Introduction
With the goal of supporting EMC in selecting an advanced physics suite for the GFS v16, due to be implemented operationally in 2021, the GMTB tested four configurations of NOAA’s Unified Forecast System (UFS). The runs were started on 04 December 2018 and were completed on 09 February 2019. This test was conducted collaboratively between the various groups as described below:
● Global Model Test Bed (GMTB): Conduct model runs and produce verification and diagnostics to complement EMC's evaluations
● NOAA Environmental Modeling Center (EMC): Provide three model configurations (one baseline and two advanced physics configurations) and the workflow, prepare the test plan, and conduct verification and diagnostics
● NOAA Earth System Research Laboratory (ESRL) Global Systems Division (GSD): Provide one advanced physics configuration
● Navy Research Laboratory (NRL), NOAA ESRL Physical Sciences Division (PSD), and NCAR: Participate in an independent expert panel to evaluate test results
Model Configuration
The experiment consisted of four configurations of the GFS listed in the following table, all of which employed the Finite-Volume Cubed-Sphere dynamical core (FV3). Suite 1 was consistent with the model configuration used in the retrospective tests conducted by EMC to inform the GFS v15 operational implementation. Suite 2 was the same as Suite 1 except it employed a different boundary layer scheme. Suite 3 used the components that were developed by Climate Process Team (CPT) across multiple research centers and universities, including Colorado State University, University of Utah, NASA, NCAR, and EMC. Its individual parameterizations have been applied primarily to medium and long range predictions. Suite 4 was derived from the operational Rapid Refresh and High-Resolution Rapid Refresh (RAP/HRRR) modeling systems. It was assembled and developed by GSD from years of community contributions through the WRF community modeling system for mesoscale applications. Both Suites 3 and 4 were aerosol aware (AA), while Suite 3 used a constant and horizontally homogeneous aerosol distribution, and Suite 4 used a time-evolving aerosol distribution that was initialized from monthly climatology. Note that Suite 1 was exclusively used for atmospheric radiation, gravity wave drag, ozone, and stratospheric water photochemistry processes. Suite 4 was the only configuration run using the Common Community Physics Package (CCPP), which the other suites did not employ.
Initial Conditions
Model integrations used atmospheric initial conditions (ICs) based on the ECMWF model to eliminate model-climate biases that might be introduced if ICs from the Global Data Assimilation System (GDAS), which uses GFS physics, were applied. ECMWF ICs were obtained with the help of NOAA Physical Sciences Division and were interpolated to the FV3GFS grid by EMC prior to the start of the test.
Soil ICs were taken from the retrospective runs of FV3GFS done during the testing geared to the GFS v15 implementation. The initialization for the RUC LSM was done internally in the FV3GFS model by interpolating the information from the levels supplied (the four levels of Noah LSM) to the RUC levels.
Aerosol ICs were suite-specific: Suite 3 was initialized with a horizontally-homogeneous and time-invariant distribution set in the model code itself, while Suite 4 was initialized from a space- and time-dependent climatology of water- and ice-friendly aerosols (Thompson and Eidhammer, 2014). The GSD group provided the aerosol climatology file QNWFA_QNIFA_SIGMA_MONTHLY.dat.nc, which is also distributed with the Weather Research and Forecast model. GSD also provided a conversion utility to interpolate the aerosol climatology horizontally and vertically to the FV3GFS grid.
Before the start of the test, EMC created one IC file for each case, containing the atmospheric, soil, and aerosol fields. All suites were initialized from the same files, with aerosol information ignored by Suites 1, 2, and 3.
Cases Run
Runs were initialized every five days between 1 January 2016 and 31 December 2017, alternating between 00 and 12 UTC. In addition, 16 cases studies were added under the recommendation of the EMC Model Evaluation Group. The list of cases and a brief description can be found below.
Tropical Cyclone (TC) cases:
● 10/01/15 00z TC Joaquin and flooding in SC
● 10/02/16 00z TC Matthew
● 08/26/17 00z TC Harvey
● 09/07/17 00z TC Irma
● 10/04/17 00z TC Nate
● 08/19/18 00z TC Lane
● 09/11/18 12z TC Florence
● 07/31/17 00z TC Noru
Other cases:
● 01/18/16 12z Blizzard of 2016 - progressive
● 04/22/16 00z Plains severe weather - progressive, also a chance to examine drylines
● 03/10/17 00z "Pi Day" Blizzard - Precipitation type
● 04/20/17 00Z Flooding in the Mississippi Valley
● 07/29/17 00z Too hot in FV3GFS in CA
● 10/16/17 12z Inversions and 2-m temperature
● 01/01/18 00z "Bomb" cyclone
● 03/15/17 00z Atmosphere river - progressive
Workflow
A simplified workflow was supplied by EMC to execute the runs. The workflow consisted of four distinct tasks:
1) establishing the directory structure and submission scripts for each case,
2) running the model,
3) running the Unified Post-Processor and archiving the output, and
4) creating sounding output in BUFR format and archiving the output
In order to more efficiently run all of the cases, each suite was split into eight streams such that eight cases could be run concurrently for each suite. The workflow consisted of a set of bash scripts, with one script submitting the next. Once a case had been successfully run through all components of the workflow, the next date was submitted to be executed.
The workflow code is version controlled in a private GitHub repository. Please contact gmtb-help@ucar.edu for additional information.
Codes
The executables were provided by GSD (Suite 4 NEMSfv3gfs) and EMC (Suites 1-2-3 NEMSfv3gfs, postprocessor, and BUFR soundings). They have provided GMTB with the information below regarding the source code and compilation.
Model
Suite 1:
VLab repository: NEMSfv3gfs; tag nemsfv3gfs_beta_v1.0.12 (implementation version) and submodules therein.
To compile:
compile.sh directory jet " HYDRO=N 32BIT=Y" 1
Suites 2, 3:
VLab repository: NEMSfv3gfs; tag NEMSfv3gfs_suite23
VLab repository: FV3; tag NEMSfv3gfs_suite23
VLab repository: NEMS: tag NEMSfv3gfs_suite23
To compile:
compile.sh directory jet " HYDRO=N 32BIT=Y" 1
Suite 4:
Code for Suite 4, including NEMS, FV3, and CCPP-related materials are located in a private GitHub repository. Please contact gmtb-help@ucar.edu for additional information.
To compile:
./compile.sh directory 'OPENMP=N INTEL18=Y 32BIT=Y SION=Y CCPP=Y HYBRID=N STATIC=Y SUITE=suite_FV3_GSD.xml'
Unified post-processor
VLab repository: EMC_post; tag ncep_post.v8.0.27
Two files were changed to be able to post-process MG3: GFIP3.f and INITPOST_GFS_NEMS_MPIIO.f. The differences are recorded in Appendix B.
Aerosol tools and climatology
https://github.com/NCAR/aeroconv; hash #6d3aa
Archives
The following archives have been created on the NOAA HPSS:
Initial conditions: /BMC/gmtb/5year/Phys_Test_FV3GFSv2/FV3_IFC_GMTB
Post-processed files: /BMC/gmtb/5year/Phys_Test_FV3GFSv2/POST_ARCH
Soundings in BUFR format: /BMC/gmtb/5year/Phys_Test_FV3GFSv2/BUFR_ARCH
