Statistics Component

The Statistics (stat) component is the core of VCasT. It reads forecast and reference fields directly from GRIB2, NetCDF, or Zarr files, optionally interpolates them onto a common target grid, and computes verification metrics without requiring a MET pre-processing step.

Unlike the MET Stat component (which post-processes existing MET .stat output), the Statistics component operates on raw model and observation files directly, driven by a file-template configuration that tells VCasT how to locate forecast/reference pairs across a date range.

Supported metrics (set via stat_name in the configuration file):

• rmse, bias, mae, stdev, corr: continuous error metrics

• quantiles: Q1/median/Q3, IQR, and outlier bounds

• gss, fbias, pod, far, sr, csi: categorical/contingency-table metrics, each requiring a forecast threshold, an observation threshold, and a radius of influence, e.g. "gss:290:290:2"

• fss: Fractions Skill Score, requiring a forecast threshold, an observation threshold, and a neighborhood window size, e.g. "fss:20:20:1"

Results are written to a single tab-separated output file with one row per date/lead-time/variable/level (and, for ensemble runs, per member).

Note

Ensemble mode (stat_type: ens) currently supports the fss and reliability metrics, computed across all listed members at once rather than per-member.

• Statistics YAML Configuration
• Statistics Use Case: HRRR Grid-to-Grid Verification