Windy is not a forecast. It is a renderer — it paints one numerical weather model's raw output onto a map and lets you scrub through time. Almost every mistake people make with it comes from reading the colour and skipping the three questions underneath it.
The wave legend from your link. Note the breaks: 0.5, 1, 1.5, 2 — then a jump to 6, then 9. The scale is not linear. Half the colour range covers the first two metres, because that is where nearly all of the world's sea state lives. A small colour shift down at the calm end is a bigger real change than a large one up at the red end.
The frame
Every pixel on Windy answers a specific question you may not have asked. Get in the habit of confirming all three before you interpret anything. Two of them sit in corners of the screen most people never click.
RIGHT EDGE
Wind, gusts, waves, swell, rain, thunder, temperature, cloud, pressure, currents, visibility, radar, satellite. These are different physical quantities from different sub-models. Waves is combined sea state; swell is only the long-travelled component. Confusing the two is the most common misread on the site.
BOTTOM RIGHT
The label showing ECMWF, GFS, ICON or a local high-resolution model. It also tells you the grid spacing and when the run was issued. Switching models can change the picture completely — and that change is information, not noise.
BOTTOM STRIP
The timeline. Windy opens on the nearest step, not necessarily now. Near-term steps are 3-hourly and widen further out. Drag it — a sea state that looks fine at 06:00 can be a different animal by 15:00, and a still frame hides that.
Practical consequence: a screenshot of Windy is nearly worthless without those three facts attached. When someone sends you one, ask which model and which hour.
Waves — part one
Wave layers report significant wave height, written Hs. It is not the average wave and it is not the biggest wave. It is the average height of the highest third of the waves — a definition that exists because it matches what an experienced observer reports by eye.
Half the waves are smaller than Hs. But across any few hours, the largest single wave you meet runs to roughly 1.9 × Hs, and a rogue can exceed twice it. So when Windy paints 2 m, plan for the occasional 3.5–4 m face. This one conversion accounts for a large share of "the forecast was wrong" stories.
Predict first
Windy shows Hs = 1.5 m for your afternoon crossing. Roughly what is the largest wave you should expect to meet?
B. Hs is the mean of the top third, so the single largest wave in a three-hour record sits near 1.9 × that — about 2.8 m here. Doubling the printed number is the safe field rule. C inverts the definition: Hs is deliberately weighted towards the big end, which is exactly why it matches eyeball estimates.
Waves — part two
Wave period — seconds between crests — is buried in the click-popup and the meteogram rather than the map colour, which is why most people never look at it. It matters more than height. Period sets the wavelength, and wavelength compared to your boat is the whole story.
In deep water, wavelength L ≈ 1.56 × T² metres. Five seconds gives you crests 39 m apart; fourteen seconds puts them 306 m apart. Same height, entirely different day.
the sea, side on
Watch the magenta bar. It is a 10 m boat, drawn to the same horizontal scale as the water. Hold the height still and drag the period: the wave height barely changes, but the boat goes from spanning a good fraction of a wavelength — climbing and pitching through every crest — to a speck on a hillside that the whole sea lifts and sets down. That is why 1.5 m at 14 s is a pleasant ocean and 1.5 m at 4 s is a miserable one.
The view widens as the period grows, so that roughly three waves always fit on screen — read the scale bar, not the pixels. Height is drawn on its own exaggerated scale, which is why a long swell does not simply shrink away.
Direction
This is a genuine trap, and it is a collision of two conventions rather than a bug. Meteorology names a wind for its origin — a "northeasterly" blows from the northeast. But an animation can only sensibly move in the direction the energy is travelling. So the particles stream southwest while the forecast text says northeast, and both are correct.
Cross-check that never fails: wind waves are built by the wind, so they travel broadly with it. If the wave animation flows the same way the wind particles flow, you are reading both correctly.
