2.5. Tutorials

This chapter describes tutorial data that originally came from an SRW Tutorial. It should be updated in the future.

2.5.1. Sample Forecast #4: Halloween Storm

Objectives:

• Compare forecast outputs for similar experiments that use different IC/LBC sources.

• Use verification tools to assess forecast quality.

2.5.1.1. Weather Summary

A line of severe storms brought strong winds, flash flooding, and tornadoes to the eastern half of the US.

Weather Phenomena: Flooding and high winds

• Storm Prediction Center (SPC) Storm Report for 20191031

Fig. 2.5 Halloween Storm 2019

2.5.1.2. Data

Data for the Halloween Storm is publicly available in S3 data buckets. The Rapid Refresh (RAP) data can be downloaded from the SRW App data bucket using wget. Make sure to issue the command from the folder where you want to place the data.

wget https://noaa-ufs-srw-pds.s3.amazonaws.com/develop-20240618/halloween_rap.tgz
tar -xzf halloween_rap.tgz

This will untar the halloween_rap.tgz data into a directory named RAP.

The SRW App can pull HRRR data directly from the HRRR data bucket. Users do not need to download the data separately.

2.5.1.3. Load the workflow

To load the workflow environment, source the lmod-setup file and load the workflow conda environment by running:

source /path/to/dtc-vx-workflow/setup_conda.sh

After loading the workflow, users should follow the instructions printed to the console. Usually, the instructions will tell the user to run conda activate srw_app. For example, a user on Hera with permissions on the nems project may issue the following commands to load the workflow (replacing User.Name with their actual username):

source /scratch1/NCEPDEV/nems/User.Name/ufs-srweather-app/etc/lmod-setup.sh hera
module use /scratch1/NCEPDEV/nems/User.Name/ufs-srweather-app/modulefiles
module load wflow_hera
conda activate srw_app

2.5.1.4. Configuration

Navigate to the ufs-srweather-app/ush directory. The default (or “control”) configuration for this experiment is based on the config.community.yaml file in that directory. Users may copy this file into config.yaml if they have not already done so:

cd /path/to/ufs-srweather-app/ush
cp config.community.yaml config.yaml

Users can save the location of the ush directory in an environment variable ($USH). This makes it easier to navigate between directories later. For example:

export USH=/path/to/ufs-srweather-app/ush

Users should substitute /path/to/ufs-srweather-app/ush with the actual path on their system. As long as a user remains logged into their system, they can run cd $USH, and it will take them to the ush directory. The variable will need to be reset for each login session.

2.5.1.4.1. Experiment 1: RAP Data

Edit the configuration file (config.yaml) to include the variables and values in the sample configuration excerpts below.

Hint

To open the configuration file in the command line, users may run the command:

vi config.yaml

To modify the file, hit the i key and then make any changes required. To close and save, hit the esc key and type :wq to write the changes to the file and exit/quit the file. Users may opt to use their preferred code editor instead.

Start in the user: section and change the MACHINE and ACCOUNT variables. For example, when running on a personal MacOS device, users might set:

user:
   RUN_ENVIR: community
   MACHINE: macos
   ACCOUNT: none

For a detailed description of these variables, see Section %s.

Users do not need to change the platform: section of the configuration file for this tutorial.

In the workflow: section of config.yaml, update EXPT_SUBDIR, CCPP_PHYS_SUITE, PREDEF_GRID_NAME, DATE_FIRST_CYCL, DATE_LAST_CYCL, and FCST_LEN_HRS.

workflow:
  USE_CRON_TO_RELAUNCH: false
  EXPT_SUBDIR: halloweenRAP
  CCPP_PHYS_SUITE: FV3_RAP
  PREDEF_GRID_NAME: RRFS_CONUS_13km
  DATE_FIRST_CYCL: '2019103012'
  DATE_LAST_CYCL: '2019103012'
  FCST_LEN_HRS: 36
  PREEXISTING_DIR_METHOD: rename
  VERBOSE: true
  COMPILER: intel

Note

Users may also want to set USE_CRON_TO_RELAUNCH: true and add CRON_RELAUNCH_INTVL_MNTS: 3. This will automate submission of workflow tasks when running the experiment. However, not all systems have cron.

EXPT_SUBDIR: This variable can be changed to any name the user wants from “halloweenRAP” to “HalloweenStorm1” to “askdfj” (but note that whitespace and some punctuation characters are not allowed). However, the best names will indicate useful information about the experiment. Since this tutorial helps users to compare the output from RAP and HRRR forecast input data, this tutorial will use halloweenRAP for the Halloween Storm experiment that uses RAP forecast data.

