WRF Tutorial

For Version 2.2 / Synoptic Lab

10/3/2007Robert Fovell

rfovell@ucla.edu

http://tinyurl.com/2uagrc orhttp://www.atmos.ucla.edu/~fovell/WRF/wrf_tutorial_2007.ppt.htm

Background on WRF model

• “Weather Research and Forecasting”• Co-developed by research and operational

communities– ARW core “Advanced Research WRF”– NMM core “Nonhydrostatic Mesoscale Model”

• Supercedes MM5 and Eta models• Current version 2.2• Platforms include Linux and Mac OS X

WRF advantages

• Better numerics than MM5– Arakawa C grid, R-K scheme, odd order advection

w/ implicit diffusion– Much less diffusive, larger effective resolution,

permits longer time steps

• Better handling of topography than Eta (original NAM)– NAM model is now WRF-NMM

• Fortran 95 (MM5 was F77)• NetCDF, GRIB1 and GRIB2

Further advantages

• MPI from the ground up• Allows real data and idealized simulations in

same framework• Plug-in architecture (different groups will

supply WRF “cores”)• Recently added: moving nests and nudging• NetCDF output - many great tools such as

NetCDF operators: http://nco.sourceforge.net/

WRF disadvantages

• Bleeding edge• Smaller range of physics choices (tho more

modern)• Software design is unintuitive for physical

scientists• Can take hours to compile

– But does not need frequent recompiling

• Comparatively slower than MM5• NetCDF files can be huge

WRF and related software

• WRF Preprocessing System (WPS)– Replaces/supercedes WRF SI

• WRF-ARW model– Single node and MPI

• WRF postprocessing software– RIP (read/interpolate/plot)– GrADS

• Specific to “hurricane” Synoptic Lab environment

• Neglecting for now: GRIB2, ARWpost

Web resources• WRF model users site

http://www.mmm.ucar.edu/wrf/users/user_main.html

• ARW users’ guidehttp://www.mmm.ucar.edu/wrf/users/docs/user_guide/contents.html

• WRF-ARW/WPS online tutorialhttp://www.mmm.ucar.edu/wrf/OnLineTutorial/index.htm

• WRF namelist descriptionhttp://www.mmm.ucar.edu/wrf/users/docs/user_guide/users_guide_chap5.html#Nml

• Tutorial presentationshttp://www.mmm.ucar.edu/wrf/users/tutorial/

tutorial_presentation.htm

My resources

• This presentation (PPT format)http://www.atmos.ucla.edu/~fovell/WRF/wrf_tutorial_2007.ppt

• WRF on Mac OS X http://www.atmos.ucla.edu/~fovell/WRF/WRF_ports.html

http://macwrf.blogspot.com

Setup on “hurricane” machines

Presumed:• tcsh environment

• Intel Fortran compiler (64-bit)• my environment setup employed

• precompiled versions of WRF, WPS, RIP and wrf_to_grads

Environment setup> …is the command line prompt

• If you don’t have a .cshrc file (worth saving) - recommended> cp /home/fovell/.cshrc .> source .cshrc

• If you want to keep your present .cshrc> cp /home/fovell/cshrc_fovell.csh .> ./cshrc_fovell.csh[you need to have the compiler environment set up

already]

# RGF additions [abridged]

setenv RIP_ROOT /home/fovell/RIP4setenv GADDIR /home/fovell/lib/gradssetenv GASCRP /home/fovell/gradslib

#alias lsm 'ls -alt | more'alias rm 'rm -i'alias cp 'cp -i'alias mv 'mv -i'alias trsl ' tail -f rsl.out.0000'alias mpirun 'nohup time /home/fovell/mpich-1.2.7p1/bin/mpirun'alias w2g '/home/fovell/WRF2GrADS/wrf_to_grads'

setenv P4_GLOBMEMSIZE 4096000setenv P4_SOCKBUFSIZE 65536unlimitlimit coredumpsize 0

This environment uses my versions ofnetcdf, mpich, grads, RIP

Set up a run directory

> cd

> mkdir FELIX

> cd FELIX

> cp /home/fovell/WRFtutorial/make_all_links.csh .

