Using several models together

A simple set of three

More models do not mean more certainty. Two or three that you understand are more useful than ten you cannot read. Add a model when it has a reason: different physics, a finer grid, different starting data. Which ones suit your area is in Which model should I use?

What to compare

• Timing. When does the front or the wind shift arrive? A six-hour difference may decide whether you are in port or at sea.
• Strength. Is the peak 25 kn or 35 kn? The gap between the models is more useful than either number.
• Direction and sea state. A wind shift of 40 degrees can turn a good beat into a bad lee shore. Wave height, period and direction are covered in Waves.
• Same time, latest run. Compare the same valid time from the latest run of each model. An old run is not a second opinion: When does a new run arrive?

Spread as a measure of confidence

The distance between models, and between the members of an ensemble, is the most honest sign of confidence you have. A narrow spread says that small changes in the start made little difference, so the forecast is stable. A wide spread says the outcome depends on details nobody knows yet. Spread normally grows with lead time: see How far ahead can you trust a forecast? and Ensembles and probabilities.

When to trust agreement

• Short range: yes. At 24 to 48 hours, a global model, a regional model and an ensemble that give the same picture are very likely to be close to what happens.
• Local effects: not fully. Models share observations and some assumptions, so they can be wrong together. This is most likely for things none of them resolve, such as squalls, fog and the wind close to a coast: What a model cannot see.
• Day seven: check the ensemble. If only single runs agree and the ensemble is wide, the agreement may be a coincidence.

When they disagree

1. Find out what they disagree about: timing, strength, track or direction. They often agree on the main feature and differ on the detail.
2. Check the ensemble. Does the disagreement fit inside its spread? Then it is normal uncertainty. If one model is outside the spread, doubt that one.
3. Look for the reason: resolution, run time, or treatment of the coast. See Why models disagree.
4. Check which has been closer in recent days against a nearby station or buoy: Checking a forecast against reality.
5. Plan for the worse case that is plausible. If your plan survives it, go. If it only works in the best case, wait or change the route.
6. Look again at the next run. The disagreement usually narrows as the time approaches.

Before you leave: a short checklist

1. Open the latest run of a global model, the best regional model for your waters and an ensemble.
2. Compare the time of the critical wind shift or front, and the peak wind and gust, in all of them.
3. Note the spread. Is it small enough for your plan?
4. Add a margin of about a force near coasts, headlands and in showery weather, and use gusts for the reef decision.
5. Check wave height, period and direction against your course and the entrance of your destination.
6. Check tides, streams and, if it matters, surge for the harbour entrances.
7. Read the official forecast or warning for the area and compare it with the models.
8. Decide your turn-back point and a safe port, and agree it with the crew.
9. Look again just before you cast off, and at each chance on the way.

Models are guidance. The skipper decides.