PREDEF_GRID_NAME: This experiment uses the RRFS_CONUS_13km, rather than the default RRFS_CONUS_25km grid. This 13-km resolution is used in the NOAA operational Rapid Refresh (RAP) model and is the resolution envisioned for the initial operational implementation of the Rapid Refresh Forecast System (RRFS). For more information on this grid, see Section %s.

CCPP_PHYS_SUITE: The FV3_RAP physics suite contains the evolving parameterizations used operationally in the NOAA Rapid Refresh (RAP) model; the suite is also a prime candidate under consideration for initial RRFS implementation and has been well-tested at the 13-km resolution. It is therefore an appropriate physics choice when using the RRFS_CONUS_13km grid.

DATE_FIRST_CYCL, DATE_LAST_CYCL, and FCST_LEN_HRS set parameters related to the date and duration of the forecast. Because this is a one-cycle experiment that does not use cycling or data assimilation, the date of the first cycle and last cycle are the same.

For a detailed description of other workflow: variables, see Section 3.1.3.

In the task_get_extrn_ics: section, add USE_USER_STAGED_EXTRN_FILES and EXTRN_MDL_SOURCE_BASEDIR_ICS. Users will need to adjust the file path to point to the location of the data on their system.

task_get_extrn_ics:
  EXTRN_MDL_NAME_ICS: RAP
  USE_USER_STAGED_EXTRN_FILES: true
  EXTRN_MDL_SOURCE_BASEDIR_ICS: /path/to/RAP/for_ICS

For a detailed description of the task_get_extrn_ics: variables, see Section %s.

Similarly, in the task_get_extrn_lbcs: section, add USE_USER_STAGED_EXTRN_FILES and EXTRN_MDL_SOURCE_BASEDIR_LBCS. Users will need to adjust the file path to point to the location of the data on their system.

task_get_extrn_lbcs:
  EXTRN_MDL_NAME_LBCS: RAP
  LBC_SPEC_INTVL_HRS: 3
  USE_USER_STAGED_EXTRN_FILES: true
  EXTRN_MDL_SOURCE_BASEDIR_LBCS: /path/to/RAP/for_LBCS

For a detailed description of the task_get_extrn_lbcs: variables, see Section %s.

Users do not need to modify the task_run_fcst: section for this tutorial.

In the rocoto:tasks: section, increase the walltime for the data-related tasks and metatasks. Then include the YAML configuration file containing the plotting task in the rocoto:tasks:taskgroups: section, like this:

rocoto:
  tasks:
    task_get_extrn_ics:
      walltime: 06:00:00
    task_get_extrn_lbcs:
      walltime: 06:00:00
    metatask_run_ensemble:
      task_make_lbcs_mem#mem#:
        walltime: 06:00:00
      task_run_fcst_mem#mem#:
        walltime: 06:00:00
    taskgroups: '{{ ["parm/wflow/prep.yaml", "parm/wflow/coldstart.yaml", "parm/wflow/post.yaml", "parm/wflow/plot.yaml"]|include }}'

Note

Rocoto tasks are run once each. A Rocoto metatask expands into one or more similar tasks by replacing the values between # symbols with the values under the var: key. See the Rocoto documentation for more information.

For more information on how to turn on/off tasks in the workflow, please see Section 2.3.3.2.2.2.

In the task_plot_allvars: section, add PLOT_FCST_INC: 6. Users may also want to add PLOT_FCST_START: 0 and PLOT_FCST_END: 36 explicitly, but these can be omitted since the default values are the same as the forecast start and end time respectively.

task_plot_allvars:
  COMOUT_REF: ""
  PLOT_FCST_INC: 6

PLOT_FCST_INC: This variable indicates the forecast hour increment for the plotting task. By setting the value to 6, the task will generate a .png file for every 6th forecast hour starting from 12z on October 30, 2019 (the 0th forecast hour) through the 36th forecast hour (November 1, 2019 at 0z).

After configuring the forecast, users can generate the forecast by running:

./generate_FV3LAM_wflow.py

To see experiment progress, users should navigate to their experiment directory. Then, use the rocotorun command to launch new workflow tasks and rocotostat to check on experiment progress.

cd /path/to/expt_dirs/halloweenRAP
rocotorun -w FV3LAM_wflow.xml -d FV3LAM_wflow.db -v 10
rocotostat -w FV3LAM_wflow.xml -d FV3LAM_wflow.db -v 10

Users will need to rerun the rocotorun and rocotostat commands above regularly and repeatedly to continue submitting workflow tasks and receiving progress updates.

