Research

We publish the method and the data behind it. The code for every number is public.

Ship manoeuvring and resistance models under sparse AIS sampling: what can be predicted, identified and bounded

Vishmayraj Zala, Kathan Patel. Under review, 2026.

Code and data bundle, Zenodo

Shore-side monitoring sees a ship only through AIS reports, seconds to minutes apart. We measured what those reports let ship physics predict, identify and bound: 47,739,929 reports from four networks over eleven days in the Baltic and North Sea, tested on held-out ships.

  • On course changes of 45° or more, textbook turn models predict where the ship went about 22% worse at the 95th percentile than averaging the two reports.
  • A hull’s steering constants can’t be identified from AIS. It carries no rudder angle.
  • A Holtrop–Mennen speed screen is crossed by 4 of 8,472 ships up to Froude number 0.45, and by 181 of 199 above it.
  • Under sparse sampling, ship physics is good for bounds, not for filling in the gaps.