WeatherBlend

Multi-model forecast blending for Sennen, Cornwall

Dry-window models

Per-(station, window) P(N-hour dry block in 09–18 local). 3b LightGBM + 3p copula MC. Brier, lower better. Δ vs best single NWP — negative = blend wins.

Dry window — Trengwainton — 2-hour

Phase 3q · v2026-08-02_154011_phase3q Δ +0.000 vs prev train

Phase 3q blender. Trained 2026-08-02. Metric: Test Brier.

Lead Blend Best single Δ vs best
+24h 0.000 (0.000 val)
+48h 0.000 (0.000 val)
+72h 0.000 (0.000 val)
Verify history (13 runs)

Mon + Thu rolling Brier. Per-lead cells turn red when the rolling metric breaches the lead-specific drift threshold; check the verify report for the per-cell breakdown. Version column names the trained model — a fresh champion takes ~5-9d to show.

Run (UTC) Version N +24h+48h+72h+96h+120h
v2026-07-26_154305_phase3q 8 0.000
0.000
0.000
v2026-07-19_152830_phase3q 22 0.000
0.000
0.000
v2026-07-19_152830_phase3q 10 0.000
0.000
0.000
v2026-07-12_153204_phase3q 22 0.000
0.000
0.000
v2026-07-12_153204_phase3q 10 0.000
0.000
0.000
v2026-07-05_152333_phase3q 14 0.000
0.000
0.000
v2026-07-05_152333_phase3q 10 0.000
0.000
0.000
v2026-06-28_223429_phase3q 16 0.000
0.000
0.000
v2026-06-28_164125_phase3q
v2026-06-28_223429_phase3q
8 0.000
0.000
0.000
v2026-06-21_152916_phase3q 18 0.000
0.000
0.000
v2026-06-14_153851_phase3q
v2026-06-21_152916_phase3q
23 0.000
0.000
0.003
v2026-06-14_153851_phase3q 21 0.007
0.004
0.005
v2026-06-14_153851_phase3q 9 0.016
0.009
0.001

Dry window — Trengwainton — 3-hour

Phase 3b · v2026-08-02_153737 Δ -0.004 vs prev train

53-feature LightGBM per-(station, window). Trained 2026-08-02. Metric: Test Brier.

Lead Blend Best single Δ vs best
+24h 0.036 has_dry_window_mf (0.036) +1.8%
+48h 0.032 has_dry_window_icon (0.093) -65.1%
+72h 0.033 has_dry_window_icon (0.093) -64.7%
Verify history (13 runs)

Mon + Thu rolling Brier. Per-lead cells turn red when the rolling metric breaches the lead-specific drift threshold; check the verify report for the per-cell breakdown. Version column names the trained model — a fresh champion takes ~5-9d to show.

Run (UTC) Version N +24h+48h+72h+96h+120h
v2026-07-26_154025 8 0.000
0.001
0.001
v2026-07-19_152622 22 0.000
0.001
0.001
v2026-07-19_152622 10 0.000
0.001
0.001
v2026-07-12_152939 22 0.000
0.001
0.003
v2026-07-12_152939 10 0.000
0.001
0.016
v2026-07-05_152108 14 0.000
0.001
0.000
v2026-07-05_152108 10 0.000
0.001
0.000
v2026-06-28_223229 16 0.000
0.001
0.001
v2026-06-28_164003
v2026-06-28_223229
8 0.000
0.001
0.000
v2026-06-21_152620 18 0.000
0.001
0.000
v2026-06-14_153642
v2026-06-21_152620
23 0.000
0.001
0.001
v2026-06-14_153642 21 0.024
0.004
0.002
v2026-06-14_153642 9 0.057
0.006
0.002

Phase 3q · v2026-08-02_154011_phase3q Δ +0.000 vs prev train

Phase 3q blender. Trained 2026-08-02. Metric: Test Brier.

Lead Blend Best single Δ vs best
+24h 0.000 (0.000 val)
+48h 0.000 (0.000 val)
+72h 0.000 (0.000 val)
Verify history (13 runs)

Mon + Thu rolling Brier. Per-lead cells turn red when the rolling metric breaches the lead-specific drift threshold; check the verify report for the per-cell breakdown. Version column names the trained model — a fresh champion takes ~5-9d to show.

