--analysis : what is the real effect of that option ?

All other topics about postprocessing model data (GrADS and other software), about other numerical weather prediction software (including WRF-NMM and WRF-ARW discussion unrelated to UEMS/WRF EMS), and general meteorology talk go in this forum.
Post Reply
mehdi
Posts: 7
Joined: Wed Apr 19, 2017 8:27 am

--analysis : what is the real effect of that option ?

Post by mehdi » Wed Oct 16, 2019 10:39 am

Dear all,
As every year or so, I'm getting back to UEMS to try to decipher the secrets of NWP. This time I'm pretty confident I will understand something as I have a robust setup based on a working GCP virtual machine.
I've tried to run a 15 days long downscaling test based on the cfsrv2 dateset. It fails miserably after 4 days of prediction. I figured out that there was a --analysis option for the ems_prep routine that might be more than useful. But as I managed to download the required files and as the simulation worked for a few days, I wonder what really happened and what is the real effect of the --analysis option.
On a pseudo-historical dataset like all the reanalysis data sets, what is different with and without --analysis ? Is it a matter of boundary conditions/time interpolation treatment ?
Further question : I read this topic viewtopic.php?t=170 and I wonder if there would be any difference for a given date YYYYMMHH starting the simulation 36 hours, 24 hours before, or even at this date YYYYMMHH itself ? If you want to downscale a reanalysis dataset with uems, is there any motivated guideline regarding the model warm up duration ?

Any hint or pointer would be appreciated, a lot. Thanks to all.

meteoadriatic
Posts: 1602
Joined: Wed Aug 19, 2009 10:05 am

Re: --analysis : what is the real effect of that option ?

Post by meteoadriatic » Wed Oct 16, 2019 6:43 pm

mehdi wrote:
Wed Oct 16, 2019 10:39 am
Dear all,
As every year or so, I'm getting back to UEMS to try to decipher the secrets of NWP. This time I'm pretty confident I will understand something as I have a robust setup based on a working GCP virtual machine.
I've tried to run a 15 days long downscaling test based on the cfsrv2 dateset. It fails miserably after 4 days of prediction. I figured out that there was a --analysis option for the ems_prep routine that might be more than useful. But as I managed to download the required files and as the simulation worked for a few days, I wonder what really happened and what is the real effect of the --analysis option.
On a pseudo-historical dataset like all the reanalysis data sets, what is different with and without --analysis ? Is it a matter of boundary conditions/time interpolation treatment ?
I believe that it tells download script to receive only analysis, not (re)forecast gribs.

mehdi wrote:
Wed Oct 16, 2019 10:39 am
Further question : I read this topic viewtopic.php?t=170 and I wonder if there would be any difference for a given date YYYYMMHH starting the simulation 36 hours, 24 hours before, or even at this date YYYYMMHH itself ? If you want to downscale a reanalysis dataset with uems, is there any motivated guideline regarding the model warm up duration ?
Generally it should be started at least 6 hours before your time of interest to allow enough spinup time. It also depends much on resolution. You need enough number of model timesteps to spin it up. So for very coarse resolutions like ~30km of dx/dy, you need maybe even 12 hours of spinup. However on fine grids like 1km, spinup (at least inside that nest) will happen much faster, probably within hour or two.

Post Reply