GEFS v12 AWS data - seems to have few discrepancies in the data: missing files, missing documentation. #809
Comments
|
Reaching out to see if I can find out the answer to this |
|
Hi Zac, I also noticed that the lead time resolution goes from 3 hourly to 6 hourly when I move from D1-10 to D>10 . This is not in the documentation. |
chiaral
changed the title
|
Hello - I am expanding this issue, because I am finding more un-documented things as i move through the data. In particular, for the Day 1-10 the data should be at a 3 hourly interval. That makes for 80 time steps (8 per day for the first 10 days). I have quite a few variables, however, that when read with The two What does happen if I plot the two time series? for another variable where the quantities is not an accumulated measure but an average (but for which I have the same double coordinates) I have:
When I look into the file through a wgrib2 reader, I have 80 time steps (corresponding to the 80 3 hourly interval in 240 hours) but these are the deltas (sorry I will copy everything for completeness) : Note how: I have the first 3 hourly interval, then a 6 hourly, then a 3 hourly from 6-9, then a 6hourly... and so on.. so the 3-6, is missing, so is the 9-12 - and it is instead swapped with the 6hourly accumulation. I also want to note that the Delta time steps in the Can we have someone from NOAA chiming in? Thanks SOOO much! |
|
for completeness - the delta of a regular variable with just one forecast_time coordinate looks like this (albeit it's not an interval). |
|
After reading the table with more attention - and not the documentation only - this behavior is documented! so it's all good - it might require some wrangling in the zarr version but I will figure it out. The missing files tho and the different resolution for D>10 doesn't seem to be in the documentation. |
|
Ok - here I finalize the list of missing documentation to the best of my knowledge - so far - in the dataset. The link to the Presentation "Meanwhile, a presentation on GEFSv12 can be found here." listed in the documentation is not publicly available - I get a "you need access" warning. Probably some of these missing details are clarified in that presentation, but I cannot be sure.
in fact, if I download one file for accumulated convective precip and use wgrib2 to explore it: for Tmax However I have noticed this varying resolution in other variables that are not listed in the documentation, for the averaging period: Downward Long-Wave Rad. Flux Downward Short-Wave Radiation Flux Ground Heat Flux Latent Heat Net Flux Water Runoff Momentum Flux, V-Component Upward Short-Wave Radiation Flux Upward Long-Wave Rad. Flux Upward Long-Wave Rad. Flux Momentum Flux, U-Component Total Cloud Cover [%] Sensible Heat Net Flux Large-Scale Precipitation (non-convective) For some of these variables it makes sense (I expect NCPCP to have the same resolution as the other accumulate precip products) but I was not expecting some of the others. Two are the cases - either these variables are indeed accumulated/averaged with this alternating resolution, or the grib files have an erroneous forecast time width. For the precip/tmax it's obvious the different resolution (especially for the accumulated variables), in fact, as I reported above, the 2 time series (the one referring to 3hours width forecast period, and the one for the 6hour width data) are one half the other. Below I plot the 2 time series, in particular they are the global average of the files with 2000010100_c00 start time/ensemble - for APCP for NPCP Water runoff too... but for the average variables it's harder to say whether they are averaged on different time periods: for LHTFL for TCDC I think this is all from me now. |
|
Chiara, all pressure levels after day+10 are in one file without the "_abv700". The documentation will be updated to reflect this. |
|
This issue has been addressed by @ThomasMoreHamill - with the new documentation https://noaa-gefs-retrospective.s3.amazonaws.com/Description_of_reforecast_data.pdf Thanks so much! |
I have done a quick random check (I have not checked all folders) but it appears as all the files related to subset of fields above 700mb are missing for the D>10 days forlders.
In the documentation it says:
I initially thought that because of the similar lat/lon grid they had one grib file for the whole Day 1-16/35 period, but nope, the file in the Day 1-10 folder has only lead times for 3 to 240 hours.
The file names are
where for
ensi mean the ensemble members,c00andp##The same issue seems to be present for those days with 11 members and forecast times up to Day 35.
The text was updated successfully, but these errors were encountered: