Comparing seasonal models and monthly runs

Two ways to compare seasonal forecasts: several centres at one initialization, or several monthly runs from one centre. When each comparison is useful.

Written and maintained by seasonforecasts · Published 11 September 2026 · Reviewed 12 September 2026 · Sources: C3S seasonal forecast catalogue and product descriptions

Two questions, two modes

The workspace has two comparison modes. Model mode places several centres side by side for the same initialization, and answers “do the models agree?”. Run mode places several initializations from one centre side by side, and answers “is this model’s outlook stable?”. In both, every panel shares the valid period, the field, the forecast type and the colour scale, so the display supports a consistent comparison. Each system still has its own hindcast distribution and forecast characteristics.

Agreement between models

Agreement across different systems is useful context, but these systems are not statistically independent: they can share observations, model components and errors. Agreement is not a calibrated probability. Look for the sign of the anomaly first and its magnitude second. Models differ in how strongly they express a signal, so a weak anomaly in one centre next to a strong one in another is agreement on direction, not disagreement. Models can also strongly favour opposite outcomes. The tercile summary is calculated within each model, so disagreement between models does not itself make a panel neutral.

Stability between runs

Similar outlooks in successive runs indicate persistence despite updated initial conditions. A change can reflect new information, ensemble sampling or a system update; it does not prove that the earlier signal was weak. For the same valid period, a newer run has a shorter lead but need not be sharper or more extreme. Successive runs are related forecasts, not independent confirmations. Check the system labels when comparing runs across an upgrade.

Using point inspection

Click a location to get the source-grid value from every open panel at once. Values come from the nearest node in each model’s C3S 1° output grid. They describe a broad grid area, with differing model terrain and coastal representation, rather than a local city forecast.

Sharing a comparison

The URL records the mode, panels, field, forecast type, period, region and whether navigation is linked. Copy it to share exactly what you are looking at, or export the board as a labelled PNG with source, system version, initialization, lead and valid period on each panel.

Sources and method

These references explain the underlying science and provider conventions. Descriptions of this site’s calculations, colours and coverage were checked against its processing code and imported data.