Run (UTC) Version N +24h+48h+72h+96h+120h
v2026-07-26_154305_phase3q 8 0.000
0.000
0.000
v2026-07-19_152830_phase3q 22 0.000
0.000
0.000
v2026-07-19_152830_phase3q 10 0.000
0.000
0.000
v2026-07-12_153204_phase3q 22 0.000
0.000
0.001
v2026-07-12_153204_phase3q 10 0.000
0.001
0.002
v2026-07-05_152333_phase3q 14 0.000
0.000
0.000
v2026-07-05_152333_phase3q 10 0.000
0.000
0.000
v2026-06-28_223429_phase3q 16 0.000
0.000
0.001
v2026-06-28_164125_phase3q
v2026-06-28_223429_phase3q
8 0.001
0.000
0.000
v2026-06-21_152916_phase3q 18 0.000
0.000
0.001
v2026-06-14_153851_phase3q
v2026-06-21_152916_phase3q
23 0.000
0.000
0.016
v2026-06-14_153851_phase3q 21 0.032
0.020
0.028
v2026-06-14_153851_phase3q 9 0.068
0.045
0.016

Dry window — Trengwainton — 4-hour

Phase 3b · v2026-08-02_153805 Δ -0.004 vs prev train

53-feature LightGBM per-(station, window). Trained 2026-08-02. Metric: Test Brier.

Lead Blend Best single Δ vs best
+24h 0.070 has_dry_window_icon (0.114) -38.9%
+48h 0.062 has_dry_window_icon (0.129) -52.1%
+72h 0.055 has_dry_window_icon (0.143) -61.3%
Verify history (13 runs)

Mon + Thu rolling Brier. Per-lead cells turn red when the rolling metric breaches the lead-specific drift threshold; check the verify report for the per-cell breakdown. Version column names the trained model — a fresh champion takes ~5-9d to show.

Run (UTC) Version N +24h+48h+72h+96h+120h
v2026-07-26_154053 8 0.001
0.001
0.001
v2026-07-19_152644 22 0.001
0.001
0.000
v2026-07-19_152644 10 0.001
0.001
0.000
v2026-07-12_153005 22 0.001
0.000
0.002
v2026-07-12_153005 10 0.001
0.000
0.005
v2026-07-05_152133 14 0.001
0.001
0.000
v2026-07-05_152133 10 0.001
0.001
0.000
v2026-06-28_223252 16 0.224
0.001
0.000
v2026-06-28_164026
v2026-06-28_223252
8 0.448
0.001
0.000
v2026-06-21_152640 18 0.053
0.001
0.000
v2026-06-14_153657
v2026-06-21_152640
23 0.186
0.002
0.179
v2026-06-14_153657 21 0.046
0.024
0.015
v2026-06-14_153657 9 0.085
0.032
0.001

Phase 3q · v2026-08-02_154011_phase3q Δ +0.000 vs prev train

Phase 3q blender. Trained 2026-08-02. Metric: Test Brier.

Lead Blend Best single Δ vs best
+24h 0.000 (0.000 val)
+48h 0.000 (0.000 val)
+72h 0.000 (0.000 val)
Verify history (13 runs)

Mon + Thu rolling Brier. Per-lead cells turn red when the rolling metric breaches the lead-specific drift threshold; check the verify report for the per-cell breakdown. Version column names the trained model — a fresh champion takes ~5-9d to show.

Run (UTC) Version N +24h+48h+72h+96h+120h
v2026-07-26_154305_phase3q 8 0.001
0.001
0.002
v2026-07-19_152830_phase3q 22 0.001
0.000
0.003
v2026-07-19_152830_phase3q 10 0.000
0.000
0.001
v2026-07-12_153204_phase3q 22 0.001
0.002
0.004
v2026-07-12_153204_phase3q 10 0.001
0.003
0.008
v2026-07-05_152333_phase3q 14 0.000
0.001
0.003
v2026-07-05_152333_phase3q 10 0.000
0.001
0.001
v2026-06-28_223429_phase3q 16 0.202
0.001
0.004
v2026-06-28_164125_phase3q
v2026-06-28_223429_phase3q
8 0.404
0.001
0.001
v2026-06-21_152916_phase3q 18 0.050
0.002
0.005
v2026-06-14_153851_phase3q
v2026-06-21_152916_phase3q
23 0.178
0.001
0.194
v2026-06-14_153851_phase3q 21 0.076
0.050
0.061
v2026-06-14_153851_phase3q 9 0.154
0.099
0.033

Dry window — Trengwainton — 5-hour

Phase 3q · v2026-08-02_154011_phase3q Δ +0.000 vs prev train

Phase 3q blender. Trained 2026-08-02. Metric: Test Brier.

