WeatherBlend

Multi-model forecast blending for Sennen, Cornwall

Temperature models

2b (lean) + 2c (rich). MAE °C, lower better. Δ vs best single NWP — negative = blend wins.

Temperature — Sennen Cove

Phase 2c · v2026-06-15_090237_phase2c Δ +0.000 vs prev train

Rich blender, 88 features (adds dew/RH/cloud/wind/pressure). Trained 2026-06-15. Metric: Test MAE (°C).

Lead Blend Best single Δ vs best
+24h 0.255 temp_ecmwf (0.284) -10.1%
+48h 0.303 temp_aifs (0.310) -2.3%
+72h 0.365 temp_aifs (0.382) -4.4%
+96h 0.423 temp_aifs (0.464) -8.9%
+120h 0.492 temp_aifs (0.479) +2.7%
Verify history (2 runs)

Mon + Thu rolling MAE (°C). 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
v2026-06-07_145248_phase2c 384 0.507
temp_ecmwf: 0.223
0.450
temp_ecmwf: 0.275
0.224
temp_ecmwf: 0.377
0.276
temp_mf: 0.400
v2026-06-07_145248_phase2c 48 0.588
temp_aifs: 0.092
By actual NWP forecast lead (6h buckets)

Same data grouped by ValidTime − freshest contributing NWP cycle (6h buckets) instead of trained-lead label. Reveals MAE structure within a trained bucket once predict spread to hourly outputs (2026-05-04+). Buckets start at the trained lead — earlier figures measured from the cron-fire time, which made offset-day models look like sub-lead forecasts.

Run (UTC) 24-29h30-35h36-41h42-47h48-53h54-59h60-65h66-71h72-77h78-83h84-89h90-95h96-101h102-107h
0.5060.4600.5610.6120.4790.4070.3040.4810.2110.2870.2410.2200.2540.520
0.5820.656

Phase 2b · v2026-06-14_153323 Δ -0.009 vs prev train

Lean blender, 13 features. Trained 2026-06-14. Metric: Test MAE (°C).

Lead Blend Best single Δ vs best
+24h 0.264 temp_ecmwf (0.284) -7.0%
+48h 0.302 temp_aifs (0.310) -2.3%
+72h 0.367 temp_aifs (0.382) -3.9%
+96h 0.440 temp_aifs (0.464) -5.3%
+120h 0.498 temp_aifs (0.479) +4.0%
Verify history (2 runs)

Mon + Thu rolling MAE (°C). 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
v2026-06-07_145236 384 0.313
temp_ecmwf: 0.223
0.205
temp_ecmwf: 0.275
0.230
temp_ecmwf: 0.377
0.207
temp_mf: 0.400
v2026-06-07_145236 48 0.403
temp_aifs: 0.092
By actual NWP forecast lead (6h buckets)

Same data grouped by ValidTime − freshest contributing NWP cycle (6h buckets) instead of trained-lead label. Reveals MAE structure within a trained bucket once predict spread to hourly outputs (2026-05-04+). Buckets start at the trained lead — earlier figures measured from the cron-fire time, which made offset-day models look like sub-lead forecasts.

Run (UTC) 24-29h30-35h36-41h42-47h48-53h54-59h60-65h66-71h72-77h78-83h84-89h90-95h96-101h102-107h
0.3250.2590.3070.3770.2090.1890.1920.2400.1880.3100.4580.4360.1960.326
0.4000.438