HRRR

At a glance

Convection-allowing, which is the point.

Most weather models are too coarse to contain a thunderstorm, so they estimate the effect of convection statistically rather than simulating the storms themselves. At 3 km, HRRR is fine enough to grow individual storms in the model, which is what makes a forecast radar image meaningful at all.

That is also the honest limitation. A convection-allowing model produces storms that are realistic in character — the right kind of weather, in roughly the right area, at roughly the right time — without being right about any particular cell. Treat the shape and timing of a system as useful, and the exact position of a cell as not.

Why the end of the timeline moves.

HRRR runs every hour, but only four of those runs each day go the distance. The hourly runs reach 18 hours ahead; the 00, 06, 12 and 18 Z runs reach 48.

The app always prefers the newest frame available for a given hour, so the timeline is a mix: recent hours come from a run that may be minutes old, while the far tail comes from the last synoptic run. As that run ages, the tail gets shorter — a 48-hour run launched at 12 Z only reaches 36 hours from now by 00 Z. When the next synoptic run posts, the tail springs back out.

This is why the slider length appears to breathe over the course of a day. Nothing is wrong; you are seeing the age of the most recent long run.

When frames appear.

A model run is not available the moment its cycle time passes — it has to be computed and published first. Measured against the published data: HRRR posts its 19–48 hour frames about 1 hour 45 minutes after the cycle time. Shorter-range hours arrive sooner.

The app polls for new frames continuously and shows the newest it has. There is no refresh button because there is nothing for you to trigger.

What Radar Forecast takes from it.

• Composite reflectivity — the forecast radar frames themselves, rendered with the same colour table as the observed MRMS frames so the two are comparable.
• Winds aloft — the wind field at each level the wind table reports, sampled along the route every 50 nautical miles at the time you cross each point.

Observed radar to the left of "now" is MRMS, not HRRR. Surface fronts come from WPC and forecast weather type from NDFD; those are separate products on the same timeline.

Compared with what actually happened

Rather than ask you to take the model's word for it, the site publishes side-by-side frames of the forecast against the observed radar for the same valid time.

See forecast versus actual →