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 · v2026-08-02_153504_phase2c Δ -0.003 vs prev train

Rich blender, 88 features (adds dew/RH/cloud/wind/pressure). Trained 2026-08-02. Metric: Test MAE (°C).

Lead Blend Best single Δ vs best
+24h 0.382 temp_ecmwf (0.375) +2.0%
+48h 0.398 temp_aifs (0.349) +14.0%
+72h 0.421 temp_aifs (0.389) +8.1%
+96h 0.454 temp_aifs (0.429) +5.9%
+120h 0.547 temp_aifs (0.487) +12.5%
Verify history (15 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 +12h+24h+48h+72h+96h+120h
v2026-07-19_152422_phase2c
v2026-07-26_153754_phase2c
624 2.619
temp_aifs: 0.133
2.567
temp_aifs: 0.154
2.398
temp_ecmwf: 0.317
2.406
temp_ecmwf: 0.308
2.435
temp_aifs: 0.270
2.269
temp_aifs: 0.264
v2026-07-12_152714_phase2c
v2026-07-19_152422_phase2c
624 2.192
temp_aifs: 0.367
2.116
temp_ecmwf: 0.390
2.086
temp_ecmwf: 0.499
2.103
temp_ecmwf: 0.424
2.249
temp_aifs: 0.346
1.875
temp_aifs: 0.625
v2026-07-12_152714_phase2c
v2026-07-19_152422_phase2c
624 2.390
temp_aifs: 0.337
2.396
temp_aifs: 0.313
2.719
temp_aifs: 0.372
2.619
temp_aifs: 0.338
2.377
temp_aifs: 0.348
2.072
temp_aifs: 0.493
v2026-07-05_151842_phase2c
v2026-07-12_152714_phase2c
624 3.515
temp_aifs: 0.373
3.353
temp_aifs: 0.351
3.249
temp_aifs: 0.313
3.266
temp_aifs: 0.329
3.123
temp_aifs: 0.294
3.133
temp_ecmwf: 0.545
v2026-07-05_151842_phase2c
v2026-07-12_152714_phase2c
624 3.927
temp_ecmwf: 0.361
3.606
temp_aifs: 0.302
2.929
temp_aifs: 0.637
3.351
temp_aifs: 0.504
3.792
temp_aifs: 0.505
3.506
temp_ecmwf: 0.635
v2026-06-28_153524_phase2c
v2026-07-05_151842_phase2c
624 1.235
temp_ecmwf: 0.948
1.214
temp_ecmwf: 0.982
1.254
temp_ecmwf: 0.975
1.448
temp_aifs: 0.635
1.574
temp_aifs: 0.267
0.478
temp_aifs: 0.823
v2026-06-28_153524_phase2c
v2026-07-05_151842_phase2c
624 1.100
temp_ecmwf: 0.747
0.941
temp_mf: 0.600
1.055
temp_aifs: 0.586
0.719
temp_aifs: 0.610
0.348
temp_aifs: 0.696
0.342
temp_aifs: 0.880
v2026-06-21_152422_phase2c
v2026-06-28_153524_phase2c
624 1.591
temp_ecmwf: 0.180
1.509
temp_ecmwf: 0.208
1.479
temp_ecmwf: 0.215
1.106
temp_aifs: 0.385
0.757
temp_ecmwf: 0.204
1.018
temp_ecmwf: 0.551
v2026-06-21_152422_phase2c
v2026-06-28_153524_phase2c
624 1.801
temp_aifs: 0.101
1.797
temp_aifs: 0.071
2.132
temp_aifs: 0.621
2.187
temp_ecmwf: 0.480
1.807
temp_ecmwf: 0.428
1.409
temp_ecmwf: 0.536
v2026-06-15_090237_phase2c
v2026-06-21_152422_phase2c
600 2.481
temp_aifs: 0.510
2.539
temp_aifs: 0.465
2.590
temp_aifs: 0.950
3.559
temp_aifs: 0.767