> make_all_links.csh

> cp /home/fovell/WRFtutorial/namelist.* .[copies namelist.input, namelist.wps]

WRF for real-data run

Hurricane Felix (2007)

[This example uses data thatwill not remain online]

WPS overview

• Tasks– (1) set up a domain (can be reused)

• geogrid.exe

– (2) unpack parent model data (e.g., from GFS, NAM, etc.)• ungrib.exe

– (3) prepare unpacked data for WRF• metgrid.exe

• Controlled by namelist.wps

namelist.wps

&share wrf_core = 'ARW', max_dom = 1, start_date = '2007-09-02_00:00:00','2007-09-02_00:00:00', end_date = '2007-09-03_12:00:00','2007-09-03_12:00:00', interval_seconds = 10800, io_form_geogrid = 2,/

• For start_date, end_date need one column for each domain• interval_seconds is parent model data frequency (here, 3 h)

namelist.wps (cont.) &geogrid parent_id = 1, 1, parent_grid_ratio = 1, 3, i_parent_start = 1, 53, j_parent_start = 1, 65, e_we = 70, 259, e_sn = 40, 199 geog_data_res = '2m','2m', dx = 36000, dy = 36000, map_proj = 'lambert', ref_lat = 15.0 ref_lon = -75.0, truelat1 = 29.6, truelat2 = 29.6, stand_lon = -75.0, geog_data_path = '/home/fovell/WPS_GEOG/geog'/

there is more…

geogrid - create domain

> geogrid.exe

* creates geo_em.d01.nc (a NetCDF file)* look for “Successful completion of geogrid.”

> plotgrids.exe

* creates gmeta

> idt gmeta

* uses NCAR graphics tool to view domain

‘gmeta’ file

> ncdump geo_em.d01.nc | more

netcdf geo_em.d01 {dimensions: Time = UNLIMITED ; // (1 currently) DateStrLen = 19 ; west_east = 69 ; south_north = 39 ; south_north_stag = 40 ; west_east_stag = 70 ; land_cat = 24 ; soil_cat = 16 ; month = 12 ;variables: char Times(Time, DateStrLen) ; float XLAT_M(Time, south_north, west_east) ; XLAT_M:FieldType = 104 ; XLAT_M:MemoryOrder = "XY " ; XLAT_M:units = "degrees latitude" ; XLAT_M:description = "Latitude on mass grid" ; XLAT_M:stagger = "M" ;

Parent model data issues

• Sources include GFS, NAM, NARR reanalysis data, etc.

• Need a different Vtable (variable table) for each source– e.g., Vtable.GFS, Vtable.AWIP (NAM), Vtable.NARR, etc.

– Look in /home/fovell/WRFtutorial

Accessing parent model data> link_grib.csh /home/fovell/2007090200/gfs

* links to where parent model for case resides** data files start with ‘gfs*’

> ln -sf /home/fovell/WRFtutorial/Vtable.GFS Vtable

* specifies appropriate Vtable

> ungrib.exe

* extracts parent model data* look for “Successful completion of ungrib.”

Next step: metgrid> metgrid.exe...hopefully you see ...

“Successful completion of metgrid.”

...Output looks like...

met_em.d01.2007-09-02_00:00:00.nc met_em.d01.2007-09-02_21:00:00.ncmet_em.d01.2007-09-02_03:00:00.nc met_em.d01.2007-09-03_00:00:00.ncmet_em.d01.2007-09-02_06:00:00.nc met_em.d01.2007-09-03_03:00:00.ncmet_em.d01.2007-09-02_09:00:00.nc met_em.d01.2007-09-03_06:00:00.ncmet_em.d01.2007-09-02_12:00:00.nc met_em.d01.2007-09-03_09:00:00.ncmet_em.d01.2007-09-02_15:00:00.nc met_em.d01.2007-09-03_12:00:00.ncmet_em.d01.2007-09-02_18:00:00.nc

ncdump on a metgrid filenetcdf met_em.d01.2007-09-02_00:00:00 {dimensions: Time = UNLIMITED ; // (1 currently) DateStrLen = 19 ; west_east = 69 ; south_north = 39 ; num_metgrid_levels = 27 ; num_sm_levels = 4 ; num_st_levels = 4 ; south_north_stag = 40 ; west_east_stag = 70 ; z-dimension0012 = 12 ; z-dimension0016 = 16 ; z-dimension0024 = 24 ;

