Spaghetti chart and ERA5 verification: decoding the forecast history

1. Why overlay all the runs?

A weather model is relaunched several times a day. Each launch — called a run — produces a new forecast for the following days. For a given target date, there are therefore dozens of runs issued over several days, all of which forecast the conditions of that moment from different distances in time.

Most apps only show the latest run. That's the "photo" — useful, but it says nothing about how the forecasts have evolved over time.

Solano's spaghetti chart overlays all these runs on a single chart. The image that forms — tight or scattered, converging or chaotic — gives information that the value alone cannot: was the situation predictable?

      For each atmospheric model Solano analyses (AROME HD, ICON-EU, ARPEGE, ECMWF, GFS, UKMO…), the spaghetti chart reconstructs up to 5 days of run history — that's several dozen curves for models that run 4 to 8 times a day.
    

2. How to open the Spaghetti panel

Place a point on the map and wait for the analysis to load. In the results panel, each model is shown with its stability score. Click a model to open the corresponding Spaghetti panel.

The Spaghetti panel opens on the right (on desktop) or at the bottom (on mobile). By default it shows the Mean wind variable for the day selected in the strip.

Shortcut: the stability score shown on the strip matches what you'll see in the spaghetti — an 8/10 score means grouped curves, a 3/10 score means scattered curves.

3. Reading the chart: axes, colours, tabs

The axes

Horizontal axis — the hours of the selected day, from 0h to 23h.
Vertical axis — the value of the displayed variable (knots for wind, mm/h for rain, degrees for direction…).

The curve colours

Each curve is a different run. The colour indicates its relative age: recent runs are drawn in a saturated shade, older runs in a pale shade. The most recent run is always the most prominent curve — it's the "current forecast".

The variable tabs

At the top of the chart, a set of tabs lets you change the displayed variable:

Wind — mean wind speed in knots
Gusts — maximum speed in knots (often more unstable than the mean wind)
Direction — wind heading in degrees (note: 355° and 5° are almost identical)
Rain — precipitation in mm/h
Temperature — in °C
Cloud — cloud cover in %
Pressure — sea-level pressure in hPa

Curves tab of the Consistency panel — multi-model spaghetti with wind, gusts and marine data — The Curves tab — each coloured line is a different run. Atmospheric models (ARPEGE, ECMWF, ICON) and marine models (waves, currents) are shown separately. The left sidebar lets you filter by variable and by model.

Advice: always start with Gusts for sailing. Gusts are more volatile than the mean wind and their run-to-run stability matters more for safety.

4. The patterns to recognise

A few configurations come up regularly. Learn to identify them and you'll read a spaghetti in a few seconds.

📌 Tight curves — strong consensus

All runs overlap almost perfectly. The forecast has barely changed for several days. High stability score (8–10).

→ Reliable forecast. Plan with confidence.

📌 Scattered curves — structural uncertainty

Old and recent runs diverge markedly and with no clear trend. The weather situation is inherently hard to predict for this model. Low score (2–5).

→ Don't commit. Recheck at J-1.

📌 Curves converging towards the right

Older runs diverged, but recent runs are tightening. The situation is clarifying as the target day approaches. Score rising.

→ Positive signal: the forecast is gaining accuracy. Watch 1–2 more runs.

📌 Curves diverging towards the right

Recent runs spread apart while older ones agreed. The models are losing confidence as the target day approaches — often the sign of a disturbance that's hard to locate. Score falling.

→ Warning sign. The forecast is degrading — maximum vigilance.

📌 Late swing — a single very offset run

All runs agreed, then the latest recent run breaks sharply with the trend (e.g. +10 knots all at once). It could be a genuine change of situation or an initialisation failure.

→ Wait for the next run to confirm. An isolated swing should be taken with caution.

Very tight runs, high stability score — **Tight runs** — the curves almost overlap: successive runs tell the same story, score 8.2/10.

Spread runs, divergent forecasts — **Spread runs** — same area, a different variable and a different model: the runs no longer agree, the uncertainty is real.

5. The ERA5 Verif. tab — forecast vs. reality

The 4th tab of the Spaghetti panel — 📡 Verif. — adds another dimension: the comparison with what actually happened.

ERA5T: the reference reanalysis

ERA5T is ECMWF's near-real-time atmospheric reanalysis. It's not a direct field measurement: it's a coherent hour-by-hour reconstruction of the state of the atmosphere, obtained by data assimilation of sparse observations (weather stations, radiosondes, satellites, buoys) into a state-of-the-art numerical model. It is available with roughly a 1-to-2-day delay and represents the best available estimate of what happened.

In the Verif. tab, the ERA5T line is drawn as a black dashed line. The coloured curves are the selected model's runs for that same day. The question becomes simple: were the models right?

ERA5 Verif. tab — black dashed ERA5T line and coloured ARPEGE, ECMWF, ICON runs — The ERA5 Verif. tab — the black dashed line is the ERA5T reanalysis. The coloured curves (ARPEGE in blue, ECMWF in green, ICON in brown) show the successive runs issued for that day. The day selector in the sidebar is independent of the main strip.

How to navigate past days

A dedicated day selector appears in the Spaghetti panel in Verif. mode — independent of the main strip. You can therefore compare past forecasts without changing the selected day in the rest of the app.

The last 15 days are available. The run labels show "J-N · DD/MM" to place each curve relative to the observed day.

What Verif. does not cover

Recent ERA5T doesn't provide every variable. Wind direction, visibility and marine data (waves, currents) are not available in this tab. The variables covered are: mean wind, gusts, precipitation, temperature, cloud cover and pressure.

Advanced use: the ★ button in the day selector jumps straight to J-1 at noon, with the 12Z runs — the most interesting window for seeing how the models forecast yesterday against the ERA5 reality.

6. Usage tips

Compare several models. If ECMWF shows tight curves but ARPEGE scattered ones, the two models don't agree — a sign of uncertainty that the multi-model score will also capture.
Look at gusts first. Gusts are the most volatile variable between runs. A scattered "gusts" spaghetti while the mean wind is stable indicates peak uncertainty that conventional weather hides.
Use ERA5 Verif. to calibrate your confidence. If a model is systematically right at J-2 on your spot, you can give it more credit. If its J-4 runs are always far from ERA5, they deserve less attention.
Low-score days are worth watching. If a day's score is low at J-3, come back at J-2: it frequently happens that the score rises as the models converge on the approach.