Statistics Use Case: HRRR Grid-to-Grid Verification

This use case demonstrates how to run VCasT’s Statistics module to compute a mix of continuous and categorical verification metrics directly from GRIB2 forecast and reference files, without a MET pre-processing step.

Prerequisites

Before running the example, make sure VCasT is installed. Follow the installation steps in Quick Start Guide.

Run the Example

Clone the test repository:

git clone https://github.com/NOAA-GSL/VCasT-tests
cd VCasT-tests/examples/Stats/all_stats-0001

Run VCasT with the provided YAML file:
```
vcast stat.yaml
```
This locates the forecast and reference GRIB2 files under ../input_files/HRRR, optionally interpolates them, and writes the computed metrics to stat.out.

YAML Configuration Explained

Below is the content of the example stat.yaml, which configures VCasT to:

Process a single hourly time step of 2-meter temperature (T2M)
Compare HRRR forecast fields against HRRR reference fields
Compute continuous metrics (rmse, bias, quantiles, mae, corr, stdev) as well as categorical metrics thresholded at 290 K with a 2-grid-point radius of influence (gss, fbias, pod, far, csi, sr)

Sample stat.yaml configuration

start_date: "2024-04-01_02:00:00"
end_date: "2024-04-01_02:00:00"
interval_hours: "1"

start_lead_time: 1
end_lead_time: 1
interval_lead_time: 1

vars: "T2M"

fcst_model: "HRRR"
fcst_file_template: "../input_files/HRRR/hrrr.{year}{month}{day}{hour}.wrfprsf{lead_time}.T2M.grib2"

ref_model: "HRRR"
ref_file_template: "../input_files/HRRR/hrrr.{valid_year}{valid_month}{valid_day}{valid_hour}.wrfprsf00.T2M.grib2"

output_dir: "."
output_filename: "stat.out"

stat_type: "det"
stat_name:
  - "rmse"
  - "bias"
  - "quantiles"
  - "mae"
  - "gss:290:290:2"
  - "fbias:290:290:2"
  - "pod:290:290:2"
  - "far:290:290:2"
  - "csi:290:290:2"
  - "corr"
  - "stdev"
  - "sr:290:290:2"

interpolation: false
target_grid: "../input_files/REF/ref_2024040102_f001.nc"

processes: 1

Output

The output file (stat.out) is a tab-separated file with one row per date/lead-time/variable/level, and one column per requested statistic (with quantiles expanding into six columns: 25p, 50p, 75p, IQR, LW, UW).

To feed these results into the Aggregation or Plotting components, point their input_file / vars settings at this output file.