UEMS wind parametrization

Questions and other topics related to UEMS 15.
skosecki
Posts: 20
Joined: Wed Aug 31, 2016 7:37 am

UEMS wind parametrization

Post by skosecki » Wed Aug 31, 2016 11:29 am

Hello,
Here is my first post in this forum, so I would like to say hi to everyone :)
I'm PHD student of physical oceanography. I deal with hydrodynamical modelling. My area of interest are West Spitsbergen fjords, especially Hornsund and Kongsfjord. I'm trying to make 2010-2015 hindcast and I get to the point where I don't have reliable wind field data. This is very tricky because of complex orografy in fjords and still unresolved.. Cause of my poor experience in atmospherics modelling I've decided to use UEMS, with I used before occasionally. I've created "parent domain" with is mapping West Spitsbergen and part of Greenland Sea. I've decided to make it with 4km horizontal resolution, thus the "child domains" for Hornsund and Kongsfjord could have 1.3km fine resolution. I'm running it on supercomputer so I have I lot of computation power (24 cores per node), but since my domain dimensions are 125x125, 55x40, 52x60 I'm limited to use 9 processor cores(decomp =0, numtiles =1, time_step = adaptive). I'm using CFSRV1 and CFSRV2 reanalysis data sets.

I know that it is very important to do proper calibration/parameterization. My question is how to perform such wind parametrization in UEMS.
First of all I would like to know how is it possible to run several simulations simultaneously? For now, 1 hour of computations give me 42 hour simulation, but I can use different nodes. My idea was to copy domain directory in /run. Localize it (ems_domain --localize) and run several cases of physics in parallel. I got data form one of meteorological stations in Hornsund. I had an idea to perform calibration for 2010 and then compute 2011 - 2015 with chosen parametrization.

Sorry for my poor english - I'm trying my best! :)
I'll appreciate every advice or suggestion.

alfe
Posts: 98
Joined: Thu Nov 25, 2010 8:13 pm

Re: UEMS wind parametrization

Post by alfe » Wed Aug 31, 2016 2:54 pm

Hi skosecki ,
What an interested topic !
As a first approach, you might be interested in reading this topic viewtopic.php?f=9&t=1061
What is the mesh size of CFSRV1 and CFSRV2 ?

skosecki
Posts: 20
Joined: Wed Aug 31, 2016 7:37 am

Re: UEMS wind parametrization

Post by skosecki » Thu Sep 01, 2016 8:03 am

Thanks alfe!
I've checked this topic before. I don't have such problem so far :) and I'm not so sure for now if the parameters that he chosen to change (RA_SW_PHYSICS, TOPO_WIND, CU_PHYSICS) are good to try with in my case. I haven't even compare my preliminary results with meteo station measurements, so far.
My problem is basically how to run few simulations simultaneously with different parameters and if one year of such "calibration time" is enough or maybe to much.
CFSRV1/CFSRV2 have 0.5deg resolution, which means that im my field I have aprox 12x56km (if I'm correct). Thankfully I don't have very large instabilities near boundaries.

alfe
Posts: 98
Joined: Thu Nov 25, 2010 8:13 pm

Re: UEMS wind parametrization

Post by alfe » Fri Sep 02, 2016 5:43 pm

I don't know much about parallel runs.
For your calibration, it depends on the needed accuracy. I think that 1 year is not bad because you will see the seasonal effects.
WRF gives some over estimations for 10m above ground wind speed. Maybe topowind=1 is a good idea for the finest domain. By the way what is the PBL used ? YSU is not bad. I use Morrison MP with RA_LW = 01 and RA_SW = 05. I have found that it gives quite accurate 2m temperature values.
Never use toposhading option because it is buggy.
Also the surface physics is important, especially because in your area there is a lot of snow. I use NOAH MP because it gives a good estimation of the snow cover.
You can also use sloperad=1 for the finest domain, because the terrain is very steep in your area !

There are many subtopics to talk about...

skosecki
Posts: 20
Joined: Wed Aug 31, 2016 7:37 am

Re: UEMS wind parametrization

Post by skosecki » Tue Sep 06, 2016 9:43 am

The first step of validation will be compare the wind roses from the Hornsund meteo station and model output. I hope that they will look similar :) That's the accuracy I need for now. Correlation of 0.85 will be good for me.