4.026
temp_aifs: 0.783
2.159
temp_ecmwf: 1.008
v2026-06-15_090237_phase2c
v2026-06-21_152422_phase2c
600 2.122
temp_aifs: 0.430
2.267
temp_ecmwf: 0.433
1.330
temp_aifs: 0.480
1.083
temp_aifs: 0.438
1.200
temp_aifs: 0.370
1.323
temp_ecmwf: 0.923
v2026-06-07_145248_phase2c
v2026-06-14_153341_phase2c
v2026-06-15_090237_phase2c
720 0.858
temp_aifs: 0.349
0.828
temp_aifs: 0.349
0.878
temp_aifs: 0.453
0.473
temp_aifs: 0.443
0.520
temp_aifs: 0.383
0.414
temp_aifs: 0.432
v2026-06-07_145248_phase2c
v2026-06-14_153341_phase2c
v2026-06-15_090237_phase2c
720 0.961
temp_ecmwf: 0.086
0.989
temp_ecmwf: 0.242
0.551
temp_aifs: 0.348
0.446
temp_aifs: 0.416
0.443
temp_aifs: 0.403
0.380
temp_aifs: 0.435
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) 12-17h18-23h24-29h30-35h36-41h42-47h48-53h54-59h60-65h66-71h72-77h78-83h84-89h90-95h96-101h102-107h108-113h114-119h120-125h126-131h132-137h138-143h144-149h
2.6252.4532.5672.3512.3622.2372.3482.4892.5812.4842.3342.5772.6142.5152.4062.5132.5232.4392.2492.2722.3862.372
2.1932.1512.0852.2172.2232.0181.9922.2782.4832.3181.9052.6892.8732.7152.2192.5812.0911.9981.8561.9431.9091.858
2.3932.2912.3992.3662.9092.8022.6912.8102.7912.6812.6372.6402.4982.5032.3992.3732.2732.2312.1002.0491.9451.933
3.5233.3163.3373.3653.4663.3483.1983.3853.4333.2883.2913.2503.0493.1793.1213.1424.0633.8723.0553.2683.4163.2961.695
3.9453.2863.6433.1952.8952.7922.7643.2483.4463.3383.1353.7624.0403.9363.5604.2834.5394.2963.3843.7263.9743.8621.695
1.2241.4991.2161.2301.1621.2291.1881.3911.4741.5451.2991.7932.0872.2351.4962.4290.7130.7600.4030.5600.6690.6961.695
1.0672.2800.9081.3021.0110.9121.0640.9980.9931.1030.7450.6870.6910.6410.3190.3600.4700.5230.3190.3970.3840.438
1.5971.4541.5551.4161.3861.3391.5471.3331.2491.3131.1950.9510.7800.5270.7571.3411.2831.1281.0770.9450.8550.756
1.8011.7821.7972.3262.2642.2092.0732.2762.2942.1362.1942.2142.1871.9771.8711.6681.6841.4761.3971.4821.4271.287
2.4782.5692.4992.6782.5602.6772.3833.0183.2073.5093.3924.0004.1964.1564.0262.1902.3022.6231.9862.4272.6162.904
2.1202.1942.2671.1621.1211.2181.2751.4991.4341.3231.0081.2021.2971.4521.0971.3791.4821.7461.1391.6361.8602.419
0.8650.6690.8750.7360.6840.6410.9170.8060.7380.6250.5020.3620.3180.4270.5200.4600.3660.3930.4070.3870.3820.701
0.9611.0210.9890.5610.6260.4530.5440.5630.5470.6150.4020.4870.5890.5900.4490.4710.3740.4340.3820.3610.3120.646
0.5060.4600.5610.6120.4790.4070.3040.4810.2110.2870.2410.2200.2540.520
0.5820.656

