WRF installation

This guide explains how to install and set up the WRF weather model and its tools. It shows how to run the model automatically for operational forecasting. Before starting:

1. Check your computer's CPU count - WRF can run on one core but works better with multiple cores
2. Make sure you have enough disk space:
   • 100GB for geographical data
   • 100-200GB for model files (depends on your domain size)
   • For testing: 200-300GB is enough
   • For operational use: at least 1TB is recommended

For complete WRF documentation, visit the WRF User's Guide

Automated Installation

Follow these steps for easy installation:

1. Get the code:

   git clone [email protected]:fmidev/WRF_installation.git
   cd WRF_installation
   

2. Make the script runnable:

   chmod +x installation.sh
   

3. Run the installation (default location for the installation is /home/$USER/WRF_Model, but can be modified inside the installation script):

   ./installation.sh 
   

4. The script will:

   • Install all needed dependencies
   • Download and build required libraries
   • Compile WRF, WPS, WRFDA and UPP
   • Set up environment variables
   • Create folders for operational use
   • Set up a schedule template (crontab)
   • Download geographical data and CRTM coefficients

Manual Installation

The installation file contains step-by-step instructions for installing all libraries and compiling the WRF software and tools. When running WRF, you need specific environment variables set up by the env.sh script.

Domain Setup

You can easily create WRF domains using WRF Domain Wizard:

1. Click "New" in the sidebar and draw your main domain on the map. This will be your outer domain with the base resolution (default 12 km).
2. Click "Add Nest" to add an inner domain and adjust it on the map.
3. Set the inner domain resolution using "Parent_grid_ratio" - the default is 3, which means 4 km resolution if your outer domain is 12 km.
4. Click "Save" to download the domain settings (namelist.wps) for your WRF setup.

Work Flow

Environment Setup

The env.sh script in the WRF_Model/scripts directory sets up all necessary settings for WRF:

1. Main Settings:

   • Library Paths: Where to find NetCDF, HDF5, OpenMPI and other libraries
   • Program Locations: Where WRF, WPS and WRFDA programs are installed
   • Run Settings: How long to forecast, time between runs, and other basic options
   • Folders: Where to find and save data, observations, and results

2. On/Off Switches: The script has simple yes/no options to control which parts run:

   export RUN_CHECK_BOUNDARY_FILES=true  # Check if boundary files exist
   export RUN_GET_OBS=true               # Download observations
   export RUN_WPS=true                   # Run preprocessing
   export RUN_WRF=true                   # Run WRF model
   export WRFDA=false                    # Use data assimilation
   export RUN_UPP=false                  # Run post-processor
   export RUN_VERIFICATION=false         # Run verification tools
   

3. Using the Script: The scripts use this automatically, but you can also test/use it directly:

   source WRF_Model/scripts/env.sh
   

Boundary Data

WRF needs boundary data in GRIB format. Use the Download_GFS/get_gfs.sh script to download GFS data. The Download_GFS/gfs.cnf file sets the area, resolution, and forecast hours. Make sure the data goes to the correct folder (default: /home/{user}/WRF_Model/GFS). The script always downloads the newest data.

./get_gfs.sh

Preprocessing

Before running WPS (WRF Preprocessing System), you need to download geographical data (if installing manually). This data includes terrain height, land use types, and soil information:

# Download full dataset (~100GB)
wget -P /path/to/WPS_GEOG/ https://www2.mmm.ucar.edu/wrf/src/wps_files/geog_complete.tar.gz 
# Or download smaller high-resolution data (~30GB)
wget -P /path/to/WPS_GEOG/ https://www2.mmm.ucar.edu/wrf/src/wps_files/geog_high_res_mandatory.tar.gz 
# Extract the files
cd /path/to/WPS_GEOG/
tar -zxvf geog_complete.tar.gz --strip-components=1
tar -zxvf geog_high_res_mandatory.tar.gz --strip-components=1

Next, set up your domain in WRF_Model/scripts/Run_WPS.sh and WRF_Model/scripts/Run_WRF.sh:

1. Domain Settings: Copy settings from your WRF Domain Wizard file (namelist.wps):

   • Center point location (ref_lat, ref_lon)
   • Grid size (e_we, e_sn)
   • Grid spacing (dx, dy)
   • Nesting ratios (parent_grid_ratio)
   • Map type (map_proj)

2. File Locations: Check all folder paths (these should be correct if you used automated installation):

   • WPS program location
   • Geographical data location
   • Input/output folders

3. CPU Usage: Set the number of CPUs based on your computer:

   mpirun -np 24 ./geogrid.exe  # Uses 24 CPU cores
   

Test your WPS setup:

# Format: ./Run_WPS.sh <year> <month> <day> <hour> <forecast_hours> <output_folder>
# Example: 48-hour forecast starting September 10, 2024, 01:00 UTC
./Run_WPS.sh 2024 09 10 01 48 /home/username/WRF_Model/out/

If successful, this creates the input files needed for WRF.

Observations (Optional for Data Assimilation)

For data assimilation, you need observation data. The system uses NCEP's real-time database.

1. Data Types: The get_obs.sh script downloads satellite data and global observations from NCEP.

2. Limitations:

   • NCEP only keeps 2-3 days of recent data
   • Local data, radar data, and custom observations need special formatting (not included)

3. Getting Data: Run the script with date and time:

   # Format: ./get_obs.sh <year> <month> <day> <hour>
   # Example: Get data for September 10, 2024, 01:00 UTC
   ./get_obs.sh 2024 09 10 01
   

4. Result: The script organizes files for WRFDA to use.

Running the Model

The Run_WRF.sh script manages the WRF model run, including optional data assimilation.

1. Setup: Before running, check these settings:

   • Model Options: Make sure physics options, time steps, and domain settings match your needs. See the WRF User's Guide for all options.

   • Folder Paths: Check all paths (should be correct if using automated installation):

     • WRF program location
     • Input/output folders
     • Working folders

   • CPU Count: Set the number of processors for best performance

   • Data Assimilation: Check these settings if using WRFDA

2. Running the Model:

   # Format: ./Run_WRF.sh <year> <month> <day> <hour> <forecast_hours> <output_folder>
   # Example: 48-hour forecast starting September 10, 2024, 01:00 UTC
   ./Run_WRF.sh 2024 09 10 01 48 /home/username/WRF_Model/out/
   

3. Using Data Assimilation:

   • Set WRFDA=true in your env.sh file
   • Make sure CRTM files are available for satellite data
   • Provide proper background error statistics

Post-processing

UPP (Unified Post Processor) converts WRF output (NetCDF) to GRIB format. The installation file shows how to compile UPP. The setup_upp file explains how to set up UPP (not needed with automated installation). The Run_scripts/execute_upp.sh script converts NetCDF to GRIB automatically.

Verification

Instructions for using HARP verification with WRF will be added in the future.

Cleaning and Automation

The clean_wrf script removes old GFS, WRF, and UPP files. You can set this up to run automatically once a day.

For fully automated operation, use the control_run_WRF.sh script to run all steps. It's best to test each part separately first. Run it with:

./control_run_WRF.sh <analysis_hour>