For my "basic"- version 0 run i use:

Code: Select all

     Model Physics

       Cumulus Scheme            :  None,None,None
       Shallow Cumulus Scheme    :  Off,Off,Off
       Microphysics Scheme       :  Thompson Aerosol-Aware,Thompson Aerosol-Aware ,Thompson Aerosol-Aware
       PBL Scheme                :  Yonsei University,Yonsei University,Yonsei University
       YSU Wind Correction       :  No Wind Correction,No Wind Correction,No Wind Correction
       Gravitational Settling    :  No Gravitational Settling ,No Gravitational Settling ,No Gravitational Settling
       Land Surface Scheme       :  Noah 4-Layer LSM,Noah 4-Layer LSM,Noah 4-Layer LSM
       Snow Cover Physics        :  New School,New School,New School
       Number Land Categories    :  21 (MODIS + Lakes),21 (MODIS + Lakes),21 (MODIS + Lakes)
       Number Soil Categories    :  16,16 ,16
       Number Soil Layers        :  4,4,4
       Surface Layer Physics     :  Monin-Obukhov,Monin-Obukhov ,Monin-Obukhov
       Sfc Heat & Moisture Flux  :  Model Computed u* and Sfc Fluxes Model Computed u* and Sfc Fluxes Model Computed u* and Sfc Fluxes
       Long Wave Radiation       :  RRTM ,RRTM ,RRTM
       Short Wave Radiation      :  Goddard (GFDL) Short Wave ,Goddard (GFDL) Short Wave,Goddard (GFDL) Short Wave
       SW Update Between Calls   :  Update SW Radiation,Update SW Radiation,Update SW Radiation
       Slope Radiation Effects   :  No Slope Effects,No Slope Effects,No Slope Effects
       Topography Shading        :  No Shading Effects,No Shading Effects,No Shading Effects
I run parallelly version 1 (TOPO_WIND=1, YSU_TOPDOWN_PBLMIX = 0) and version 2 (TOPO_WIND=0, YSU_TOPDOWN_PBLMIX = 1). But I - silly me - I didn't specify for each domain like TOPO_WIND = 0,1,1.
Since I run those 3 parallel simulations my performance decrees from 43 h/h to 33 h/h, despite that I used different processors.

Here is my 2 domain (hgt refined) - Hornsund
Image

BTW - alfe - do You use restarts in yours simulations?

j0nes2k
Posts: 56
Joined: Wed Feb 03, 2010 5:06 pm

Re: UEMS wind parametrization

Post by j0nes2k » Tue Sep 06, 2016 2:06 pm

Hello,

interesting topic - would be great if you could share your results which configuration works best for you and your area.

From my experience for domains over Europe there is no clear "winning" configuration pattern available - the best accuracy gains I got was from increasing the resolution (and number of CPUs used). Adaptive timestep should probably be used. I could not find huge differences in wind speed forecasts when using different configurations - the one with most impact seemed to be TOPO_WIND. However if you have the chance to increase the resolution this is probably a better way to go. For the verification part: the larger your area is and the more complex the terrain is, the more difficult the verification and tuning will be.

Regarding parallel runs, I simply ran different configurations after each other, but not in parallel. I copied the corresponding domain in the runs directory, made my changes and added cronjob entries. After all runs were finished I could examine and compare the results.


Best regards,

Jonas

alfe
Posts: 98
Joined: Thu Nov 25, 2010 8:13 pm

Re: UEMS wind parametrization

Post by alfe » Tue Sep 06, 2016 7:16 pm

BTW - alfe - do You use restarts in yours simulations?
No, I don't. It's like a rocket, once it starts I just let it go

skosecki
Posts: 20
Joined: Wed Aug 31, 2016 7:37 am

Re: UEMS wind parametrization

Post by skosecki » Wed Dec 21, 2016 12:04 pm

As I promised, I will like to present to you my preliminary results of wind/temperature parameterization for Svalbard - Hornsund fjord.

Here is a basic overview how my verison "zero" (v0) is set:

Code: Select all

                 *  This is a 1-way nested simulation

         *  Simple ocean mixed layer model turned ON


           EMS MODEL RUN SUMMARY FOR THE "west-spistsbergen" DOMAIN


                                     WRF ARW core             WRF REAL program
    ******************************************************************************

     System Name                                Scheduled Processors

       tryton-ap                 :          09                     09

     Total Processors            :          09                     09
     Domain Decomposition        :        Internal                 Internal


     Active Domains                  Domain 01                    Domain 02                    Domain 03
    ***************************************************************************************************************

     Domain & Run Information

       Domain Type               :  Limited Area                 Limited Area                 Limited Area
       Primary Time Step         :  Adaptive                     Adaptive                     Adaptive
       Step to Output Time       :  Yes                          Yes                          Yes
       Grid dimensions (NX x NY) :  125 x 125                    55 x 40                      52 x 61
       Vertical Layers (NZ)      :  45                           45                           45
       Grid Spacing              :  4.00 km                      1.33 km                      1.33 km
       Top of Model Atmosphere   :  50mb                         50mb                         50mb
       Parent Domain             :  NA                           Domain 01                    Domain 01
       Nesting Feedback          :  Feedback Off                 Feedback Off                 Feedback Off


     Timing Information

       Start Date                :  2010 Jan 01 18:00 UTC        2010 Jan 01 18:00 UTC        2010 Jan 01 18:00 UTC
       End Date                  :  2011 Mar 31 18:00 UTC        2011 Mar 31 18:00 UTC        2011 Mar 31 18:00 UTC
       Simulation Length         :  10896 Hours                  10896 Hours                  10896 Hours
       Boundry Update Freq       :  06 Hours


