One weather hour, two forecast leads

Suppose you want to compare the GFS forecast for 00:00 UTC on October 31. In the October 30 18:00 UTC run, that is F006. In the October 30 12:00 UTC run, it is F012. The initialization times differ by six hours, but both graphics describe the same target hour.

A time-alignment example, not a current weather forecast
Model runForecast leadValid time
Oct 30 · 18:00 UTCF006Oct 31 · 00:00 UTC
Oct 30 · 12:00 UTCF012Oct 31 · 00:00 UTC

The rule is simple: valid time equals initialization time plus forecast lead. To find the other run’s lead, subtract its initialization time from that shared valid time. Try changing the run times in the calculator below.

How comparison works in Model Lab

Open a parameter, choose Compare, and select the model and run for map B. The slider shows the intersection of the published valid times for the two selected products. If one run has hourly steps and another has three-hour steps, only the shared times are available for comparison.

This is different from filling the gaps with invented intermediate frames. If the runs do not share a published valid time, the viewer says so. Changing the map region also checks model coverage; a CONUS-only grid cannot be used for a Hawaii comparison.

Separate timing, placement, and intensity

First ask whether a feature is arriving earlier or later. Then look at where it is placed and how strong the field is. For a rain band, compare the leading edge and the period of heaviest rain as well as the maximum accumulation. Two runs can give a similar total to different places or during different hours.

Keep the parameter, units, map region, and optional layers consistent. A total accumulated since initialization covers a longer period in an older run; matching valid times alone does not make those accumulation periods equivalent. Use a common interval when making that comparison.

A change is evidence to examine

A later run incorporates a different forecast starting point, but a single dramatic frame does not establish a reliable trend. Compare several runs and more than one model when coverage permits. Agreement can help identify a consistent scenario; disagreement is useful information about forecast uncertainty.

Two deterministic models are not an ensemble probability. Counting how many of two maps show rain does not produce a calibrated chance of rain. Use the official local forecast and supported probabilistic products for that question.

Share the comparison with its context

Include each model’s initialization time, the common valid time, and the parameter when you describe a comparison. In the viewer, Export A exports the primary graphic, so a download from comparison mode is not automatically a two-map montage. The shared page link preserves both selected runs.

The calculator performs clock arithmetic. It does not query a model’s archive or guarantee that a parameter exists at the calculated lead. Model Lab makes those availability checks when it builds the comparison.

TRY THE CALCULATION

Align two model runs

All inputs and results below use UTC. Change a run time or lead to see which forecast hour in run B matches run A.

Run A F006 and run B F012 both verify at Oct 31, 2026, 00:00 UTC.

Time arithmetic only. Model Lab checks whether that forecast hour and parameter are actually published. Your entries stay in this browser; this tool makes no data requests.

Sources & corrections

This guide explains the Mammatus website’s behavior and uses the original product definitions below. The publisher is Mammatus Weather; it is not an agency forecast or an endorsement by a data provider.

Found an error? Send a correction with the page link and the detail that needs checking. Read our sources and methods.

Compare forecasts in Model Lab