Incomplete SEAS5.1 (ECMWF 51) seasonal forecasts dataset (March 1st init)

Dear all,

I’ve been working with SEAS5(.1) for a several years and so far I had no issue with the download process. I wanted to update my dataset and notice that 2024 and 2025 contained more ensemble members. So I checked earlier years, and noticed that the data I had downloaded in Dec 2024 were not anymore 100% compatible with the updated runs (spatial correlation ~0.9-0.95). So I though I’d re-download everything to avoid mixing data. So far all good.

I’m downloading several fields, but let’s take Z200 as an example. I’m selecting 1981-2025 (each year downloaded separately) and two initialization dates, May 1st and March 1st. For May 1st, I get exactly the number of time steps I select (01.05→ 02.10, 154 days).

For March 1st, I ran into troubles. I’m selecting all available time steps (01.03-01.10, 215 days) but when I check the data, the first 42 ensemble members only contain March→August and Sept and Oct are missing. I tried downloading both via the website and via the python api script and the result is the same.

Has anyone encountered the same issue? Is there a reason this is happening?
One can cut out the incomplete ensembles but it is a potential risky source of errors in the preprocessing of the data.

Any hint on how to proceed or explanation on why is happens would be great!
Thank you in advance,
best wishes,

Giorgia

One more thing, I see the problem only with variables on pressure levels (Z200, TPL 850, U850 and V850)
while T2m, OLR, SST and MSLP are fine

Hello @Giorgia_Di_Capua !

I have had a quick look at what you reported and I found a couple of intriguing (unexpected) behaviours on the CDS, that deserves some investigation. I was going to propose you to record this as an issue in the ECMWF support system, but I have just seen you already did that a couple of days ago. Someone at ECMWF will follow-up that issue with you through that route.

Regards!

Eduardo Penabad
C3S seasonal forecasts production analyst