Scheduled Downtime
On Friday 21 April 2023 @ 5pm MT, this website will be down for maintenance and expected to return online the morning of 24 April 2023 at the latest

Gfs analysis files

This post was from a previous version of the WRF&MPAS-A Support Forum. New replies have been disabled and if you have follow up questions related to this post, then please start a new thread from the forum home page.


I want to do a test by comparing wrf output files with analysis files from GFS. I found .sfcanl extension files that have my variables of interest (temperature and humidity at 2m) but I can’t find how are this files created and what is it’s accuracy.
Are they appropiate to validate wrf output? Is there any other dataset you could recommend me (analysis or reanalysis, not sure about the difference) to use?
We also compare WRF with GFS. GFS analysis data is kind of reliable because it assimilates huge amounts of observations.
However, you may need to pay attention to difference in resolution between WRF and GFS.

As for other surface data, we also use STAGE IV and GDAS in-situ observations. Note that StageIV is only for CONUS, while GDAS is global. You need to choose datasets based on your study area.
Thanks Ming Cheng,
As for what you said about the resolution, I'm aware of this and use a regridding function over the global GFS sfcanl files to match WRF resolution. I'm using Pearson coefficient to numerically evaluate the accuracy of the simulation. I found that this coefficient suddently goes down (around 0.5) at 18h while being aroung 0.9 for the other cycles (0h, 06h, 12h). Is it because 18 cycle files are more accurate to real data? Does it make sense to you?
I am not 100% sure what could be the possible reason that the result at 18hr forecast is worse than in other time. One possible reason is that the model cannot well reproduce the situation at the transitional time from day to night. PBL scheme performs better in the daytime when the boundary is unstable. However, in the nighttime when the PBL becomes stale, the PBL scheme is less capable. if 18hr is the local sunset time, then poor PBL simulation might be a reason.
Thanks for your reply.
I saw a pattern in the decay of the Pearson coefficient for all the configurations submitted to test but Conus (with MYJ as PBL scheme) performed better in general for T2 than YSU scheme (I also checked Q2 and the difference wasn't that big between PBL schemes).
I'm working in a domain centred in Panama, so 18h UTC really is midday in that country (local time = UTC - 5), so I don't know it could be related to sun hours...