How We Forecast

The three-tier model

Every date you plan around falls into one of three tiers, and the app always tells you which:

• Real forecast: 0 to roughly 14 days out. A genuine weather forecast, sourced from the best-performing model for each location.
• Extended-range: roughly 14 days to 7 months out. A seasonal model estimate, less precise than a forecast but still grounded in current model runs.
• Climatology: roughly 7 months to 24 months out. What the weather has historically done at that time of year, not a prediction of what it will do this time.

The full planning horizon is 24 months. Nothing on this site presents climatology as a forecast, or a forecast as more certain than the model agreement actually supports.

Sources

Real forecasts and extended-range estimates combine the best-performing models per region, drawing from national weather agencies including NOAA (US), ECMWF (Europe), DWD (Germany), Météo-France, and Environment Canada.

Climatology draws on a separate source: ECMWF's ERA5 reanalysis, via the Copernicus Climate Change Service. It blends real station, aircraft, buoy, radar, and satellite observations with a physics model to build a consistent global grid, not a climate-change projection.

The confidence indicator

Every forecast carries a confidence indicator reflecting how far out you're planning and how much the underlying models agree with each other. Close to your travel dates, it's high. Months out, it's a guide to model agreement, not a promise. Treat it as guidance, not certainty.

The Travel Window Score

Climatology-tier destination pages carry a Travel Window Score: a 0–10 rating per month, used to identify a city's best travel window. It's a weighted composite of exactly three inputs, the only climatology-tier variables in our dataset:

• Daytime comfort (40%): how close the month's average daytime high sits to 24°C, our comfort anchor.
• Rain-day count (30%): how many days that month typically see at least 1.0mm of rain (the WMO wet-day threshold).
• Precipitation total (30%): the month's typical total rainfall.

When two adjacent months score within half a point of each other, the site shows a two-month "shoulder span" instead of a single month, because the underlying data is monthly-resolution only. We never interpolate a week or a date out of it.

Why our numbers can differ from official normals

Historical weather figures on this site come from a climate reanalysis dataset (ERA5), not a single weather station. A reanalysis blends real observations (weather stations, satellites, radar, buoys, aircraft) with a physics model to produce a consistent grid covering the whole planet, so every destination gets a value, not just the ones near an official station.

Two things follow from that, and they're separate:

Different basis. A national weather agency's published "normal" (Météo-France, the UK Met Office, Japan's JMA, and others) reports what one specific station measured. Our figures describe a wider grid area around each destination. For a city with an unusual microclimate, or a station in a specific spot within a larger metro area, these can genuinely differ.

Different period. Official normals are typically a fixed 30-year reference window (commonly 1991–2020). We use a rolling 10-year window instead (the last complete decade, computed at render time) because a recent decade describes what a traveler is likelier to encounter than a benchmark that can be decades old by the time you read it.

A documented pattern for Europe, specifically. Published research comparing ERA5 to dense rain-gauge networks (Bandhauer, Isotta, et al., 2022, International Journal of Climatology) found ERA5 systematically overestimates precipitation across the European regions studied, tied to counting slightly too many wet days. Paris, for example, runs roughly 20% above Météo-France's published normal, consistent with that documented pattern. If you compare our figures for a European city against an official station average and ours runs higher, this is likely why. We're disclosing that rather than adjusting our numbers to match official figures. A "corrected" number we can't independently verify is worse than an honest one that's sourced and explained.

What we don't claim

We don't claim precision beyond what a tier's nature supports. A real forecast is a prediction; an extended-range estimate is a model's best guess; climatology describes a pattern, not a promise for your specific dates. We make no claims about variables we don't measure, and no superlative on this site is written by hand. Every "driest month" or "best time to visit" is derived directly from the data next to it, so it can never drift out of sync.

How to cite us

Every climatology-tier city page has a "Cite this page" button with a ready-to-copy citation and a clean canonical URL. If you're a journalist or writer working from our data, use the canonical URL of the page you're citing (for example, weatheritinerary.com/explore/weather-in-paris). Questions about the methodology? Reach me directly at leo [at] weatheritinerary.com. I read everything.

Who's behind this

I'm Leo Vicente, the founder and engineer who built WeatherItinerary. No "Travel Team" byline, no anonymous editorial voice. This methodology is something I'm personally accountable for.