How Our Forecasts Work
WFY24 combines DWD ICON forecast models, short-range nowcasting and localized place metadata to assemble each weather page.
Actual city weather page flow
This diagram follows the real weather-page architecture used by WFY24, while staying at a public-safe level of detail.
/{locale}/weather/{slug}
The user requests the canonical weather page for a specific place slug.
Canonical place resolution
The page resolves the slug into a canonical place object with coordinates and localized geographic context.
Forecast assembly
The initial HTML is assembled on the server from forecast and climate layers so the page can render fast and cleanly.
Current / hourly / daily / nearby
The initial page is rendered with the core weather sections already present for SEO and fast first render.
Live refinement
After render, the live nowcasting layer can refine current conditions and short-range weather signals.
The product clearly separates the core SEO render from the live overlays that are added afterwards.
Model layers already used by the product
The facts here are already documented publicly in the product and do not reveal internal mechanics.
2 km grid for Central Europe, 0-48 hours, 8 updates daily.
7 km grid for Europe and nearby regions, roughly 30-120 hours, updated every 3 hours.
13 km worldwide grid, 0-180 hours, runs at 00/06/12/18 UTC.
Short-range layer on top of the base forecast for live conditions and 0-6h nowcast.
1. Forecast model stack
WFY24 is built around the DWD ICON model family. Depending on geography and lead time, forecast pages can draw on ICON-D2 for Central Europe at 2 km resolution, ICON-EU for Europe at 7 km resolution, and ICON-Global for worldwide coverage at 13 km resolution.
In the current product documentation, ICON-D2 covers 0-48 hours and updates 8 times daily, ICON-EU covers roughly 30-120 hours and updates every 3 hours, and ICON-Global covers 0-180 hours with 4 daily runs at 00, 06, 12 and 18 UTC.
2. Update cadence and nowcasting
The base forecast refreshes whenever new model output becomes available. For the shortest range, WFY24 also uses a proprietary nowcasting layer powered by XGBoost, to blend real-time observations with model guidance for the next 0-6 hours.
This is why some WFY24 experiences can surface faster-changing signals such as real-time temperature, precipitation timing, lightning and severe-weather context without waiting for a full global model cycle.
3. How a place page is assembled
A weather page is not just a raw grid cell dump. WFY24 resolves a place slug, loads the place metadata and coordinates, fetches forecast data for the place itself, then formats the output for the active locale, units and page type.
- Localized place names and breadcrumbs are applied when available.
- The forecast is tied to the place coordinates so the homepage card and the place page stay aligned.
- Time zone, sunrise/sunset, air quality and nearby modules are added when that page template supports them.
4. Accuracy and limits
WFY24 already explains in its FAQ that forecast accuracy is usually strongest in the first 1-3 days and becomes less certain further out. Mountain ranges, coastlines, convective storms and local microclimates can all diverge from a model grid average.
WFY24 is designed to make model output understandable and local, but it does not replace official warnings from national meteorological or civil-protection agencies during high-impact events.