Phase 2b · v2026-08-02_153440 Δ -0.003 vs prev train

Lean blender, 13 features. Trained 2026-08-02. Metric: Test MAE (°C).

Lead Blend Best single Δ vs best
+24h 0.391 temp_ecmwf (0.375) +4.3%
+48h 0.391 temp_aifs (0.349) +12.2%
+72h 0.403 temp_aifs (0.389) +3.5%
+96h 0.447 temp_aifs (0.429) +4.2%
+120h 0.524 temp_aifs (0.487) +7.7%
Verify history (15 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
v2026-07-19_152405
v2026-07-26_153731
624 2.363
temp_aifs: 0.154
1.730
temp_ecmwf: 0.317
1.702
temp_ecmwf: 0.308
2.049
temp_aifs: 0.270
1.511
temp_aifs: 0.264
v2026-07-12_152651
v2026-07-19_152405
624 2.103
temp_ecmwf: 0.390
1.486
temp_ecmwf: 0.499
1.407
temp_ecmwf: 0.424
1.957
temp_aifs: 0.346
0.993
temp_aifs: 0.625
v2026-07-12_152651
v2026-07-19_152405
624 2.409
temp_aifs: 0.313
2.318
temp_aifs: 0.372
1.851
temp_aifs: 0.338
2.242
temp_aifs: 0.348
1.123
temp_aifs: 0.493
v2026-07-05_151819
v2026-07-12_152651
624 3.034
temp_aifs: 0.351
2.835
temp_aifs: 0.313
2.455
temp_aifs: 0.329
2.907
temp_aifs: 0.294
2.656
temp_ecmwf: 0.545
v2026-07-05_151819
v2026-07-12_152651
624 3.312
temp_aifs: 0.302
2.509
temp_aifs: 0.637
2.536
temp_aifs: 0.504
3.656
temp_aifs: 0.505
3.054
temp_ecmwf: 0.635
v2026-06-28_153504
v2026-07-05_151819
624 1.143
temp_ecmwf: 0.982
0.768
temp_ecmwf: 0.975
0.624
temp_aifs: 0.635
1.493
temp_aifs: 0.267
0.436
temp_aifs: 0.823
v2026-06-28_153504
v2026-07-05_151819
624 0.831
temp_mf: 0.600
0.619
temp_aifs: 0.586
0.397
temp_aifs: 0.610
0.344
temp_aifs: 0.696
0.472
temp_aifs: 0.880
v2026-06-21_152403
v2026-06-28_153504
624 1.440
temp_ecmwf: 0.208
1.043
temp_ecmwf: 0.215
0.491
temp_aifs: 0.385
0.728
temp_ecmwf: 0.204
0.811
temp_ecmwf: 0.551
v2026-06-21_152403
v2026-06-28_153504
624 1.618
temp_aifs: 0.071
1.814
temp_aifs: 0.621
1.490
temp_ecmwf: 0.480
1.652
temp_ecmwf: 0.428
1.168
temp_ecmwf: 0.536
v2026-06-14_153323
v2026-06-21_152403
648 2.585
temp_aifs: 0.465
2.274
temp_aifs: 0.950
2.889
temp_aifs: 0.767
3.792
temp_aifs: 0.783
1.628
temp_ecmwf: 0.961
v2026-06-14_153323
v2026-06-21_152403
648 2.247
temp_ecmwf: 0.433
0.863
temp_aifs: 0.462
0.619
temp_aifs: 0.438
1.003
temp_aifs: 0.370
0.902
temp_ecmwf: 0.840
v2026-06-07_145236
v2026-06-14_153323
720 0.704
temp_aifs: 0.335
0.520
temp_aifs: 0.406
0.234
temp_aifs: 0.445
0.400
temp_aifs: 0.383
0.438
temp_aifs: 0.432
v2026-06-07_145236
v2026-06-14_153323
720 0.809
temp_ecmwf: 0.242
0.339
temp_aifs: 0.348
0.381
temp_aifs: 0.416
0.468
temp_aifs: 0.403
0.430
temp_aifs: 0.435
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) 24-29h30-35h36-41h42-47h48-53h54-59h60-65h66-71h72-77h78-83h84-89h90-95h96-101h102-107h108-113h114-119h120-125h126-131h132-137h138-143h144-149h
2.3632.3552.3622.1931.6911.8101.8711.7481.6431.8581.8681.7372.0312.1262.0881.9551.5061.5091.5531.524
2.0642.2432.2361.9591.4011.6381.8721.7121.2321.9272.1041.8861.9302.2631.9441.7900.9941.0280.9560.908
2.4122.3752.5142.3222.2992.4082.3482.2031.8711.8791.7241.6812.2632.2572.1352.0311.1591.0900.9520.980
3.0363.0583.0762.7912.8072.9292.9392.7722.4922.4072.2182.2702.9102.8663.8653.6312.5952.7592.8842.7741.274
3.3582.8023.0652.9522.3442.8303.0242.9122.3332.9473.1733.0133.4324.1394.3734.1222.9413.2643.4953.3281.274
1.1411.1571.1131.1970.7050.8910.9721.0870.5030.9491.0771.1871.4322.1720.6580.7140.3730.5320.5910.5511.274
0.8031.1410.9180.8240.6390.5350.5390.6730.4020.3970.4180.3960.3130.3750.4650.4930.4140.6590.5720.551
1.4841.3541.3111.2851.1090.9160.7830.9030.5690.3470.1480.1590.7281.2271.1240.9630.8470.7690.7010.654
1.6182.2942.2292.1231.7651.9351.9571.7921.5061.5001.4581.2711.7091.5461.5171.3571.1611.2261.1621.075
2.5392.7212.6392.7562.0832.6412.8573.1222.7583.1733.4003.4933.7913.8152.0692.3111.5281.8371.8171.983
2.2681.7750.9170.9480.8310.9920.9120.7830.5740.7060.7380.7660.9021.1901.2431.4800.8071.1031.1131.336
0.7430.6340.5900.5470.5720.4170.3510.3360.2680.1560.1060.0960.4000.4570.3690.4140.4510.3880.3760.601
0.8090.3970.4040.3390.3370.3540.3240.3480.3680.3620.4770.4000.4810.4690.3850.4640.4760.3150.2830.501
0.3250.2590.3070.3770.2090.1890.1920.2400.1880.3100.4580.4360.1960.326
0.4000.438