results deviate from NCEP data

I am computing the ufs-mrweather-app locally and in doing so I did compare my own simulations with the original forecast of the NCEPs. I computed
different versions of the app (v15p2, v16beta), different resolutions (1 degree, 0.5 degree, 0.25 degree) and different initialisations (GRIB, nc).

So far my simulations result in deviations from the NCSP forecasts with a significant increase in deviations after 10 days. Those discrepancies
can be for example as large as T2m of 6 degrees.

My questions are:
Is it a given that ufs-mrweather-app simulations do deviate intensely from NCEP data or is this a sure sign that my computations do contain an error?
Do you know of analysis that were done of the precision of the different models with regards to the NCEP model?
The single most confusing issue to me is that my simulations with v16beta do deliver bigger discrepancies than the simulations with v15p2.

Any input on your end is much appreciated.

Best regards 

Attached you will find a sample:

base: 24.8.2021
grid: 12° E, 51° N
red line: ncep gfs, 0.25x0.25 degree
green line: ufs-mrweather-app v15p2, initialisation nc-files, 0.5x0.5 degree

Hi, Ulrich,

Do you have surface flux plots (e.g., latent heat flux, sensible heat flux, shortwave radiation, longwave radiation, soil heat flux, etc.)? The surface temperature (T2) seems drift away after 10 days and the surface energy budget can give a hint on the issue.



Hi, Ulrich,

It seems have big difference in the downward short wave radiation. You may check the 2d cloud (upper, middle, low, and total cloud) plots over that area. 



Hi Linlin,


thanks for the answer. I attached the 2d cloud (upper, middle, low, and total cloud) plots. Because of the effect that the simulation seems to get coarser after T240 and deviates more from the NCEP data, I am still looking for a mistake I am making.

Is it possible to find a data set of a simulation from v15p2 or v16beta that was calculated up to T384 (any base) to compare this data with my data.

I have only been able to find plots but no data. I have also attached the two files model_configure and atm_in with my settings. fhmax is set to 384.0.

Otherwise I had not made any changes.

Thanks Ulrich



-- model_configure ----

ENS_SPS: .false.
PE_MEMBER01: 120
app_domain: global
atmos_nthreads: 1
calendar: julian
cpl: .false.
debug_affinity: .false.
dt_atmos: 300
filename_base: atm sfc
iau_offset: 0
ideflate: 1
imo: 1536
jmo: 768
memuse_verbose: .false.
nbits: 14
ncores_per_node: 8
nfhmax_hf: 12
nfhout: 3
nfhout_hf: 1
nhours_fcst: 732
nsout: -1
num_files: 2
output_1st_tstep_rst: .false.
output_file: netcdf
output_grid: gaussian_grid
output_history: .true.
print_esmf: .false.
quilting: .true.
restart_interval: 6
start_day: 24
start_hour: 0
start_minute: 0
start_month: 8
start_second: 0
start_year: 2021
total_member: 1
use_hyper_thread: .false.
write_dopost: .false.
write_fsyncflag: .true.
write_groups: 1
write_nemsioflip: .true.
write_tasks_per_group: 24


Hi, Ulrich,

During the UFS MRW tutorial, there are some examples runs.

On the supported machines (e.g., Hera, Jet, Cheyenne, ...), there are some example cases data available. You can also found some data according to the instructions in MRW App users' guide:

MRW App workflow will automatic generate configure the needed files. So you can try to use that to get the configure files.