This data source has 27 vertical levels.This will vary with source.

WRF model steps

• Tasks– Run real.exe (to finish creation of WRF

model input data)– Run wrf.exe

• Both use namelist.input– Configured separately from namelist.wps

but includes overlapping information

namelist.input &time_control run_days = 0, run_hours = 36, run_minutes = 0, run_seconds = 0, start_year = 2007 , 2007 , start_month = 09 , 09 , start_day = 02 , 02 , start_hour = 00 , 00 , start_minute = 00, 00, start_second = 00, 00, end_year = 2007 , 2007 , end_month = 09 , 09 , end_day = 03 , 03 , end_hour = 12 , 12 , end_minute = 00, 00, end_second = 00, 00,

For start_*, end_*, one column per domain

namelist.input (cont.)

interval_seconds = 10800 input_from_file = .true., .true., history_interval = 60, 60, frames_per_outfile = 6, 6, restart = .false., restart_interval = 5000,

interval_seconds matches namelist.wpsinput_from_file should normally be ‘true’ for each domainhistory_interval - how frequently (in min) output createdframes_per_outfile - number of writes in each history file

If wish to restart mode, restart = .true.(and set model start_* data to restart time)

restart_interval = frequency (min) for writing restart files

namelist.input (cont.) &domains time_step = 150, time_step_fract_num = 0, time_step_fract_den = 1, max_dom = 1, s_we = 1, 1, 1, e_we = 70, 259, 94, s_sn = 1, 1, 1, e_sn = 40, 199, 91, s_vert = 1, 1, 1, e_vert = 31, 31, 31, num_metgrid_levels = 27 dx = 36000, 12000, 333, dy = 36000, 12000, 333, grid_id = 1, 2, 3, parent_id = 0, 1, 2, i_parent_start = 0, 53, 30, j_parent_start = 0, 65, 30, parent_grid_ratio = 1, 3, 3, parent_time_step_ratio = 1, 3, 3,

namelist.input (cont.) &physics mp_physics [Microphysics] = 1 , 1 , ra_lw_physics [Longwave rad] = 1 , 1 , ra_sw_physics [Shortwave rad] = 1 , 1 , radt [Radiation time step; min] = 10 , 10 , sf_sfclay_physics [Surface layer] = 1 , 1 , sf_surface_physics [Surface] = 1 , 1 , bl_pbl_physics [Boundary layer] = 1 , 1 , bldt [Boundary layer time step; min]= 0, 0, cu_physics [cumulus scheme] = 1 , 0 , cudt [cumulus time step; min] = 5 , isfflx = 1, ifsnow = 0, icloud = 1, surface_input_source = 1, num_soil_layers = 5 , mp_zero_out = 0 ,

Notes on physics

• Need to use SAME microphysics (mp) scheme in each domain, but can use different cumulus (cu) schemes

• Some physics combinations work better than others, some don’t work at all -- this is only lightly documented

• bldt = 0 means boundary layer scheme is called every time step

namelist.input (cont.) &dynamics w_damping = 0, diff_opt [subgrid turbulence] = 1, km_opt [ “ ] = 4, diff_6th_opt [numerical smoothing] = 0, diff_6th_factor [ “ ] = 0.12, base_temp = 290. damp_opt = 0, zdamp = 5000., 5000., 5000., dampcoef = 0.01, 0.01, 0.01 khdif = 0, 0, 0, kvdif = 0, 0, 0,

Only some diff_opt/km_opt combinations make sense,and choices are resolution-dependent.

