Kvarken Log test scan. June 2026.
I've been building a field tool for the Godzilla expedition. A GoPro placed on the bow will take a photograph every 60 seconds through the day. Each evening I pull the frames onto my phone, run them through the app, and it generates a reading for each one.
The readings aren't about the birds. They're about the place.
Each species reads the same water differently. Divers don't sit on water randomly. Where a bird settles, how long it stays, whether it dives or moves on, all of that is a response to the place: the water, the weather, what else is out there. That's what the app is trying to read. Red-throated and black-throated divers hunt most efficiently in clear water. A diving bird is a clue, not a confirmation. Absence matters too. It doesn't mean nothing was there.
What I'm not doing is speaking for the bird. The camera takes one visual slice. The bird is working across sound, smell, water temperature, what other species are doing around it. The app reads what the image shows, and says so when it can't.
GPS runs alongside it, tracked on my phone throughout the day. Weather goes in: wind, pressure, cloud cover. There's a sounds field too, for what the camera missed. Each reading has a position, a time, and a weather state.
The app uses an AI vision model to scan each frame. It works across hundreds of frames at once and picks up things no single observer would catch. What it produces over a full crossing is part of what I want to find out. But it makes mistakes: it misidentifies birds at distance, it can't hear the site, it works from a single image. The readings are a starting point. It's also made me think differently about what looking actually is.
I've been testing it at Nydala this week, freshwater, different birds. The Kvarken is the real thing. We leave Holmsund on 28 June.
I'll post the readings here as we go.