An issue with the ERA5-Land monthly averaged data from 1950 to present dataset was identified in January 2024 on both CDS discs and MARS tapes.
It affects the following part of the dataset:
Product: Monthly averaged reanalysis
Time period: from September 2022 to February 2024
Area: Global
Variables: all accumulated variables (please refer to Table 3 in ERA5-Land documentation)
Incorrect data for period September 2022 to February 2024 is no longer available to download using the interactive CDS Download data form over the web, but remains accessible for the time being via CDS API or direct MARS access. **Incorrect data from September 2022 to February 2024 will not be replaced on CDS discs nor MARS tapes. Instead users can recover the same data using the product monthly averaged reanalysis by hour of day at 00:00.
A fix is planned to be implemented for ERA5-Land monthly averaged data stored on CDS discs in time for the release of the March 2024 monthly averaged data (i.e. from March 2024 onwards).
For reference, this issue has been recorded as a “Known issue" in the ERA5-Land: data documentation.
We apologise for any inconvenience this may have caused.
Should you have any questions, please contact us via our Support Portal.
ECMWF Support
on behalf of the Reanalysis team
Corrected ERA5-Land monthly averaged data (accumulated variables) now available from March 2024 onwards.
Incorrect data from September 2022 to February 2024 will not be replaced on CDS discs nor MARS tapes. Instead users can recover the same data using the product monthly averaged reanalysis by hour of day at 00:00.
I cannot believe the incorrect September 2022 to February 2024 data is not being replaced on the servers, that is not a decent long-term solution at all(!) How do you ensure that users aren’t unknowingly downloading the “Monthly averaged reanalysis” data and using it in scientific publications? This is a huge issue affecting loads of scientists using this data.
And here we are, one more year later, and of course I ran into exactly this problem…
I completely agree with @Dave_van_Wees posting. This is not OK. Absolutely not.
Expecting from users to fix the data on their own is not efficient when the source has understood the causes and can do this for everyone once. Furthermore, leaving flawed data accessible in the default access option inevitably leads to unintended errors, wrong trend estimates, etc. pp. you name it. In this case, users might only notice the problem b/c it is so obviously large in most of the affected variables. But not everyone looks into the Known Issues each time they load data. There must be a better solution.
Unless I completely misunderstood the top comment from 2024-02, it is not true that the false data are no longer available from the Download data form. I say that because in this case this is exactly how I requested and downloaded the data: Filling the form (monthly avg rea), including several affected variables (‘e’, ‘tp’ amongst others), globally, netCDF; waiting until processed, loaded, used the data.
And the faulty period very much is in the requested data. Is this really desired by ECMWF? I find this very, very hard to believe and it does not agree with the standards I firmly believe ECMWF should follow (and most often is, and I appreciate that a lot!).
I know first hand that I am not the only person running into this issue. You can find multiple forum entries where people have no clue what is going on.
Yes, there might be the option to retrieve the correct values via the 00:00 hourly option, but this cannot seriously be the long-term solution. It leaves me kind of speechless. This can be done better.
We strongly believe that the incorrect data for September 2022 to February 2024 should be formally corrected and replaced in the official “monthly averaged reanalysis” product.
Leaving this period permanently uncorrected is likely to cause ongoing confusion and potential problems, especially for users who are unaware of this issue or have already downloaded and used the affected data.