Lead Blend Best single Δ vs best
+24h 0.000 (0.000 val)
+48h 0.000 (0.000 val)
+72h 0.000 (0.000 val)
Verify history (13 runs)

Mon + Thu rolling Brier. Per-lead cells turn red when the rolling metric breaches the lead-specific drift threshold; check the verify report for the per-cell breakdown. Version column names the trained model — a fresh champion takes ~5-9d to show.

Run (UTC) Version N +24h+48h+72h+96h+120h
v2026-07-26_154305_phase3q 8 0.005
0.005
0.007
v2026-07-19_152830_phase3q 22 0.003
0.002
0.009
v2026-07-19_152830_phase3q 10 0.003
0.002
0.007
v2026-07-12_153204_phase3q 22 0.004
0.008
0.014
v2026-07-12_153204_phase3q 10 0.006
0.014
0.026
v2026-07-05_152333_phase3q 14 0.001
0.004
0.012
v2026-07-05_152333_phase3q 10 0.001
0.004
0.005
v2026-06-28_223429_phase3q 16 0.159
0.005
0.015
v2026-06-28_164125_phase3q
v2026-06-28_223429_phase3q
8 0.315
0.003
0.006
v2026-06-21_152916_phase3q 18 0.049
0.009
0.017
v2026-06-14_153851_phase3q
v2026-06-21_152916_phase3q
23 0.165
0.005
0.197
v2026-06-14_153851_phase3q 21 0.231
0.193
0.101
v2026-06-14_153851_phase3q 9 0.288
0.295
0.142

Dry window — Trengwainton — 6-hour

Phase 3b · v2026-08-02_153831 Δ -0.004 vs prev train

53-feature LightGBM per-(station, window). Trained 2026-08-02. Metric: Test Brier.

Lead Blend Best single Δ vs best
+24h 0.097 has_dry_window_icon (0.186) -47.9%
+48h 0.092 has_dry_window_icon (0.214) -56.9%
+72h 0.102 has_dry_window_icon (0.171) -40.7%
Verify history (13 runs)

Mon + Thu rolling Brier. Per-lead cells turn red when the rolling metric breaches the lead-specific drift threshold; check the verify report for the per-cell breakdown. Version column names the trained model — a fresh champion takes ~5-9d to show.

Run (UTC) Version N +24h+48h+72h+96h+120h
v2026-07-26_154121 8 0.013
0.005
0.009
v2026-07-19_152707 22 0.011
0.005
0.006
v2026-07-19_152707 10 0.011
0.005
0.006
v2026-07-12_153030 22 0.023
0.006
0.011
v2026-07-12_153030 10 0.032
0.009
0.041
v2026-07-05_152159 14 0.010
0.004
0.007
v2026-07-05_152159 10 0.009
0.004
0.003
v2026-06-28_223313 16 0.150
0.006
0.005
v2026-06-28_164047
v2026-06-28_223313
8 0.284
0.006
0.004
v2026-06-21_152700 18 0.064
0.008
0.010
v2026-06-14_153711
v2026-06-21_152700
23 0.154
0.004
0.233
v2026-06-14_153711 21 0.209
0.157
0.149
v2026-06-14_153711 9 0.213
0.097
0.154

Phase 3q · v2026-08-02_154011_phase3q Δ +0.000 vs prev train

Phase 3q blender. Trained 2026-08-02. Metric: Test Brier.

Lead Blend Best single Δ vs best
+24h 0.000 (0.000 val)
+48h 0.000 (0.000 val)
+72h 0.000 (0.000 val)
Verify history (13 runs)

Mon + Thu rolling Brier. Per-lead cells turn red when the rolling metric breaches the lead-specific drift threshold; check the verify report for the per-cell breakdown. Version column names the trained model — a fresh champion takes ~5-9d to show.

Run (UTC) Version N +24h+48h+72h+96h+120h
v2026-07-26_154305_phase3q 8 0.012
0.013
0.018
v2026-07-19_152830_phase3q 22 0.009
0.007
0.025
v2026-07-19_152830_phase3q 10 0.009
0.006
0.018
v2026-07-12_153204_phase3q 22 0.010
0.019
0.033
v2026-07-12_153204_phase3q 10 0.015
0.029
0.052
v2026-07-05_152333_phase3q 14 0.005
0.011
0.027
v2026-07-05_152333_phase3q 10 0.004
0.012
0.014
v2026-06-28_223429_phase3q 16 0.127
0.013
0.033
v2026-06-28_164125_phase3q
v2026-06-28_223429_phase3q
8 0.244
0.010
0.016
v2026-06-21_152916_phase3q 18 0.050
0.022
0.040
v2026-06-14_153851_phase3q
v2026-06-21_152916_phase3q
23 0.143
0.015
0.184
v2026-06-14_153851_phase3q 21 0.212
0.198
0.100
v2026-06-14_153851_phase3q 9 0.289
0.358
0.406

Dry window — Lands End — 2-hour

Phase 3q · v2026-08-02_154012_phase3q Δ +0.000 vs prev train

Phase 3q blender. Trained 2026-08-02. Metric: Test Brier.

