| Version 20 (modified by carlos, 6 years ago) (diff) |
|---|
Ensemble experiments
One of the advantages of using WRF4G is the possibility of defining several types of ensemble case studies in the same experiment to easily compare results between them. Each ensemble section will generate at least one independent realization depending always on the indicated date_time value. This is not only applicable to WRF namelist configurations, it can be perturbed other parameters available in the experiment configuration such as app or namelist_version. Thus, as a result, users can ran WRF simulations with different WRF model versions, execute the same WRF model compiled with different compilers, or check if the same simulation produces the same output on various resources. In order to better illustrate this WRF4G feature, several examples are provided:
- Example of WRF4G experiment configuration file to set two ensemble sections in which namelist parameters are going to be perturbed.
[DEFAULT] # Experiment configuration name = test # Simulation domain max_dom = 1 # Experiment time-specification # start_date | end_date | chunk_size_h date_time = 2011-08-28_12:00:00 | 2011-08-30_00:00:00 | 12 hours calendar = standard timestep_dxfactor = 6 # Running options np = 1 requirements = ARCH = "x86_64" clean_after_run = yes save_wps = no parallel_env = MPIRUN parallel_real = yes parallel_wrf = yes # Vtables must exist as Vtable.[input_extdata] extdata_vtable = GFS # Seconds between global analysis input times extdata_interval = 21600 preprocessor = default postprocessor = SFC domain_path = /home/user/wrf4g/repository/domains/Santander_50km extdata_path = /home/user/wrf4g/repository/input/NCEP/GFS output_path = /home/user/wrf4g/test/output app = wrf_all_in_one | bundle | /home/user/repository/apps/WRF/WRFbin-3.4.1_r2265_gfortran.tar.gz namelist_version = 3.4.1 [ensemble/phys1] namelist_values = spec_bdy_width | 10 spec_zone | 1 relax_zone | 9 feedback | 0 history_interval | 180 frames_per_outfile | 3 e_vert | 28 radt | 15 mp_physics | 5 cu_physics | 1 ra_lw_physics | 1 ra_sw_physics | 1 sf_sfclay_physics | 2 bl_pbl_physics | 2 sf_surface_physics | 2 [ensemble/phys2] namelist_values = spec_bdy_width | 10 spec_zone | 1 relax_zone | 9 feedback | 0 history_interval | 0 frames_per_outfile | 0 e_vert | 28 radt | 15 mp_physics | 4 cu_physics | 1 ra_lw_physics | 1 ra_sw_physics | 1 sf_sfclay_physics | 1 bl_pbl_physics | 1 sf_surface_physics | 2
- Example of WRF4G experiment configuration file to simulate the same WRF experiment using intel and gfortran compilers.
[DEFAULT] # Experiment configuration name = test # Simulation domain max_dom = 1 # Experiment time-specification # start_date | end_date | chunk_size_h date_time = 2011-08-28_12:00:00 | 2011-08-30_00:00:00 | 12 hours calendar = standard timestep_dxfactor = 6 # Running options np = 1 requirements = ARCH = "x86_64" clean_after_run = yes save_wps = no parallel_env = MPIRUN parallel_real = yes parallel_wrf = yes # Vtables must exist as Vtable.[input_extdata] extdata_vtable = GFS # Seconds between global analysis input times extdata_interval = 21600 preprocessor = default postprocessor = SFC domain_path = /home/user/wrf4g/repository/domains/Santander_50km extdata_path = /home/user/wrf4g/repository/input/NCEP/GFS output_path = /home/user/wrf4g/output # WRF-namelist parameters. Override namelist.input variables here namelist_version = 3.5.1 namelist_values = spec_bdy_width | 10 spec_zone | 1 relax_zone | 9 feedback | 0 history_interval | 180 frames_per_outfile | 3 e_vert | 28 radt | 15 mp_physics | 5 cu_physics | 1 ra_lw_physics | 1 ra_sw_physics | 1 sf_sfclay_physics | 2 bl_pbl_physics | 2 sf_surface_physics | 2 [ensemble/intel] app = wps | command | module load WPS/intel/3.5.1 wrf | command | module load WRF/intel/3.5.1 [ensemble/gfortran] app = wps | command | module load WRF/gfortran/3.5.1 wrf | command | module load WRF/gfortran/3.5.1
- Example of WRF4G experiment configuration file to execute the same simulation on two computing resources. This obviously involves different path for each resource.
[DEFAULT] # Experiment configuration name = test # Simulation domain max_dom = 1 # Experiment time-specification # start_date | end_date | chunk_size_h date_time = 2011-08-28_12:00:00 | 2011-08-30_00:00:00 | 12 hours calendar = standard timestep_dxfactor = 6 # Running options np = 1 clean_after_run = yes save_wps = no parallel_env = MPIRUN parallel_real = yes parallel_wrf = yes # Vtables must exist as Vtable.[input_extdata] extdata_vtable = GFS # Seconds between global analysis input times extdata_interval = 21600 preprocessor = default postprocessor = SFC app = wps | command | module load WRF/intel/3.5.1 wrf | command | module load WRF/intel/3.5.1 # WRF-namelist parameters. Override namelist.input variables here namelist_version = 3.5.1 namelist_values = spec_bdy_width | 10 spec_zone | 1 relax_zone | 9 feedback | 0 history_interval | 180 frames_per_outfile | 3 e_vert | 28 radt | 15 mp_physics | 5 cu_physics | 1 ra_lw_physics | 1 ra_sw_physics | 1 sf_sfclay_physics | 2 bl_pbl_physics | 2 sf_surface_physics | 2 [ensemble/resource_one_member] requirements = HOSTNAME = "resource_one" [ensemble/resource_two_member] requirements = HOSTNAME = "resource_two" [resource/resource_one] domain_path = /home/user/repository/domains/Santander_50km extdata_path = /home/user/repository/input/NCEP/GFS output_path = /home/user/test/output [resource/resource_two] domain_path = /home/user2/repository/domains/Santander_50km extdata_path = /home/user2/repository/input/NCEP/GFS output_path = /home/user2/test/output
