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.

Bovey Tracey

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)
97%99%99% confident dry (P=97% · run 3–9h) 15:00Z (96%)
100%100%100% confident dry (P=100% · run 7–9h) 09:00Z (100%)
—100%100% confident dry (P=100% · run 9–9h) 15:00Z (100%)
——100% confident dry (P=100% · run 9–9h) 15:00Z (100%)
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
98%92%95% 50% confident dry (P=98% · τ=21%)
99%98%98% 83% confident dry (P=99% · τ=21%)
—98%98% 100% confident dry (P=98% · τ=22%)
——97% 100% confident dry (P=97% · τ=19%)
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)
93%96%97% confident dry (P=93% · run 3–9h) 14:00Z (93%)
100%100%100% confident dry (P=100% · run 7–9h) 08:00Z (99%)
—100%100% confident dry (P=100% · run 9–9h) 11:00Z (99%)
——100% confident dry (P=100% · run 9–9h) 12:00Z (100%)
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
92%82%83% 50% confident dry (P=92% · τ=28%)
97%96%96% 83% confident dry (P=97% · τ=28%)
—97%97% 100% confident dry (P=97% · τ=27%)
——97% 100% confident dry (P=97% · τ=35%)
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)
87%92%93% confident dry (P=87% · run 3–9h) 13:00Z (90%)
99%99%99% confident dry (P=99% · run 7–9h) 08:00Z (98%)
—100%100% confident dry (P=100% · run 9–9h) 11:00Z (99%)
——100% confident dry (P=100% · run 9–9h) 11:00Z (99%)
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)
81%85%88% confident dry (P=81% · run 3–9h) 12:00Z (84%)
96%97%96% confident dry (P=96% · run 7–9h) 08:00Z (98%)
—100%99% confident dry (P=100% · run 9–9h) 11:00Z (99%)
——99% confident dry (P=99% · run 9–9h) 10:00Z (99%)
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
81%55%36% 33% confident dry (P=81% · τ=33%)
92%96%96% 67% confident dry (P=92% · τ=33%)
—92%97% 100% confident dry (P=92% · τ=26%)
——98% 83% confident dry (P=98% · τ=26%)
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)
71%80%82% ambiguous (P=71% · run 3–9h) 11:00Z (78%)
93%96%94% confident dry (P=93% · run 7–9h) 08:00Z (96%)
—99%99% confident dry (P=99% · run 9–9h) 11:00Z (99%)
——99% confident dry (P=99% · run 9–9h) 09:00Z (99%)
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.