Note

When using cron to automate the workflow submission (as described above), users can omit the rocotorun command and simply use rocotostat to check on progress periodically.

Users can save the location of the halloweenRAP directory in an environment variable (e.g., $HRAP). This makes it easier to navigate between directories later. For example:

export HRAP=/path/to/expt_dirs/halloweenRAP

Users should substitute /path/to/expt_dirs/halloweenRAP with the actual path to the experiment directory on their system. As long as a user remains logged into their system, they can run cd $HRAP, and it will take them to the halloweenRAP experiment directory. The variable will need to be reset for each login session.

2.5.1.4.2. Experiment 2: Changing the Forecast Input

Once the halloweenRAP case is running, users can return to the config.yaml file (in $USH) and adjust the parameters for a new forecast. In this forecast, users will change the forecast input to use HRRR data and alter a few associated parameters.

In the workflow: section of config.yaml, update EXPT_SUBDIR and PREDEF_GRID_NAME. Other parameters should remain the same.

workflow:
  EXPT_SUBDIR: halloweenHRRR
  PREDEF_GRID_NAME: RRFS_CONUScompact_13km

Note

Relative to the original CONUS domain, the “compact” CONUS domains are slightly smaller. The original CONUS domains were a bit too large to run with LBCs from HRRR, so the “compact” domains were created to be just small enough to work with HRRR data.

In the task_get_extrn_ics: section, update the values for EXTRN_MDL_NAME_ICS and USE_USER_STAGED_EXTRN_FILES and add EXTRN_MDL_FILES_ICS. Users may choose to comment out or remove EXTRN_MDL_SOURCE_BASEDIR_ICS, but this is not necessary.

task_get_extrn_ics:
  EXTRN_MDL_NAME_ICS: HRRR
  USE_USER_STAGED_EXTRN_FILES: false
  EXTRN_MDL_FILES_ICS:
  - '{yy}{jjj}{hh}00{fcst_hr:02d}00'

For a detailed description of the task_get_extrn_ics: variables, see Section %s.

Update the same values in the task_get_extrn_lbcs: section:

task_get_extrn_lbcs:
  EXTRN_MDL_NAME_LBCS: HRRR
  LBC_SPEC_INTVL_HRS: 3
  USE_USER_STAGED_EXTRN_FILES: false
  EXTRN_MDL_FILES_LBCS:
  - '{yy}{jjj}{hh}00{fcst_hr:02d}00'

For a detailed description of the task_get_extrn_lbcs: variables, see Section %s.

After configuring the forecast, users can generate the second forecast by running:

./generate_FV3LAM_wflow.py

To see experiment progress, users should navigate to their experiment directory. As in the first forecast, the following commands allow users to launch new workflow tasks and check on experiment progress.

cd /path/to/expt_dirs/halloweenHRRR
rocotorun -w FV3LAM_wflow.xml -d FV3LAM_wflow.db -v 10
rocotostat -w FV3LAM_wflow.xml -d FV3LAM_wflow.db -v 10

Note

When using cron to automate the workflow submission (as described above), users can omit the rocotorun command and simply use rocotostat to check on progress periodically.

Note

If users have not automated their workflow using cron, they will need to ensure that they continue issuing rocotorun commands to launch all of the tasks in each experiment. While switching between experiment directories to run rocotorun and rocotostat commands in both directories is possible, it may be easier to finish the halloweenRAP experiment’s tasks before starting on halloweenHRRR.

As with the halloweenRAP experiment, users can save the location of the halloweenHRRR directory in an environment variable (e.g., $HHRRR). This makes it easier to navigate between directories later. For example:

export HHRRR=/path/to/expt_dirs/halloweenHRRR

Users should substitute /path/to/expt_dirs/halloweenHRRR with the actual path on their system.

2.5.1.5. How to Analyze Results

Navigate to halloweenHRRR/2019103012/postprd and/or halloweenRAP/2019203012/postprd. These directories contain the post-processed data generated by the UPP from the Halloween Storm forecasts. After the plot_allvars task completes, this directory will contain .png images for several forecast variables.

