Kelp Quest

Kelp Quest turns Washington seafloor photographs into careful little identification puzzles. Check the feature under a red target, decide whether a model's label holds water, and help the Seattle Aquarium refine a dataset for kelp-forest monitoring.

What is Kelp Quest?

Each subject is a tightly cropped piece of a remotely operated vehicle photograph, with a red dot naming the exact bit of seafloor that matters. One workflow asks for a simple yes or no on the model's prediction; another asks you to choose among categories spanning kelp, other algae, animals, substrate, debris, and genuinely unreadable patches. The trick is resisting the loudest object in the frame and judging what sits directly beneath the dot. Murky water, uneven light, and look-alike algae turn an apparently simple click into a compact lesson in field ecology.

What you can do there

• Confirm or reject a model label on a seafloor image patch
• Choose the category beneath a target when the model is uncertain
• Consult the field guide for difficult classifications

Why we picked it

The project gives a familiar image-labeling task unusually strong texture and purpose. Its photographs feel like tiny underwater mysteries, while the field guide and research notes make clear how those judgments connect to a larger attempt to map and restore Washington's kelp forests.

How to get the most from it

Start with the Yes/No workflow, read the short tutorial, and keep the crosshair's center—not the biggest shape—in charge of your answer. Open the field guide when red and brown algae blur together, and use Not sure rather than forcing a heroic guess.

Good to know

The multiple-choice workflow is marked desktop-only and can temporarily run out of images because subjects arrive from the Yes/No stage. Some training images may reveal whether your answer was correct. The interactive flow, account behavior, audio, and accessibility were not independently tested here.

Who made it, and when?

Seattle Aquarium Coastal Climate Resilience team is the credited creator or organization. The earliest supported launch date we found is March 24, 2026.