Key facts
| Fact | Detail |
|---|---|
| HRRR grid spacing | 3 km, refreshed hourly |
| GFS grid spacing | 13 km, forecasts to 16 days |
| NAM grid spacing | 12 km parent with 3 km nests |
| UKV domain spacing | 1.5 km inner, 4 km outer |
| Met Office global grid | 10 km in mid-latitudes |
| 24-hour wind error, New York Bight | RMSE 2.1 m/s, correlation 0.89 |
| 48-hour wind error, New York Bight | RMSE 2.6 m/s, correlation 0.83 |
| GFS skill at day 10 | Almost none |
What a grid box averages away
The gap matters most on marginal days, when 12 knots and 18 knots separate a comfortable drift from a wet ride home. The working skill is choosing which product to trust at the coast, how far ahead to commit, and what to check against a buoy before the trailer leaves the yard. A numerical model stores one temperature, one wind vector and one humidity value for every box in its grid. The High-Resolution Rapid Refresh divides the air over the United States into 3 km boxes and refreshes them hourly. The Global Forecast System uses 13 km boxes and runs forecasts out to 16 days. The North American Mesoscale model works at 12 km with 3 km nests inside the parent grid. Every one of those rectangles is a parcel of atmosphere. None is a reading at the boat ramp. The Met Office builds the same hierarchy at a smaller scale. Its UK variable-resolution model uses 1.5 km boxes in the inner domain, 4 km in the outer domain, and a global model at 10 km through the mid-latitudes. The design purpose is coastal detail. A nested grid represents coastlines, hills and the processes that depend on them in more detail than the coarse background. A fine grid still has a floor. A cove 300 metres across, a pierhead or a narrow inlet falls inside one box and loses its shape. Two cues show which regime you are in. Move the forecast pin 200 metres and watch the value. If nothing changes between one beach and the next bay, both points share a box or a smoothed field. If the number jumps at a headland, the model is resolving a boundary and the value depends on position. Neither cue describes the wind at the rod tip. Apps add a second smoothing layer. A point forecast is usually interpolated from the four surrounding grid points, and some products blend several models before they draw a chart. Interpolation fills the gap between boxes; it does not create detail inside one. A tidy wind curve on a phone screen can hide a 5 knot difference between two headlands a mile apart. Treat the pin as a regional number and the marine zone forecast as the plan for your stretch of water. The grid is a planning layer, and the zone text carries the timing that a grid cell cannot.
Why the model has nested domains
A grid also decides which physics the model calculates and which it approximates. At 3 km the HRRR is convection-allowing, so shower and thunderstorm cells grow from the resolved flow instead of a statistical recipe. Coarser grids parameterise convection, and a line of cells becomes a broad area of mixing and rain. That difference reaches the surface as wind, because a downdraft spreads across the water as a gust front. Grid spacing is not the smallest feature a model reproduces faithfully. Numerical smoothing erases detail at a few grid lengths, so the useful scale sits well above the headline number. Read 3 km as a label for the model family. The label does not promise a wind reading every three kilometres. A chart drawn at that spacing still shows a coastline with a few gentle bends. Height adds another correction. The 10 km global figure is approximate at the surface over southern Britain, nearer 17 km at the equator, and coarser aloft. A coastal nest therefore sits inside a background that supplies the wind aloft and the weather system. When nest and background disagree at a headland, the nest carries the local shape and the global model carries the synoptic pattern. Boundary-layer physics is the other hidden variable. The model estimates how much the surface slows the wind, and how a stable layer decouples the deck from the flow aloft. Over cold water in spring that decoupling is strong, and an offshore gradient of 20 knots arrives at deck level several knots lighter. The same gradient over warm shallow water mixes downward and arrives nearly intact. That is why two products diverge on the same afternoon. A global model can print 15 knots across a whole coast while a regional model shows 20 knots at an exposed point and 10 knots in the lee. The disagreement is two grid sizes describing one shoreline. Pick the finer product for the launch and the coarser one for the week.
How lead time changes the error
Error grows with forecast length, and the growth is measurable. Offshore wind forecasts over the New York Bight show the pattern. At 24 hours the HRRR and GFS carried a root mean square error of 2.1 m/s and a correlation of 0.89. At 48 hours the error was 2.6 m/s and the correlation at least 0.83. By day 10 the GFS had almost no skill over the same water. Translate that into boat terms. A 2 m/s error is about 4 knots, enough to move a day from comfortable to marginal. Treat the number as an error bar. A single value needs company from the next run and from the buoy, especially when the margin between a safe drift and a wet ride is two knots wide. The forecast is a range, and the range decides the plan. Run-to-run change carries more information than any single chart. Models cycle at fixed hours, and consecutive runs either hold a solution or drift. If the morning run lifts wind by 5 knots and the afternoon run holds it, the trend is real enough to plan around. If the runs oscillate around one value, the atmosphere is unresolved and the decision can wait a day. Commit inside 24 to 48 hours. Check the latest run on the morning of the trip, when the short-range guidance has the best grip on the coast. A day 7 chart is a trend line for choosing a week. Do not book a non-refundable day against it. Book flexible dates and the forecast never has to be right a week out.