More info:http://www.mmm.ucar.edu/wrf/users/tutorial/tutorial_presentation.htm

http://www.mmm.ucar.edu/wrf/users/tutorial/200707/WRF_Physics_Dudhia.pdf

• Has changed a lot since version 2.1.2• Number of vertical model levels now specified

w/ real.exe

• The num_metgrid_levels comes from parent model; you set e_vert (number of WRF levels) here– Can reset WRF levels by rerunning real.exe– Can also specify which levels you want

real.exe

e_vert = 31, 31, 31, num_metgrid_levels = 27

Setting levels in namelist.input(optional)

p_top_requested = 5000 eta_levels = 1.00,0.9969,0.9935,0.9899,0.9861,0.9821,0.9777,0.9731,0.9682,0.9629,0.9573,0.9513,0.9450,0.9382,0.9312,0.9240,0.9165,0.9088,0.9008,0.8925,0.8840,0.8752,0.8661,0.8567,0.8471,0.8371,0.8261,0.8141,0.8008,0.7863,0.7704,0.7531,0.7341,0.7135,0.6911,0.6668,0.6406,0.6123,0.5806,0.5452,0.5060,0.4630,0.4161,0.3656,0.3119,0.2558,0.1982,0.1339,0.0804,0.0362,0.0000,

• WRF uses “sigma” or “eta” coordinates (1.0 is model bottom, 0.0 is top)• Added lines to &domains in namelist.input, presuming e_vert = 51, requests a model top pressure of 50 mb (5000 Pa) and concentrates vertical resolution in lower trop

Run real.exe> mpirun -np 2 real.exewrf@iniki.atmos.ucla.edu's password: starting wrf task 0 of 2 starting wrf task 1 of 22.624u 1.248s 0:12.63 30.5% 0+0k 0+0io 0pf+0w

> tail rsl.out.0000 --> extrapolating TEMPERATURE near sfc: i,j,psfc, p target d01 2007-09-03_12:00:00 forcing artificial silty clay loam LAND CHANGE = 0 WATER CHANGE = 0 d01 2007-09-03_12:00:00 Timing for processing 0 s. LBC valid between these times 2007-09-03_09:00:00.0000 2007-09-03_12:00:00 d01 2007-09-03_12:00:00 Timing for output 0 s. d01 2007-09-03_12:00:00 Timing for loop # 13 = 0 s. d01 2007-09-03_12:00:00 real_em: SUCCESS COMPLETE REAL_EM INIT

Aside: password-less execution

• Last slide’s mpirun command asked for 2 cpus (-np 2)

• By default, 2 cpus on same workstation are accessed

• To avoid being asked for password:> cd ~/.ssh> ssh-keygen -t dsa [then hit return 4 times]Your public key has been saved in /home/wrf/.ssh/id_dsa.pub.The key fingerprint is:cc:78:50:1e:77:23:ca:8f:81:3d:f0:d2:a4:8a:2e:a7 wrf@iniki.atmos.ucla.edu> cp id_dsa.pub authorized_keys [if does not already exist]> cd ../FELIX

Run wrf.exe

• Output of real.exe is wrfbdy_d01 and wrfinput_d01 (NetCDF files)– Additional wrfinput files created for nests if max_dom > 1

• Run the model> mpirun -np 4 wrf.exe &

• Creates wrfout_d01* files keyed by simulation date, and rsl.out/rsl.error files for each CPU requested

• Namelist set up to do 36 h run• Look for at end of rsl.out.0000 file:

d01 2007-09-03_12:00:00 wrf: SUCCESS COMPLETE WRF• Output files created:

• This is because history_interval was 60 min and frames_per_outfile was 6

FELIX output

wrfout_d01_2007-09-02_00:00:00 wrfout_d01_2007-09-03_00:00:00wrfout_d01_2007-09-02_06:00:00 wrfout_d01_2007-09-03_06:00:00wrfout_d01_2007-09-02_12:00:00 wrfout_d01_2007-09-03_12:00:00wrfout_d01_2007-09-02_18:00:00