2.5.1.5.1. Copy .png Files onto Local System

Users who are working on the cloud or on an HPC cluster may want to copy the .png files onto their local system to view in their preferred image viewer. Detailed instructions are available in the Introduction to SSH & Data Transfer.

In summary, users can run the scp command in a new terminal/command prompt window to securely copy files from a remote system to their local system if an SSH tunnel is already established between the local system and the remote system. Users can adjust one of the following commands for their system:

scp username@your-IP-address:/path/to/source_file_or_directory /path/to/destination_file_or_directory
# OR
scp -P 12345 username@localhost:/path/to/source_file_or_directory /path/to/destination_file_or_directory

Users would need to modify username, your-IP-address, -P 12345, and the file paths to reflect their systems’ information. See the Introduction to SSH & Data Transfer for example commands.

2.5.1.5.2. Examining Forecast Plots at Peak Intensity

This experiment will be looking at plots from HRRR and RAP input files while the Halloween Storm is at or approaching peak intensity.

2.5.1.5.2.1. 250mb Wind

An effective weather forecast begins with analyzing a 250mb wind chart. By using this wind plot, forecasters can identify key features such as jet stream placement, jet maxima, troughs, ridges, and more. This analysis also helps pinpoint areas with the potential for the strongest severe weather.

In the 250mb wind plots below, the halloweenHRRR and halloweenRAP plots are nearly identical at forecast hour f036. This shows great model agreement. Analyzing this chart we can see multiple ingredients signaling a significant severe weather event over the eastern CONUS. The first thing to notice is the placement of the jet streak along with troughing approaching the eastern US. Also notice an extreme 150KT jet max over southern Ohio further fueling severe weather. The last thing to notice is the divergence aloft present over the eastern CONUS; seeing divergence present all the way up to 250mb indicates a strong system.

Fig. 2.6 RAP Plot for 250mb Wind

Fig. 2.7 HRRR Plot for 250mb Wind

2.5.1.5.2.2. 10m Wind

The 10m wind plots allows forecasters to pick up on patterns closer to the surface. It shows features such as convergence and pressure areas.

In the 10m wind plots below, the halloweenHRRR and halloweenRAP are once again very similar, which makes sense given that the 250mb wind plots are also so similar. We can see a few key features on this chart. The most important is the area of convergence taking place over the east coast which is driving the line of severe storms.

Fig. 2.8 RAP Plot for 10m Winds

Fig. 2.9 HRRR Plot for 10m Winds

2.5.1.5.2.3. Composite Reflectivity

Reflectivity images visually represent the weather based on the energy (measured in decibels [dBZ]) reflected back from radar. Composite reflectivity generates an image based on reflectivity scans at multiple elevation angles, or “tilts”, of the antenna. See https://www.noaa.gov/jetstream/reflectivity for a more detailed explanation of composite reflectivity.

In the composite reflectivity plots below, the halloweenHRRR and halloweenRAP models remain quite similar, as expected. Utilizing the reflectivity plots provides the final piece of the puzzle. From the previous analyses, we already had a good understanding of where the storms were likely to occur. Composite reflectivity serves as an additional tool, allowing us to visualize where the models predict storm placement. In this case, the strongest storms are indicated by higher dBZ values and appear to be concentrated in the NC/VA region.

Fig. 2.10 RAP Plot for Composite Reflectivity

Fig. 2.11 HRRR Plot for Composite Reflectivity

2.5.1.5.3. Experiment 3: Performing METplus Verification

In this experiment, we will use the METplus verification framework to evaluate the accuracy of the HRRR forecasts for the Halloween Storm Case. The METplus tools provide a robust and customizable way to assess forecast skill by comparing model output against observational data. This section will guide you through the steps to perform METplus verification for this case.

Note

For tutorial purposes we will only run this test using HRRR data, but users should feel free to do the same experiment with RAP data.

2.5.1.6. Set Up Verification

Follow the instructions below to reproduce a forecast for this event using your own model setup! Make sure to install and build the latest version of the SRW Application (v3.0.0). develop branch code is constantly changing, so it does not provide a consistent baseline for comparison.

On Level 1 systems, users can find data for the Indianapolis Severe Weather Forecast in the usual data locations (see Section 2.3 for a list).