Wind
Windy gives sustained wind and gusts as separate layers because the ratio between them is a diagnostic. Divide gust by sustained:
The absolute numbers still matter. Sustained wind in knots, for a small open boat: under 12 is comfortable, 12–18 is working weather, 18–25 is a decision, and above 25 the sea builds faster than most people expect — wave height climbs roughly with the square of wind speed, so 25 knots is not "a bit more" than 18.
Predict first
Windy shows 10 kt sustained and 27 kt gusts at your location, on a morning that otherwise looks calm. What is the model most likely telling you?
C. The mean is an average over the grid box and the hour, so a squall crossing it barely moves that number while the gust field spikes. Light-wind mornings with huge gust factors are the classic signature — and in the tropics that is very often a pre-dawn squall. Open the thunder layer, and the radar for the last few hours.
The detail view
The coloured map is a browsing tool — good for finding where weather is. Once you have a spot, click it. The panel that slides up carries far more than the map ever shows, and it is stacked in rows that reward reading top to bottom.
Confidence
Windy draws a single crisp number for every hour of the next two weeks, which is a presentational lie of omission — day 12 is rendered with exactly the same confidence as tonight. The fix is free: open several models and look at how far apart they are. Tight agreement means a well-constrained situation. A fan means the atmosphere has not decided yet, and no amount of staring at one model will resolve it.
Drag across the chart
| Horizon | What it is good for |
|---|---|
| 0–12 h | Timing. Reliable enough to plan around, though the model still cannot see a squall that has not formed yet. |
| 12–48 h | The working window. Strength and direction are usually right; timing can slip a few hours. |
| 2–5 d | Pattern, not detail. "Windy midweek" is sound; "18 knots at 3pm Thursday" is not. |
| 5–10 d | A hint. Only worth acting on when every model agrees, and even then loosely. |
| 10–14 d | Entertainment. The chart is drawn because the model runs that far, not because it knows. |
Wave forecasts sit downstream of wind forecasts, so they inherit every wind error and add their own. They also lag: swell generated by a storm error two days ago is still arriving today. Treat wave skill as roughly one notch worse than wind skill at the same horizon.
Predict first
You check a Saturday six days out. ECMWF says 12 kt, GFS says 26 kt, ICON says 17 kt. What is the useful conclusion?
B. ECMWF does score best on average, but "best on average" is not "right on this Saturday" — and a 14-knot spread is the models telling you the outcome is not yet determined. C is the worse error: averaging manufactures a confident number out of pure disagreement, and 18 kt may be the one answer no model actually supports.
Limits
These are not bugs to work around. They are consequences of putting a global model on a zoomable map, and knowing them is most of what separates a careful reader from a careless one.
Your pin
Worth saying plainly, because it changes how you should use that link: waves is close to the least informative layer for this position. The strait is narrow, shallow and sheltered on both sides, and the global wave grid barely resolves it, so it renders a flat, near-uniform wash. At zoom 5 most of what you are seeing is the South China Sea and the Andaman, not the water under your pin.
Read wind, gusts and thunder here instead, and let the sea state follow from them.
Predict first
The waves layer over the strait looks almost uniformly calm all week. What have you learned?
B. A uniform layer in confined water is usually the resolution talking, not the weather. C happens to be true — the strait is well sheltered from ocean swell — but it is not what the layer is telling you, and it is not what will spoil your day. The thing that will is a squall, and you find that on gusts, thunder and radar.
Practice
Genuinely sequential, so worth doing in order. Each step either changes what you do next or tells you to stop looking.
| You see | You should think |
|---|---|
| Hs = 2 m | Largest wave near 3.7 m |
| Period 5 s | Crests 39 m apart — steep local chop |
| Period 12 s | Crests 225 m apart — ocean swell, travelled far |
| Gust ÷ wind > 1.5 | Convection somewhere in that grid box |
| Wind × 1.4 | Sea state roughly doubles — it scales with the square |
| Spread > 30% | Not a forecast yet. Re-check nearer the time. |