WeatherBlend

Multi-model forecast blending for Bonehill Rocks, Dartmoor

Dry-window forecast

P(at least one N-hour dry block in 09:00–18:00 local). 3b champion plus Monte-Carlo challengers.

Dartmoor Nr Hexworthy

2-hour dry window

Phase 3p — Gaussian copula MC over Phase 3o hourly P(wet)

Gaussian copula MC over the rich+orographic Phase 3o's hourly P(wet) marginals. Single empirical Σ per station, fit on train-split observed daytime wet/dry binary sequences (daytime-shape autocorrelation is lead-independent for truth, so per-lead Σs pool into one). Captures within-day wet/dry autocorrelation that an iid sampler misses. Bake-off Brier 0.1064 aggregate on the 9 (station, lead) Bonehill cells.

Target date (UTC) +24h +48h +72h Confidence Best start (UTC, P dry block)
99%98%99% confident dry (P=99% · run 3–9h) 14:00Z (94%)
100%99%100% confident dry (P=100% · run 6–9h) 08:00Z (96%)
—59%69% ambiguous (P=59% · run 0–6h) 15:00Z (35%)
——100% confident dry (P=100% · run 6–9h) 08:00Z (95%)
Monte-Carlo P(an 2-hour dry block runs from each start hour) — one line per forecast horizon, soonest target day shown. The peak is the best time to set off; the height is how likely it stays dry. Each hour is its own probability, so the lines need not sum to the daily "any dry window" figure.

3-hour dry window

Phase 3b — lean (53 features)

53-feature LightGBM. Champion.

Target date (UTC) +24h +48h +72h Model agreement Confidence
93%89%89% 100% confident dry (P=93% · τ=45%)
96%95%97% 100% confident dry (P=96% · τ=45%)
—45%55% 50% confident wet (P=45% · τ=45%)
——95% 100% confident dry (P=95% · τ=41%)
Phase 3p — Gaussian copula MC over Phase 3o hourly P(wet)

Gaussian copula MC over the rich+orographic Phase 3o's hourly P(wet) marginals. Single empirical Σ per station, fit on train-split observed daytime wet/dry binary sequences (daytime-shape autocorrelation is lead-independent for truth, so per-lead Σs pool into one). Captures within-day wet/dry autocorrelation that an iid sampler misses. Bake-off Brier 0.1064 aggregate on the 9 (station, lead) Bonehill cells.

Target date (UTC) +24h +48h +72h Confidence Best start (UTC, P dry block)
95%95%96% confident dry (P=95% · run 3–9h) 14:00Z (92%)
99%98%99% confident dry (P=99% · run 6–9h) 08:00Z (94%)
—43%46% ambiguous (P=43% · run 0–6h) 14:00Z (28%)
——99% confident dry (P=99% · run 6–9h) 08:00Z (92%)
Monte-Carlo P(an 3-hour dry block runs from each start hour) — one line per forecast horizon, soonest target day shown. The peak is the best time to set off; the height is how likely it stays dry. Each hour is its own probability, so the lines need not sum to the daily "any dry window" figure.

4-hour dry window

Phase 3b — lean (53 features)

53-feature LightGBM. Champion.

Target date (UTC) +24h +48h +72h Model agreement Confidence
79%68%73% 100% confident dry (P=79% · τ=55%)
93%96%95% 100% confident dry (P=93% · τ=55%)
—40%55% 33% confident wet (P=40% · τ=46%)
——77% 100% confident dry (P=77% · τ=47%)
Phase 3p — Gaussian copula MC over Phase 3o hourly P(wet)

Gaussian copula MC over the rich+orographic Phase 3o's hourly P(wet) marginals. Single empirical Σ per station, fit on train-split observed daytime wet/dry binary sequences (daytime-shape autocorrelation is lead-independent for truth, so per-lead Σs pool into one). Captures within-day wet/dry autocorrelation that an iid sampler misses. Bake-off Brier 0.1064 aggregate on the 9 (station, lead) Bonehill cells.

Target date (UTC) +24h +48h +72h Confidence Best start (UTC, P dry block)
90%91%92% confident dry (P=90% · run 3–9h) 13:00Z (88%)
98%94%98% confident dry (P=98% · run 6–9h) 08:00Z (92%)
—29%34% ambiguous (P=29% · run 0–6h) 13:00Z (19%)
——97% confident dry (P=97% · run 6–9h) 08:00Z (89%)
Monte-Carlo P(an 4-hour dry block runs from each start hour) — one line per forecast horizon, soonest target day shown. The peak is the best time to set off; the height is how likely it stays dry. Each hour is its own probability, so the lines need not sum to the daily "any dry window" figure.

5-hour dry window

Phase 3p — Gaussian copula MC over Phase 3o hourly P(wet)

Gaussian copula MC over the rich+orographic Phase 3o's hourly P(wet) marginals. Single empirical Σ per station, fit on train-split observed daytime wet/dry binary sequences (daytime-shape autocorrelation is lead-independent for truth, so per-lead Σs pool into one). Captures within-day wet/dry autocorrelation that an iid sampler misses. Bake-off Brier 0.1064 aggregate on the 9 (station, lead) Bonehill cells.

Target date (UTC) +24h +48h +72h Confidence Best start (UTC, P dry block)
61%80%68% ambiguous (P=61% · run 3–9h) 12:00Z (72%)
95%92%95% confident dry (P=95% · run 6–9h) 08:00Z (91%)
—20%25% confident wet (P=20% · run 0–6h) 12:00Z (14%)
——94% confident dry (P=94% · run 6–9h) 08:00Z (88%)
Monte-Carlo P(an 5-hour dry block runs from each start hour) — one line per forecast horizon, soonest target day shown. The peak is the best time to set off; the height is how likely it stays dry. Each hour is its own probability, so the lines need not sum to the daily "any dry window" figure.

6-hour dry window

Phase 3b — lean (53 features)

53-feature LightGBM. Champion.

Target date (UTC) +24h +48h +72h Model agreement Confidence
49%50%51% 50% ambiguous (P=49% · τ=53%)
87%81%70% 83% confident dry (P=87% · τ=53%)
—10%19% 17% confident wet (P=10% · τ=51%)
——48% 83% ambiguous (P=48% · τ=55%)
Phase 3p — Gaussian copula MC over Phase 3o hourly P(wet)

Gaussian copula MC over the rich+orographic Phase 3o's hourly P(wet) marginals. Single empirical Σ per station, fit on train-split observed daytime wet/dry binary sequences (daytime-shape autocorrelation is lead-independent for truth, so per-lead Σs pool into one). Captures within-day wet/dry autocorrelation that an iid sampler misses. Bake-off Brier 0.1064 aggregate on the 9 (station, lead) Bonehill cells.

Target date (UTC) +24h +48h +72h Confidence Best start (UTC, P dry block)
50%74%60% ambiguous (P=50% · run 3–9h) 11:00Z (62%)
91%83%92% confident dry (P=91% · run 6–9h) 10:00Z (88%)
—14%19% confident wet (P=14% · run 0–6h) 11:00Z (11%)
——90% confident dry (P=90% · run 6–9h) 08:00Z (86%)
Monte-Carlo P(an 6-hour dry block runs from each start hour) — one line per forecast horizon, soonest target day shown. The peak is the best time to set off; the height is how likely it stays dry. Each hour is its own probability, so the lines need not sum to the daily "any dry window" figure.

Dry hour = all four 15-min readings ≤ 0.1 mm. Search bounded to 09:00–18:00 local.