Postprocessing WRF output:RIP and GrADS

(Vis5D and ARWpost also exist)

RIP

• RIP operates in batch mode, using input scripts

• RIP can overlay fields, do arbitrary cross-sections, calculate trajectories, and create Vis5D output files

• RIP tasks include– Unpack model output data (ripdp_wrf)– Create RIP plotting scripts (rip.in files)– Execute scripts (rip)

• RIP can create a LOT of output files

RIP procedure

> ripdp_wrf run1 all wrfout_d01*

[this creates a new dataset called ‘run1’and uses all wrfout_d01 files created]

> rip run1 rip.T2.in

[the rip.T2.in file is a script containingRIP plotting commands]

[the output file, rip.T2.cgm, is a graphicsmetafile]

> You can view the cgm file using idt or ictrans

36 h forecast(2 m T - color; SLP - contour; 10 m winds - vector)

RIP script

===========================================================================feld=T2; ptyp=hc; vcor=s; levs=1fb; cint=0.5; cmth=fill;> arng; cbeg=283; cend=309; cosq=0,violet,12.5,blue,25,green,37.5,>light.green,50,white,62.5,yellow,75,orange,87.5,red,100,brownfeld=U10,V10; ptyp=hv; vcmx=20.0; colr=black; linw=1; intv=2;feld=slp; ptyp=hc; vcor=s; levs=1fb; cint=4; nohl;colr=blue;linw=2;nolbfeld=map; ptyp=hb; colr=dark.blue; linw=2;feld=tic; ptyp=hb===========================================================================

http://www.mmm.ucar.edu/mm5/documents/ripug_V4.html

GrADS and wrf_to_grads

• GrADS produces beautiful graphics• Batch scriptable AND interactive• Interactive: good for overlaying different

datasets, computing difference fields [can also be done in RIP]

• Doesn’t create huge numbers of intermediate files like RIP

• Arbitrary cross-sections are very difficult to construct

GrADS procedure

• Copy control_file from /home/fovell/WRFtutorial and edit

• Select variables desired and define wrfout files to be accessed (next slide)

• w2g control_file run1g

• Creates run1g.ctl, run1g.dat

http://grads.iges.org/grads/head.html

control_file

-3 ! times to put in GrADS file, negative ignores this0001-01-01_00:00:000001-01-01_00:05:000001-01-01_00:10:00end_of_time_list ! 3D variable list for GrADS file ! indent one space to skipU ! U Compoment of windV ! V Component of wind UMET ! U Compoment of wind - rotated (diagnostic) VMET ! V Component of wind - rotated (diagnostic)W ! W Component of windTHETA ! Theta TK ! Temperature in KTC ! Temperature in C

List of available 2D fields follows

control_file (cont.) ! All list of files to read here ! Indent not to read ! Full path OKwrfout_d01_2007-09-02_00:00:00wrfout_d01_2007-09-02_06:00:00wrfout_d01_2007-09-02_12:00:00wrfout_d01_2007-09-02_18:00:00wrfout_d01_2007-09-03_00:00:00wrfout_d01_2007-09-03_06:00:00wrfout_d01_2007-09-03_12:00:00end_of_file_list

! Now we check to see what to do with the datareal ! real (input/output) / ideal / static1 ! 0=no map background in grads, 1=map background in grads-1 ! specify grads vertical grid ! 0=cartesian, ! -1=interp to z from lowest h ! 1 list levels (either height in km, or pressure in mb)1000.0950.0900.0850.0800.0750.0

Running GrADS

> gradsnc -l

[GrADS graphics output window opens]

ga-> open run1g

[ga-> is GrADS environment prompt]

ga-> /home/fovell/WRFtutorial/T2_movie.gs

[executes this GrADS script; hit return to advance a frame]

ga-> quit

[to exit]

36 h forecast(2 m T and 10 m winds)

= end =