CPA Weather Lab Home
Research surface · Forecast verification

See where the forecast held up.

A compact, provenance-first verification workspace for precipitation forecasts, ensemble spread, and retrospective MRMS truth.

Local deterministic fixture · no live data/Run vrf_demo_20260929_f002
01 · Case selection

Choose a comparable slice

✓ semantics pass
Valid-time aligned1 h accumulationTruth: retrospective onlyGrid: 3 km demo
02 · Spatial comparison

Forecast precipitation

0 2 8+ mm
Central Pennsylvania02Z–03Z UTC
N
AOI

Synthetic deterministic forecast field. Switch layers to compare the same valid interval.

Bias+0.18 mmforecast − observed
MAE0.68 mmabsolute error
Correlation0.97paired hourly values
Frequency bias1.14threshold 0.254 mm
FSS0.833×3 cells · thresholded
03 · Probabilistic view

Reliability & discrimination

raw N=5 ensemble fraction
observed frequency
forecast probability
observed ideal
Brier score 0.09
Skill vs climatology 0.71
04 · Ensemble shape

Rank histogram

truth rank
21
42
33
54
25
16

Counts across 12 fixture cases. Ties use midpoint rank.

05 · Spread / skill

Ensemble sharpness

mm
tightwide
0.42 spread1.18 RMSE

Increasing spread tracks increasing error in this demo slice.

↗
Truth semantics

Forecast issued first. MRMS verifies later.

This run archives the forecast at issue time, then compares its 1-hour interval to MultiSensor QPE Pass2 after the event. Pass2 is never used to generate the forecast.

00Z issue→02Z valid→Pass2 later
06 · Audit trail

Every score has a source

ForecastNOAA RRFS · m003demo://rrfs/m003/f002
Retrospective truthMRMS MultiSensor Pass2demo://mrms/pass2/f002
Semantics profileprecip48_rrfs_s25 · v1.01 h · ≥0.254 mm · six-dry-hour closure
RemappingConservative area overlapauthoritative area / edge products
DEMO FIXTURE. Values on this page are deterministic synthetic examples for interface review. They are never live observations, operational guidance, or a scientific conclusion.