A difference between reference time and update time

Gkikas LGPZ

@Korina
One picture says more than 1000 words!

Mark_Marsh

Thanks, for share this article

Md Eyasin Ali

@Korina India & Bangladesh post see

Korina

@Md-Eyasin-Ali Hello, what kind of information are you looking for?

Md Eyasin Ali

@Korina ok

Paspox

@Korina Thanks for this post.
I'm interested in the GFS model in particular.
When a new grib file is released by the NOAA, how long does it take Windy to show it ? Is it pretty much instantaneous (a few minutes) or does it take longer ?

vsinceac

You have full info about currently selected model via "i" control (screenshot).
About GFS: it is computed every 6h by NCEP/NOAA, starting from 00h UTC. On NCEP/NOAA site, it is available about 3h40m after each runtime and it may take a while to download and ingest it on server side: 1 full grib2 file (all parameters/levels) for one date/time validity is about 500 MB in size...

Paspox

@vsinceac Thanks a lot for your reply. When you say "it may take a while ..." do you have a rough estimate of how long that is ? Are we talking something like 1h or much less ?

Also, I'm curious about the file size that you mentioned, 500MB : do you know how big is the grib2 file, covering the whole 16day forecast ?

vsinceac

There is no 1 grib file for all validities: each of them comes in a single, dedicated grib file. There are 209 grib files for each runtime, from f000 to f384, where the number is in hours related to runtime. From f000 to f120, validities are provided every 1h, then every 3h up to f384.

You can check e.g. this one (runtime today at 06hUTC, validity for 07hUTC, 515.1 GB, all params/levels): https://nomads.ncep.noaa.gov/cgi-bin/filter_gfs_0p25.pl?file=gfs.t06z.pgrb2.0p25.f001&dir=%2Fgfs.20211014/06/atmos.
Here you can select the day, the runtime, the forecast, then optionally only partial parameters/levels/coverage to download: https://nomads.ncep.noaa.gov/cgi-bin/filter_gfs_0p25.pl?

Download time of course depends on many factors:

internet speed/latency/bandpass/etc. (it may be also regulated sometimes by NCEP/NOAA itself)
one can download only part of available grids (parameter/level) in a grib2 file (size to download would be smaller)
I don't know if Windy downloads all 209 validities (size to download would be smaller)
full size of 209 grib2 files for a runtime is about 110 GB
for an average download speed of 30 MB/s from NCEP/NOAA servers (which is not easy to obtain e.g. from Europe), you should download all grib files of a runtime in about 1h

Then I guess Windy servers spend also some time to ingest the downloaded grib2 files before being able to serve map data layers to client side.

Paspox

@vsinceac Thanks a lot for such a thorough reply. Slowly, thanks to people like you, I understand more and more about the details of these weather updates.

FYI, my interest in the "precise" timing of Windy is because I play an e-sailing game (called Virtual Regatta) and it uses real world weather data (GFS model, only wind speed and direction).
So, for example, when the 00z file is released (@3h40 UTC), the game's weather starts to update (takes around 2h) and often, it is interesting to look at Windy but I wanted to know if what Windy shows is still the old (18z) data or already the new one (00z) and when/how it goes from one to the other.

Again, thanks for your reply.

Paspox

@vsinceac Me again, with a few more questions, still concerning the GFS model.

In your message, you said "runtime today at 06hUTC, validity for 07hUTC, 515.1 GB, all params/levels"
I believe you meant 515.1 MB, right?
515MB x 209 validities = 110GB for a full run.

Do you know of a website where I can follow live the release of the different validities by the NOAA?
I know about this one (https://www.nco.ncep.noaa.gov/pmb/nwprod/prodstat/index.html#TARGET).
But is there a way to follow the release progress at the file level?

One last question:
Is there a difference in accuracy between the 4 daily updates?
I read in an old article (from 2012) that the 00z and the 12z were better than the other two.
I was surprised to read that and maybe it was the case back then but do you know if that is still the case?

Thanks

itblocksg

@matijamet said in A difference between reference time and update time:

@Korina ,

if you look on the "reference time" for NEMS, for example hier, https://www.windy.com/multimodel/44.107/15.326?44.108,14.248,8
you'll see for METEOBLUE (NEMS) :

Updated: 2020-02-27T07:27:42000Z
Reference time: 2020-02-25

Or, I don't understand something?

Thanks!

weird

idefix37

@itblocksg
This is not weird and @Korina explained the difference.
NEMS is a model provided by the Meteoblue company.
But this model is different from the model called here METEOBLUE (more exactly Meteoblue AI) in the point forecast:

Capture d’écran 2021-11-21 à 18.06.05.png

Additional information:
https://community.windy.com/topic/10316/metoblue-wildly-inaccurate-in-the-sw-usa/8?_=1637514782709

Korina

@itblocksg Hello, it would be nice if you could explain why it is weird.

Zancle10

Salve, se i tempi di riferimento ufficiali per il modello ECMWF sono 0.00 e 12.00, i due aggiornamenti in più con premium su cosa si basano? Grazie

Korina

@zancle10 Hello, this is an old article. You can find the update times at Model updates times.

ColtonAxel

I think Reference time is the time when a model starts a new forecast block. The process consists of the following: Initialisation - data are entered into the model, this creates Initial conditions (initialisation time) and then model starts calculating conditions.

youtubevanced55

@ColtonAxel thank you so much..

imran512

@Korina said in A difference between reference time and update time:

A difference between reference time and update time
Frequently Asked Questions ARTICLE 13 26 12.9k

Reply

Korina
21 Feb 2020, 18:28

Each model has a different update time and it could be confusing what connection does it have with the update time.

Here is the explanation:

Reference time - is the time when a model starts a new forecast block.
It consists of:

Initialisation - data are entered into the model, this creates Initial conditions (initialisation time) and then model starts calculating conditions.
Assimilation - normalisation on data and assembling data for computation.
Computation - calculating the future atmospheric rates of change in time increments.
Data extraction - extracting data into regular forecast intervals.
Data storage - data are stored and written in accessible format.
Data postprocessing - special postprocessing routines

For example model ECMWF has an update interval 12 hours (or 6 hours interval for Premium users), that means the first reference time will be 00:00:00Z (Zulu time) and second one 12:00:00Z.

Update time is time between the reference time and the actual update.
The whole process of data computation takes approximately from 8 to 9 hours.
So for ECMWF model the update times are 07:15:00Z and 19:15:00Z.
See the exact update times here - When and how often are the weather data updated?.

Where to find info about reference and update time:

Thanks for the clarification! The distinction between reference time and update time makes a lot more sense now. It’s interesting to see how much processing goes into generating accurate forecasts.