Reading the coast through the land mask
Every grid box is flagged as land, sea or a blend, and that flag sets surface roughness and heat exchange. A 13 km box can hold open water, a beach and a ridge of land in one value. A 3 km box separates them and still smooths a seawall, a pier or a narrow inlet. Wind accelerates around a headland and shadows in its lee, and the mask decides which boxes get which. Local circulations add timing the grid cannot fix. A sea breeze needs a weak pressure gradient and a strong temperature contrast between land and water, then builds through the afternoon and fades near sunset. A regional model can show the onshore push and still put the onset an hour early or late. An angler on the beach sees the shift before the next cycle does. Night brings the opposite problem. Over cold water the lower atmosphere stabilises, surface wind drops, and a model wind at 10 metres can sit well above what a deck feels. Drainage flow off coastal hills adds a light offshore breeze before sunrise. An early start on a calm forecast often meets more wind than the afternoon does, and the lee of a headland is the place to wait it out. The correction runs in both directions. Exposure tells you an open point will see more wind than a sheltered launch. Timing tells you the sea breeze will build after lunch and die at sunset. Carry both, and the forecast number becomes a starting point.
Using a buoy as a point check
A buoy is a point measurement with its own sensor height, averaging period and exposure. A forecast is a grid average at a standard level. The two are different quantities, and a direct comparison misleads in both directions. A buoy tucked behind a headland reads lighter than the surrounding water. A buoy on an exposed shoal reads stronger than the beach two miles away. Compare the trend, not the single reading. If the nearest buoy has held 3 knots above the forecast for the same wind direction all morning, local wind is stronger than the model. If it has held below, the model is overstating exposure at that point. Direction matters as much as speed: a buoy swinging from southwest to west tells you a front has passed, whatever the speed says. Check the averaging period before drawing conclusions. A buoy number is an average over several minutes, while a gust is instantaneous. A 14 knot average with 22 knot gusts is a different day from a steady 14 knots, and the difference decides whether a drift holds or runs. The same gap sets how much correction a trolling pass needs. Read the gust spread before rigging. Distance matters more than a station list suggests. A buoy across the bay samples different water and a different fetch. A buoy outside the inlet samples the swell before it feels the tide. Choose the buoy whose water you will actually fish, even when it sits further down the list. When buoy and forecast disagree, believe the observation for the next hour and the forecast for the next day. That split settles the common case at the ramp: the buoy tells you what the water is doing now, and the model tells you where it is heading.
What to check before committing
Work through a fixed sequence and the forecast becomes a decision rather than a mood. Read the marine zone text from the local forecast office, which carries timing and hazard language that a grid cannot. Open the highest-resolution model available for the coast. Use the global model for the synoptic pattern and the nearshore wave guidance. Then check the buoy trend. Two corrections cover most of the residual error. Exposure says an open point sees more wind than a sheltered launch, so pick the launch that matches the wind you have. Timing says a sea breeze builds into the afternoon and dies near sunset, so an early start and an early finish beat a midday gamble on an exposed shore. Apply both before the boat leaves the trailer. Wave guidance deserves its own look. Nearshore wave models are driven by the same wind grids, so a wind error becomes a wave error, and a small overforecast compounds on a shallow bank where the chop stands up. Read height and period together, then check that the launch sits inside the area the guidance covers. A wave model built for the open coast says little about a two mile bay with a narrow entrance. On a marginal day, a sheltered bay with a short run to the mark is a smaller plan than an exposed headland with a long return. That choice belongs at the ramp, not on the water, because turning back costs fuel and daylight that a shore session would have used. One habit improves the next forecast. Save the morning run for your headland, note the buoy reading at launch, and compare the two when you get home. After a season the local error has a direction and a size, and the next decision takes less time.
Related reading
Planning a weather window Gusts versus sustained wind Wind speed thresholds for small boats
Sources
- NCEP Numerical Forecast/Analysis Systems — NOAA Environmental Modeling Center · T1
- Technical Implementation Notice 14-32: High-Resolution Rapid Refresh — NOAA National Weather Service · T1
- Numerical weather prediction models — Met Office · T1
- Weather DataHub: Frequently Asked Questions — Met Office · T1
- Evaluation of Hub-Height Wind Forecasts Over the New York Bight — Wind Energy, Wiley · T1