On other systems, users need to download the HalloweenStormData.tar.gz file using any of the following methods:

1. Download directly from the S3 bucket using a browser. The data is available at https://noaa-ufs-srw-pds.s3.amazonaws.com/HalloweenStormData/HalloweenStormData.tar.gz

2. Download from a terminal using the AWS command line interface (CLI), if installed:

aws s3 cp https://noaa-ufs-srw-pds.s3.amazonaws.com/HalloweenStormData/HalloweenStormData.tar.gz HalloweenStormData.tar.gz

1. Download from a terminal using wget:

wget https://noaa-ufs-srw-pds.s3.amazonaws.com/HalloweenStormData/HalloweenStormData.tar.gz

This tar file contains IC/LBC files, observation data, model/forecast output, and MET verification output for the sample forecast. Users who have never run the SRW App on their system before will also need to download (1) the fix files required for SRW App forecasts and (2) the NaturalEarth shapefiles required for plotting. Users can download the fix file data from a browser at https://noaa-ufs-srw-pds.s3.amazonaws.com/experiment-user-cases/release-public-v2.2.0/out-of-the-box/fix_data.tgz or visit Section 3.2.3.1 for instructions on how to download the data with wget. NaturalEarth files are available at https://noaa-ufs-srw-pds.s3.amazonaws.com/develop-20240618/NaturalEarth.tar.gz. See the Section 2.3.3.2.3 for more information on plotting.

After downloading HalloweenStormData.tar.gz using one of the three methods above, untar the downloaded compressed archive file:

tar xvfz HalloweenStormData.tar.gz

Save the path to this file in the HalloweenDATA environment variable:

cd HalloweenStormData
export HalloweenDATA=$PWD

Note

Users can untar the fix files and Natural Earth files by substituting those file names in the commands above.

2.5.1.6.1. Load the Workflow

To load the workflow environment, run:

source /path/to/dtc-vx-workflow/setup_conda.sh

Users running a csh/tcsh shell would run source /path/to/etc/lmod-setup.csh <platform> in place of the first command above.

After loading the workflow, users should follow the instructions printed to the console. Usually, the instructions will tell the user to run conda activate srw_app.

2.5.1.6.2. Configure the Verification Sample Case

Once the workflow environment is loaded, copy the out-of-the-box configuration:

cd /path/to/ufs-srweather-app/ush
cp config.community.yaml config.yaml

where /path/to/ufs-srweather-app/ush is replaced by the actual path to the ufs-srweather-app/ush directory on the user’s system.

Then, edit the configuration file (config.yaml) to include the variables and values in the sample configuration excerpt below (variables not listed below do not need to be changed or removed). Users must be sure to substitute values in <> with values appropriate to their system.

 user:
    MACHINE: <your_machine_name>
    ACCOUNT: <my_account>
 platform:
    # Add EXTRN_MDL_DATA_STORES variable to config.yaml
    EXTRN_MDL_DATA_STORES: aws
 workflow:
    USE_CRON_TO_RELAUNCH: true
    EXPT_SUBDIR: halloweenstormHRRRMETPLUS
    CCPP_PHYS_SUITE: FV3_RAP
    PREDEF_GRID_NAME: RRFS_CONUScompact_13km
    DATE_FIRST_CYCL: '2019103012'
    DATE_LAST_CYCL: '2019103012'
    FCST_LEN_HRS: 24
    PREEXISTING_DIR_METHOD: rename
    VERBOSE: true
    # Change to gnu if using a gnu compiler; otherwise, no change
    COMPILER: intel
 task_get_extrn_ics:
    # Add EXTRN_MDL_NAME_ICS and EXTRN_MDL_FILES_ICS variables to config.yaml
    EXTRN_MDL_NAME_ICS: HRRR
    USE_USER_STAGED_EXTRN_FILES: false
    EXTRN_MDL_FILES_ICS:
    -  '{yy}{jjj}{hh}00{fcst_hr:02d}00'
 task_get_extrn_lbcs:
    # Add EXTRN_MDL_NAME_LBCS, USE_USER_STAGED_EXTRN_FILES, and EXTRN_MDL_FILES_LBCS variables to config.yaml
    EXTRN_MDL_NAME_LBCS: HRRR
    LBC_SPEC_INTVL_HRS: 3
    USE_USER_STAGED_EXTRN_FILES: false
    EXTRN_MDL_FILES_LBCS:
    -  '{yy}{jjj}{hh}00{fcst_hr:02d}00'
 task_plot_allvars:
   COMOUT_REF: ""
   PLOT_FCST_INC: 6
 verification:
   VX_FCST_MODEL_NAME: FV3_GFS_v16_CONUS_25km
   CCPA_OBS_DIR: /path/to/HalloweenStormData/obs_data/ccpa/proc
   MRMS_OBS_DIR: /path/to/HalloweenStormData/obs_data/mrms/proc
   NDAS_OBS_DIR: /path/to/HalloweenStormData/obs_data/ndas/proc
 rocoto:
tasks:
 task_get_extrn_ics:
   walltime: 06:00:00
 task_get_extrn_lbcs:
   walltime: 06:00:00
 metatask_run_ensemble:
   task_make_lbcs_mem#mem#:
     walltime: 06:00:00
   task_run_fcst_mem#mem#:
     walltime: 06:00:00
 taskgroups: '{{ ["parm/wflow/prep.yaml", "parm/wflow/coldstart.yaml", "parm/wflow/post.yaml", "parm/wflow/verify_pre.yaml", "parm/wflow/verify_det.yaml", "parm/wflow/test.yaml"]|include }}'