Lead Blend Best single Δ vs best
+24h 0.000 (0.000 val)
+48h 0.000 (0.000 val)
+72h 0.000 (0.000 val)
Verify history (no runs yet)

No verify rows yet for phase 3q. Next cycle: Mon/Thu 09:30 UTC, then 5d ERA5 latency.

Dry window — Lands End — 3-hour

Phase 3b · v2026-08-02_153858 Δ +0.000 vs prev train

53-feature LightGBM per-(station, window). Trained 2026-08-02. Metric: Test Brier.

Lead Blend Best single Δ vs best
+24h 0.027 has_dry_window_mf (0.093 val) -70.6%
+48h 0.028 has_dry_window_icon (0.174) -84.2%
+72h 0.030 has_dry_window_icon (0.152) -80.0%
Verify history (no runs yet)

No verify rows yet for phase 3b. Next cycle: Mon/Thu 09:30 UTC, then 5d ERA5 latency.

Phase 3q · v2026-08-02_154012_phase3q Δ +0.000 vs prev train

Phase 3q blender. Trained 2026-08-02. Metric: Test Brier.

Lead Blend Best single Δ vs best
+24h 0.000 (0.000 val)
+48h 0.000 (0.000 val)
+72h 0.000 (0.000 val)
Verify history (no runs yet)

No verify rows yet for phase 3q. Next cycle: Mon/Thu 09:30 UTC, then 5d ERA5 latency.

Dry window — Lands End — 4-hour

Phase 3b · v2026-08-02_153921 Δ +0.000 vs prev train

53-feature LightGBM per-(station, window). Trained 2026-08-02. Metric: Test Brier.

Lead Blend Best single Δ vs best
+24h 0.051 has_dry_window_mf (0.155 val) -67.1%
+48h 0.063 has_dry_window_icon (0.197) -67.9%
+72h 0.064 has_dry_window_icon (0.182) -64.9%
Verify history (no runs yet)

No verify rows yet for phase 3b. Next cycle: Mon/Thu 09:30 UTC, then 5d ERA5 latency.

Phase 3q · v2026-08-02_154012_phase3q Δ +0.000 vs prev train

Phase 3q blender. Trained 2026-08-02. Metric: Test Brier.

Lead Blend Best single Δ vs best
+24h 0.000 (0.000 val)
+48h 0.000 (0.000 val)
+72h 0.000 (0.000 val)
Verify history (no runs yet)

No verify rows yet for phase 3q. Next cycle: Mon/Thu 09:30 UTC, then 5d ERA5 latency.

Dry window — Lands End — 5-hour

Phase 3q · v2026-08-02_154012_phase3q Δ +0.000 vs prev train

Phase 3q blender. Trained 2026-08-02. Metric: Test Brier.

Lead Blend Best single Δ vs best
+24h 0.000 (0.000 val)
+48h 0.000 (0.000 val)
+72h 0.000 (0.000 val)
Verify history (no runs yet)

No verify rows yet for phase 3q. Next cycle: Mon/Thu 09:30 UTC, then 5d ERA5 latency.

Dry window — Lands End — 6-hour

Phase 3b · v2026-08-02_153944 Δ +0.000 vs prev train

53-feature LightGBM per-(station, window). Trained 2026-08-02. Metric: Test Brier.

Lead Blend Best single Δ vs best
+24h 0.103 has_dry_window_mf (0.304) -66.1%
+48h 0.115 has_dry_window_icon (0.250) -54.0%
+72h 0.126 has_dry_window_icon (0.189) -33.6%
Verify history (no runs yet)

No verify rows yet for phase 3b. Next cycle: Mon/Thu 09:30 UTC, then 5d ERA5 latency.

Phase 3q · v2026-08-02_154012_phase3q Δ +0.000 vs prev train

Phase 3q blender. Trained 2026-08-02. Metric: Test Brier.

Lead Blend Best single Δ vs best
+24h 0.000 (0.000 val)
+48h 0.000 (0.000 val)
+72h 0.000 (0.000 val)
Verify history (no runs yet)

No verify rows yet for phase 3q. Next cycle: Mon/Thu 09:30 UTC, then 5d ERA5 latency.