Can a wind forecast help a sailing drone tack? Not by the 12 points we hoped

On two months of NOAA wind, a controller given a perfect forecast of slow wind shifts recovers none of the 12-point gap to a clairvoyant tacker.

In the previous marine-hull step, a simple rule that tacks when the other tack makes more progress (header-following) reached about 88% of the ideal upwind progress on 24-hour NOAA CO-OPS wind records, against 99.7% for a clairvoyant that knows the whole day. We asked how much of that 12-point gap a forecast could close.

We gave a receding-horizon controller a persistence forecast, a 30-minute-smoothed copy of the true future, and the true future itself (30 minutes up to the whole day), planned on 1 to 30 minute blocks. July holdout, 360 windows:

ControllerShare of ideal progress
Corridor-edge tacking0.837
Header-following0.876
Forecast controller, persistence0.856
Forecast controller, smoothed perfect forecast0.855
Forecast controller, raw perfect forecast (5-min blocks)0.868
Clairvoyant0.997

No forecast variant beat the simple rule, and the horizon made no difference. The reading: the clairvoyant's edge comes from single-sample gusts that no forecast service supplies. Caveat we state in the log: finer planning blocks did worse with a perfect forecast, so our planner is not optimal, and these are observed series rather than archived forecasts. This is a negative result for this controller family, not proof that no forecast helps. Next is a planner with a day-scale terminal value.

Code, data and numbers: research/marine-hull/ (LOG.md, milestone 5b).

All writing · More in Research findings · All tools and research