Hint

To open the configuration file in the command line, users may run the command:

vi config.yaml

To modify the file, hit the i key and then make any changes required. To close and save, hit the esc key and type :wq. Users may opt to use their preferred code editor instead.

For additional configuration guidance, refer to configuring the SRW App.

2.5.1.6.3. Generate the Experiment

Generate the experiment by running this command from the ush directory:

./generate_FV3LAM_wflow.py

2.5.1.6.4. Run the Experiment

Navigate (cd) to the experiment directory ($EXPTDIR) and run the launch script:

./launch_FV3LAM_wflow.sh

To see experiment progress, users should navigate to their experiment directory. The following commands allow users to launch new workflow tasks and check on experiment progress.

cd path/to/expt_dirs/halloweenstormHRRRMETPLUS
rocotorun -w FV3LAM_wflow.xml -d FV3LAM_wflow.db -v 10
rocotostat -w FV3LAM_wflow.xml -d FV3LAM_wflow.db -v 10

Users who prefer to automate the workflow via crontab or who need guidance for running without the Rocoto workflow manager should refer to Section 2.3.4 for these options.

If a problem occurs and a task goes DEAD, view the task log files in $EXPTDIR/log to determine the problem. Then refer to Section 4.2.3.1 to restart a DEAD task once the problem has been resolved. For troubleshooting assistance, users are encouraged to post questions on the new SRW App GitHub Discussions Q&A page.

2.5.1.7. Compare

Once the experiment has completed (i.e., all tasks have “SUCCEEDED” and the end of the log.launch_FV3LAM_wflow file lists “Workflow status: SUCCESS”), users can see how the forecast verified. From the expt_dirs users should navigate to their experiment (e.g., halloweenstormHRRRMETPLUS). Then, navigate to the 2019103012 subdirectory, and finally, enter the metprd directory to view your experiment’s results.

2.5.1.7.1. MetPlus File Types

For information on different file types found in the metprd directory users should reference Section 2.13.

2.5.1.7.2. Analyzing HRRR Results

This section, will analyze how different variables were verified in the HRRR forecast. To do this, users will be examining the RMSE and MBIAS scores for Temperature and Dew Point (DPT) on the surface using the point_stat file type.

Interpretation:

• A lower RMSE indicates that the model forecast value is closer to the observed value.

• If MBIAS > 1, then the value for a given forecast variable is too high on average by (MBIAS - 1)%. If MBIAS < 1, then the forecasted value is too low on average by (1 - MBIAS)%.

• Find column 66-68 for MBIAS and 78-80 for RMSE statistics.

2.5.1.8. Temperature

To begin examining the temperature variable, navigate to path/to/2019103012/metprd. Use vi to open PointStat/, where a menu of files will appear. Navigate using the arrow and enter keys on your. Select this file:

point_stat_FV3_GFS_v16_CONUS_25km_mem000_ADPSFC_NDAS_240000L_20191031_120000V.stat

Press Enter to open this file. You will now be able to see verification data for different types of variables.

We can see that on October 31st, at 2400L, temperature received an MBIAS score of 1.7212 and an RMSE score of 2.38512. Looking at these scores, we can conclude that for this forecast hour, the temperature was over forecasted and was greater than the observed temperature.

2.5.1.9. Dew Point

Good news! The DPT variable can be examined within the same point_stat file, so there is no need to switch files. DPT received an MBIAS score of 1.5978 and an RMSE score of 2.51836. Looking at these scores, we can see that, similar to temperature, the DPT was over forecasted and was greater than the observed dewpoint.