Telltale

weather · sailing knowledge

Weather models explained for sailors

A forecast is a model’s best guess for a grid cell and an hour. Read which model, which run and which hour, then check it against the stations.

By Telltale Team · Reviewed

Before you go

  1. Open Conditions at the water you will sail and read each model’s wind for the hours you’ll be out. Conditions
  2. Note the spread. Models within 2–3 kn of each other agree; a wider gap means plan for both.
  3. Open the nearest station and compare its last few hours with each model. A station
  4. Lean on the model that matched this morning, and keep a plan for the one that didn’t.

How a forecast is made

Models start from observations and step the atmosphere forward in time.

Physics-based models solve the equations of the atmosphere on a grid. AI models such as ECMWF’s AIFS learn how weather evolves from decades of past analyses. Both start from the same kind of observations and both output a value per grid cell per hour, not the wind at your boat.

References 1

Global, regional, AI and ensemble

Pick the view that answers your question.

A global model sees the big pattern days ahead. A regional model uses a finer grid for terrain but takes its edges from a global one. An ensemble runs the same model many times to show the range.

ViewGood forKeep in mind
Global (GDPS, GFS, IFS)The pattern 2–10 days outCoastal detail is coarse
Regional (HRDPS, HRRR)Terrain, the next 1–2 daysA finer grid is not proof of accuracy
AI (AIFS)Another estimate of the patternCheck which product you are reading
EnsembleHow sure the forecast isThe spread matters more than the mean

References 2 1

When models disagree

Disagreement is information: it tells you where the forecast is weak.

Compare the same spot, the same hour and the same quantity. If the runs agree on a front but not on its timing, plan for the timing to slip. Use the newest run of each model, and note its run time.

Telltale Conditions off English Bay: four model values side by side with the best one marked, the spread between them, live readings from nearby stations, and First Narrows with its next slack.
In TelltaleEvery model and the stations near one spot. Tap the water and Conditions shows each model’s wind for that spot, how far they disagree, the fresh station readings within 20 km and how the current is running. Open Conditions

What a grid can’t see

A 2.5 km grid gets one or two values across a narrow inlet.

HRDPS runs at 2.5 km. Howe Sound is around 3 km wide in places, so the model has one or two cells to describe wind that may hug one shore. A smooth map layer is interpolated between cells; it adds no new information.

Telltale Conditions over Howe Sound: modelled wind streaks down the sound and live station readings from Squamish to Bowen Island.
Howe Sound in Telltale Conditions. The sound is a few kilometres wide, so a 2.5 km model has only a cell or two across it.

References 2

Check the forecast against the stations

The morning’s observations tell you which forecast is on track.

A station measures one place, at a known height and exposure, averaged over a known time. Line up its last hours with each model for the same hours. If they disagree by more than a few knots, trust the station for now and the trend for later.

Point Atkinson, 6 Oct 2026: forecast against measuredHourly mean wind at Point Atkinson on 6 October 2026, 08:00 to 16:00, against the HRDPS 12Z run. At 11:00 the station measured 11 kn gusting 15; the run had 1 kn.05101508:0010:0012:0014:0016:00fills inMeasuredHRDPS 12Zkn
Point Atkinson, 6 Oct 2026: forecast against measured. ECCC station Point Atkinson (hourly mean) against the HRDPS 12Z run (05:00 PDT). Real values from the committed snapshot.
Telltale Conditions with the Point Atkinson station open: an Observed row of hourly wind above rows for HRDPS, HRDPS West, HRRR and ECMWF, a spread row, and a note on which model has done best there lately.
In TelltaleMeasured against forecast, per station. Tap a station in Conditions to line up what it measured beside each model for the same hours, with the spread between the models and the one that has done best there lately. Open Point Atkinson

References 13

Model names at a glance

The name tells you who runs it and how fine the grid is.

Availability depends on the product and the region.

  • ECCC · GEM: HRDPS, RDPS, GDPS
  • NOAA: GFS, HRRR, NAM
  • ECMWF: IFS (physics) · AIFS (AI)
  • DWD: ICON global and regional
  • Met Office: global and UK models
  • Météo-France: ARPEGE · AROME

References 2 3 4 5 6 7 8 1 9 10 11

Wind, waves and current are separate forecasts

A wind arrow says nothing about the sea state or the current.

Check wave height, period and direction on their own, and the current from the CHS tables. Wind against current builds a steeper sea than the wind alone suggests.

Tides and currents

References 12

Runs, grids and delivery

Provider, family and variant are separate labels: ECCC, GEM and HRDPS continental are not interchangeable, and HRDPS West is another variant. ECMWF’s 0.25° open-data grid is a delivered product, not every calculation inside IFS. Keep the run time, the file’s arrival and the valid time apart; a run every six hours doesn’t mean the data is out the minute after.

After the sail, judge a forecast by the run that was out before you left. A later run can explain the day, but it couldn’t have helped you decide, and one striking miss doesn’t make a model bad.

References 2 15

Sources and review

Official references checked on 2026-10-11. Product details and schedules change; the provider’s page is the current word.

Take it to the water

Check the stations before you leave, then record the sail for the debrief.

Open ConditionsRecord a sail