Datasets:

openclimatefix

/

eumetsat-rss

like 1

Hi,

Thanks for pointingthose out! And yeah, there has been some issues with these files. I've been redoing these datasets (into yearly large files), so they should be available covering more time and more completely very soon. That will be replacing all of these files with newer ones. Not sure the easiest way to check the other missing files unfortunately.

Thank you Jacob,

For missing files, I think that for every year the files are named 20??_0000??-of-0000??.zarr.zip. giving this format I believe that the missing files can be found easly by looking for the missing numbers.

Personally, I don't have any issues with the current format, it's very convenient. thank you for sharing it.

Yes, there are those missing ones, but it is a lot easier as well with just yearly Zarr files, so the same format, just with a whole year instead of 1000 timesteps each. It should be easier to work with, and there are some other changes to how the data is compressed and chunked that should make it faster to read from. But yeah, glad you like the format!

The first one of the newer format has now been uploaded. More will be coming!

Thank you for uploading the new files, I just want to point out also that 2019 presents a lot of duplicates (20%). I hope that the new data resolve this problem

Yes, this should fix it. As the yearly zarrs these new files are being created from are dedpulicated, it should mean that doesn't happen anymore.

Hi,
I'm wondering why the new files are too heavy: for example in the old data of 2014 (nhv) we have 95 zarr files with 3 Gb each (2014 is a good year without missing or broken files by the way) and now we have 61 files with 50 Gb each ?

We changed the compression, so that it loads much faster, and in tests we were doing (documented here: https://github.com/openclimatefix/ocf_datapipes/issues/132) the JPEG-XL compression we were using for the previous version was a lot slower, and didn't compress much more than the Blosc2-Zstd. Times when the outputs were very small corresponded to there being a problem with the saved data (all the same value, all NaNs, etc.). The times when it did there was some issues with the output data. Each of these files also contains more timesteps than the previous one. And has a lot more timesteps that were missing from the previous version as well.

Hi,
May I please ask when we can expect to receive all the files for this dataset?

Hi, we are still working on redoing the data, but are nearly finished with it. We are also working to make the data public through Google Public Datasets, so adding them here has been a bit more delayed. I would guess another 2 months or so? If its not public through Google, then we'll start uploading them all here again in the parts. The good news is that we now have the entire archive saved, from 2008 to Feb 2023, so it is larger.

Thank you for the update! It's great to hear that you're making progress on redoing the data. I understand that the process has been delayed, but I appreciate your effort to ensure the accuracy and accessibility of the data. thank you so much for your prompt response and for the data.

Google has said they have started pulling the data to the public datasets, so I would expect it to be ready in the next few days/weeks, and I can point to that. I'll still add a dataset script here to use them, but for most cases, it might be easier just to access the data from the public buckets.

The data is now available here: https://console.cloud.google.com/storage/browser/public-datasets-eumetsat-solar-forecasting/satellite/EUMETSAT/SEVIRI_RSS/v4

Thank you so much for the update on the availability of the EUMETSAT solar forecasting satellite data on Google's public datasets.