Lead time, initialization and valid period

How seasonal forecast lead months are counted, why lead 1 is the initialization month, and how to line up different runs on the same valid period.

Three dates on every map

A seasonal forecast has an initialization, the month in which the run started; a valid period, the month or three-month window the map is about; and a lead, the number of months between them. On this site lead 1 is the initialization month itself, following the C3S convention. A September run therefore covers September at lead 1 through February at lead 6.

Comparing runs on the same valid month

The interesting comparison is rarely lead against lead but valid period against valid period. November as seen from the August run is lead 4; November from the September run is lead 3. This site always aligns panels on the valid period, so when you compare monthly runs you are looking at the same target month from successive initializations, with the lead shown on each panel.

Available months are intersected

When panels use different initializations, only the valid periods that every panel can show are offered. Two runs one month apart share five monthly periods; a three-month window has fewer, because each window needs three forecast months inside the six-month horizon. If a month you expect is missing from the picker, one of the open panels cannot reach it.

Lagged ensembles and nominal dates

Some centres, notably the UK Met Office, do not start all members on one day but run a few every day and assemble the most recent ones. C3S assigns these a nominal initialization month so they line up with the rest. The site uses that nominal month and validates it on import, so a “September” GloSea6 forecast really contains the members C3S associates with September.

Skill falls with lead

Longer leads are less certain. Lead 1 partly overlaps weather that is already unfolding; by lead 5 or 6 the ensemble spread is wide and the mean anomaly weak. Treat long-lead maps as an indication of the direction the model leans rather than a forecast to plan around. See comparing seasonal models and runs for how to use successive runs.