, FishVerify and citizen science contribution for underwater photographers.
AI applications such as iNaturalist, FishVerify and Seek can automatically identify marine species from your dive photos, often with accuracy comparable to a trained marine biologist. By submitting your images to these platforms, you feed global scientific databases used by universities and conservation programmes, with no biology training required.
In July 2022, I was diving at Banyuls-sur-Mer with a group of six students. Late afternoon, on a rocky overhang at 14 metres, I photographed a nudibranch I did not quite recognise. The morphology looked familiar but did not quite match anything in my mental catalogue. I came back with about fifty images, most of them poor quality in the flat late afternoon light.
That evening, I submitted the sharpest one to iNaturalist. Within 48 hours, two community members had confirmed the species: Hypselodoris villafranca, present along this stretch of Catalan coast but poorly documented in local databases. The observation reached Research Grade status and was automatically exported to GBIF. Six months later, a researcher from the University of Barcelona sent me a message: she was using that data to map the species' distribution in the Gulf of Lion.
It was not an outstanding photo. Sharp on the subject, correctly exposed, nothing more.
Your photos hold value you may not have measured yet. Their scientific rarity does not come from aesthetic quality. It comes from the geotagged information they carry: a species, a date, a precise location.
iNaturalist is the platform most widely used by the scientific community. Its AI identification engine, trained on millions of observations, recognises most major marine taxonomic families. The automatic suggestion is a proposal, not a verdict: definitive identification comes from volunteer community members.
For an observation to be scientifically exploitable, it needs to reach Research Grade: two concordant identifications from different experienced users. The process is fast for common species, slower for cryptic or regionally uncommon ones.
How to submit an underwater observation on iNaturalist:
Location is the critical criterion. A technically excellent photo without coordinates cannot join GBIF.
FishVerify is specifically trained on fish species. Its accuracy on Mediterranean, tropical Atlantic and Indo-Pacific species often outperforms iNaturalist for this group. The application is paid (annual subscription or one-time purchase) but delivers identifications in seconds, useful for guides or instructors during dive debriefs.
Seek, developed by the same team as iNaturalist, identifies species by pointing the camera at a subject, without automatic recording. Unusable underwater in its current form, it is still useful at the surface for checking a photo taken during a dive before submitting it formally on iNaturalist.
AI identification works from morphology, relative colour patterns and characteristic shapes. It does not need perfect colour correction. A photo taken in natural light at 12 metres in the Mediterranean, dominated by blue-green, will be analysed on its structure, not on absolute colour values.
Factors that improve accuracy:
Factors that reduce accuracy:
For invertebrates and small cryptic species, underwater macro photography techniques produce images with sufficient detail for AI to work effectively.
A regular diving photographer can build a meaningful observation portfolio within a single season. A few principles to maximise scientific value:
Precise geolocation. Your smartphone GPS taken at the surface before the dive is sufficient. Accuracy of 50 to 100 metres is broadly acceptable for biodiversity databases. If you dive from a boat, log your position before entering the water.
Document multiple individuals. A single individual sighting can be coincidental. Three or four individuals at the same site across a season constitute reliable presence data.
Photograph the surrounding habitat. An image of the substrate (sandy bottom, rock, posidonia seagrass bed, coral) enriches the observation with ecological context that distribution models rely on.
Report unusual behaviours. A fish at an atypical depth, a species observed outside its usual season, a rarely documented predator-prey interaction: these apparently anecdotal observations often carry disproportionate value for researchers calibrating behavioural models.
Marine protected areas are zones where your identification data is particularly sought by managers who have limited resources for regular field inventories.
A common pitfall: treating an AI suggestion as a certified identification. AI makes mistakes. Confusions between closely related species in complex genera such as Chromis, Gobius or nudibranchs in the genus Flabellina are frequent.
The basic rule: if the algorithm's confidence is below 70% or if you have any doubt, submit the observation without specifying a species and let the community decide. An observation labelled only "marine invertebrate" remains useful for geographic occurrence data.
The data you generate also contributes to monitoring coral bleaching events. In 2024 and 2025, citizen observations enabled researchers to map the extent of bleaching events far more rapidly than research teams alone could have managed.
Ethical underwater photography addresses the code of conduct that underpins this approach: never disturb species for a photograph, observe from a respectful distance, document without intervening.
Programmes such as REEF (Reef Environmental Education Foundation), CoralWatch and Reef Check accept data from trained divers. They require a short online certification to standardise counting protocols. AI does not replace this framework: it accelerates it by simplifying identification during preparation or post-dive analysis.
Contributing to citizen science and improving your underwater photography skills are not in tension. Wanting clearer images for better identification is a natural motivation for becoming more technically proficient underwater.
The minimal workflow for turning your next dive into a scientific contribution:
Ten minutes per dive outing is enough. Over a full year of regular diving, you can build several hundred validated observations that join global databases.
If you want to improve the technical quality of your photos so they are more easily identifiable, the AquaExposure underwater photography course covers framing and exposure techniques that produce scientifically usable images, including for small invertebrates using macro techniques.
AI colour correction for underwater photography can also recover technically imperfect photos that might otherwise have limited scientific value: a noisy or slightly soft image can become identifiable after processing.
iNaturalist is the gold standard for scientific contribution: the community validates your identifications and your data feeds into global databases. FishVerify is more precise for fish species in tropical and Mediterranean environments. Seek identifies species in real time but without history or validation. Combine iNaturalist for scientific contribution and FishVerify for quick identification after a dive.
Yes, as long as the image is sharp on the subject. AI works from morphology, contours and relative colour patterns rather than absolute colour values. A blue or green colour cast does not prevent identification if the animal's structure is readable. Below 15 metres, colours shift heavily toward blue but body shape and patterning remain identifiable.
Yes. Once an observation reaches "Research Grade" status through two independent verifications by experienced users, it is automatically exported to GBIF, the global biodiversity database used by universities and conservation programmes worldwide. Studies published in 2024 and 2025 explicitly cite data submitted by recreational divers via iNaturalist.
No. You can submit an observation with only the AI suggestion and let the community confirm or correct it. The contribution you make is the geotagged photograph. The sharper the image on diagnostic features, the faster the validation process.
Partially. Well-documented species such as common Mediterranean nudibranchs and cephalopods are reliably identified. Rare or regionally uncommon species often require human validation by specialised iNaturalist communities or dedicated forums such as NudiPixel.
Location is the critical criterion. The GPS coordinates from your smartphone taken at the surface before a dive, or from the dive site shoreline, are sufficient. Accuracy to 50 to 100 metres is broadly acceptable for biodiversity databases. A beautiful photo without location data cannot be exported to GBIF.
Species poorly documented at a specific geographic location, unusual behaviours, species observed outside their typical depth or season, and repeat observations at the same site over several years. Coral bleaching records and invasive species sightings have immediate value for active monitoring programmes.