     Data Output Information

       File Output Freq          :  01 Hour                      01 Hour                      01 Hour
       Output File Format        :  netCDF                       netCDF                       netCDF
       Aux 1 File Output Freq    :  01 Hour                      01 Hour                      01 Hour
       Aux 1 File Format         :  netCDF                       netCDF                       netCDF
       Adjust Output Times       :  Yes                          Yes                          Yes


     Model Physics

       Cumulus Scheme            :  None                         None                         None
       Shallow Cumulus Scheme    :  Off                          Off                          Off
       Microphysics Scheme       :  Thompson Aerosol-Aware       Thompson Aerosol-Aware       Thompson Aerosol-Aware
       PBL Scheme                :  Yonsei University            Yonsei University            Yonsei University
       YSU Wind Correction       :  No Wind Correction           No Wind Correction           No Wind Correction
       Gravitational Settling    :  No Gravitational Settling    No Gravitational Settling    No Gravitational Settling
       Land Surface Scheme       :  Noah 4-Layer LSM             Noah 4-Layer LSM             Noah 4-Layer LSM
       Snow Cover Physics        :  New School                   New School                   New School
       Number Land Categories    :  21 (MODIS + Lakes)           21 (MODIS + Lakes)           21 (MODIS + Lakes)
       Number Soil Categories    :  16                           16                           16
       Number Soil Layers        :  4                            4                            4
       Surface Layer Physics     :  Monin-Obukhov                Monin-Obukhov                Monin-Obukhov
       Sfc Heat & Moisture Flux  :  Model Computed u* and Sfc Fluxes Model Computed u* and Sfc Fluxes Model Computed u* and Sfc Fluxes
       Long Wave Radiation       :  RRTM                         RRTM                         RRTM
       Short Wave Radiation      :  Goddard (GFDL) Short Wave    Goddard (GFDL) Short Wave    Goddard (GFDL) Short Wave
       SW Update Between Calls   :  Update SW Radiation          Update SW Radiation          Update SW Radiation
       Slope Radiation Effects   :  No Slope Effects             No Slope Effects             No Slope Effects
       Topography Shading        :  No Shading Effects           No Shading Effects           No Shading Effects


     ARW Core Model Dynamics

       Dynamics                  :  Non-Hydrostatic              Non-Hydrostatic              Non-Hydrostatic
       Time-Integration Scheme   :  Runge-Kutta 3rd Order        Runge-Kutta 3rd Order        Runge-Kutta 3rd Order
       Diffusion Sheme           :  Simple Diffusion             Simple Diffusion             Simple Diffusion
       6th-order Diffusion       :  No 6th-Order Diffusion       No 6th-Order Diffusion       No 6th-Order Diffusion
       Eddy Coefficient Scheme   :  2D 1st Order Closure         2D 1st Order Closure         2D 1st Order Closure
       Damping Option            :  No Damping                   No Damping                   No Damping
       W Damping                 :  W Damping On                 W Damping On                 W Damping On
       Horiz Momentum Advection  :  5th Order                    5th Order                    5th Order
       Horiz Scalar Advection    :  5th Order                    5th Order                    5th Order
       Vert Momentum Advection   :  3rd Order                    3rd Order                    3rd Order
       Vert Scalar Advection     :  3rd Order                    3rd Order                    3rd Order
       Sound Time Step Ratio     :  Automatic                    Automatic                    Automatic
       Moisture Advection Option :  Positive-Definite            Positive-Definite            Positive-Definite
       Scalar Advection Option   :  Positive-Definite            Positive-Definite            Positive-Definite
       Momentum Advection Option :  Positive-Definite            Positive-Definite            Positive-Definite
       TKE Advection Option      :  Positive-Definite            Positive-Definite            Positive-Definite

Parent domain:
Image

Hornsund domain with year-mean wind velocity vector (versions 0(CFSR) - white, version 3(ERA) -black):
Image

The table with the version setups:
Image

Air temperature on 2m, model vs measurements from the Polish Polar Station
Image

Wind speed on 10m, model vs measurements from the Polish Polar Station
Image

I'm not sure and I haven't decided yet how to set up the next simulations. I hope to run max 2 of them to get to a "proper" setup.. So maybe You could help me to decide? :)

skosecki
Posts: 20
Joined: Wed Aug 31, 2016 7:37 am

Re: UEMS wind parametrization

Post by skosecki » Thu Dec 22, 2016 2:15 pm

Is there a need to re run ems_prep when run_physics.conf has been modified ? :shock:

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

Re: UEMS wind parametrization

Post by meteoadriatic » Thu Dec 22, 2016 6:16 pm

skosecki wrote:Is there a need to re run ems_prep when run_physics.conf has been modified ? :shock:
No, settings there affect only WRF, not WPS part of run.

Thank you for the presented data!

Post Reply