Following Yaa's suggestion, I'm posting a specific use case for API access to vessel track data.

Use case: machine-learning vessel arrival / port-ETA prediction (research at East China Normal University). The hourly AIS presence series (position + speed since v5) is already our strongest feature source, but arrival prediction benefits strongly from route-level signals — course-over-ground sequences, turning points on final port approaches, and deceleration zones along shipping lanes. These need actual track points; hourly aggregation discards exactly that structure.

Scale: non-commercial academic research, 2018–2020, a few thousand cargo vessels in the North Sea & Baltic region. Results will be published with GFW duly acknowledged.

Why the 10-tracks/day map export doesn't scale: excellent for manual review, but it cannot feed model training. Even a rate-limited endpoint (per-MMSI, bounded time windows) would make a real difference.

— Mohan Cui (East China Normal University) & Sandro