WeatherBlend

P(wet) vs observed wet-hour

Blue bands = observed wet hours (≥ 0.1 mm). MO Spot PoP uses a looser "any measurable precip" threshold — read it as direction-of-effect.

Chards Snowdon Hill

Phase 3c (rich)

Phase 3a (lean)

Phase comparison — +24h lead

Rolling Brier (P(wet))

Chards Snowdon Hill — 30-day rolling Brier per (version, lead). Lower better.

<24hr

Lead +24h

Lead +48h

Lead +72h

Lead +96h

Lead +120h

Rainfall amount (3f) — distributional skill

Phase 3f distributional skill — Chards Snowdon Hill. CRPS (lower better) is the primary skill metric. Coverage should sit at 0.80 for a well-calibrated 80% interval; PIT should be flat. Exceedance Brier is per-threshold P(rain ≥ T) — same units as P(wet) Brier above. Sample size shown per row.

Rolling CRPS

Rolling 80% coverage

Target = 0.80. Below = under-spread (intervals too narrow); above = over-spread.

PIT histogram (most recent verify week, all leads pooled)

Red dashed line = perfectly flat (N / 10 per bin). Bumps at 0.0 / 0.9 flag under-spread; tilt flags mean bias.

Exceedance Brier (latest verify week, per lead)