Permanent link · station × model × lead day
Minneapolis-St Paul · GraphCast (GFS init) · lead day 3
KMSP · graphcast_gfs · lead 3
Target day = initialization date + 3 days. This page is a permanent link: the numbers change as history accumulates, the address and the methodology version do not. Scores use the latest model version segment, GRAP_v100, starting 2026-01-01 — days before that ran a different version of the model and are excluded rather than silently pooled.
Every number for this combination28 rows
All four windows × both initializations × both interpolations × every variable. Rows are grouped by variable and the specification is collapsed into one monospace key (window · init · method) so the eye travels straight to the numbers. Two windows that cover exactly the same days — while the record is shorter than a year, 365d and all are the same set — share one row whose key names both, rather than printing the same numbers twice. The row matching the default view is highlighted. Errors are computed in °C and shown in °F; CI is a 95 % moving-block bootstrap interval on this group's own days, and reads — when it could not be computed (fewer than 28 scored days or fewer than 4 blocks). The ±1 °F interval is a Wilson score interval, not a bootstrap.
Skill divides by the persistence MAE on the days both have a value; that denominator and the size of the intersection are printed under the number, so the score can be checked rather than taken on trust. MAE debiased and skill, debiased remove a bias estimated on the 30 scored days before each day and applied forward — out of sample, never fitted on the day being scored; n debiased is how many days had enough history to qualify. n QC next to n counts days whose observation carries a QC flag.
| Window · init · method | n | MAE °F | MAE debiased | Bias °F | RMSE °F | ±1 °F | ±2 °F | ±3 °F | Skill | Skill, debiased (out-of-sample) | Period |
|---|---|---|---|---|---|---|---|---|---|---|---|
| instantaneous temperature t2 | |||||||||||
| 365d · all · 00Z · bilinear | 33 | 4.44[3.39, 5.54] | 3.66 | −3.00[−4.79, −1.15] | 5.49[4.13, 6.56] | 15%[7%, 31%] | 27% | 39% | +0.63[+0.49, +0.72]vs 11.94 (n=33) | +0.67n=18 | 2026-01-04 → 2026-02-06 |
| 365d · all · 00Z · nearest | 33 | 4.45[3.46, 5.47] | 3.55 | −3.06[−4.77, −1.25] | 5.47[4.19, 6.50] | 14%[6%, 29%] | 27% | 38% | +0.63[+0.50, +0.72]vs 11.94 (n=33) | +0.68n=18 | 2026-01-04 → 2026-02-06 |
| 365d · all · 12Z · bilinear | 34 | 4.40[3.39, 5.49] | 3.78 | −2.97[−4.67, −1.00] | 5.61[4.28, 6.84] | 13%[6%, 28%] | 34% | 41% | +0.63[+0.52, +0.72]vs 12.02 (n=34) | +0.68n=19 | 2026-01-04 → 2026-02-06 |
| 365d · all · 12Z · nearest | 34 | 4.39[3.45, 5.47] | 3.64 | −3.01[−4.65, −1.15] | 5.57[4.32, 6.81] | 12%[5%, 28%] | 32% | 40% | +0.63[+0.53, +0.72]vs 12.02 (n=34) | +0.70n=19 | 2026-01-04 → 2026-02-06 |
| instantaneous temperature, 00Z t2_00z | |||||||||||
| 365d · all · 00Z · bilinear | 33 | 4.81[3.52, 6.22] | 4.39 | −3.22[−5.09, −1.54] | 5.68[4.04, 7.01] | 9%[3%, 24%] | 18% | 30% | +0.57[+0.45, +0.69]vs 11.30 (n=33) | +0.62n=18 | 2026-01-04 → 2026-02-06 |
| 365d · all · 00Z · nearest | 33 | 4.75[3.55, 6.11] | 4.22 | −3.23[−4.97, −1.65] | 5.63[4.12, 6.91] | 9%[3%, 24%] | 18% | 30% | +0.58[+0.46, +0.69]vs 11.30 (n=33) | +0.63n=18 | 2026-01-04 → 2026-02-06 |
| 365d · all · 12Z · bilinear | 34 | 4.69[3.36, 6.19] | 4.19 | −3.41[−5.27, −1.83] | 5.65[3.92, 7.11] | 6%[2%, 19%] | 24% | 26% | +0.59[+0.46, +0.72]vs 11.56 (n=34) | +0.63n=19 | 2026-01-04 → 2026-02-06 |
| 365d · all · 12Z · nearest | 34 | 4.66[3.43, 6.06] | 3.98 | −3.42[−5.13, −1.95] | 5.56[3.97, 6.95] | 6%[2%, 19%] | 26% | 26% | +0.60[+0.47, +0.71]vs 11.56 (n=34) | +0.65n=19 | 2026-01-04 → 2026-02-06 |
| instantaneous temperature, 06Z t2_06z | |||||||||||
| 365d · all · 00Z · bilinear | 33 | 4.66[3.40, 5.80] | 4.67 | −3.41[−5.03, −1.87] | 5.62[4.19, 6.77] | 12%[5%, 27%] | 24% | 39% | +0.63[+0.45, +0.75]vs 12.45 (n=33) | +0.64n=18 | 2026-01-04 → 2026-02-06 |
| 365d · all · 00Z · nearest | 33 | 4.72[3.54, 5.77] | 4.51 | −3.49[−5.03, −1.99] | 5.62[4.32, 6.68] | 12%[5%, 27%] | 21% | 30% | +0.62[+0.44, +0.74]vs 12.45 (n=33) | +0.65n=18 | 2026-01-04 → 2026-02-06 |
| 365d · all · 12Z · bilinear | 34 | 4.49[3.18, 5.74] | 4.64 | −3.47[−5.00, −1.81] | 5.74[4.15, 7.05] | 12%[5%, 27%] | 35% | 41% | +0.64[+0.48, +0.77]vs 12.47 (n=34) | +0.67n=19 | 2026-01-04 → 2026-02-06 |
| 365d · all · 12Z · nearest | 34 | 4.53[3.28, 5.73] | 4.57 | −3.55[−4.99, −1.93] | 5.75[4.24, 7.06] | 12%[5%, 27%] | 32% | 41% | +0.64[+0.47, +0.77]vs 12.47 (n=34) | +0.68n=19 | 2026-01-04 → 2026-02-06 |
| instantaneous temperature, 12Z t2_12z | |||||||||||
| 365d · all · 00Z · bilinear | 33 | 5.09[3.66, 6.43] | 4.77 | −3.87[−6.29, −1.04] | 6.38[4.66, 7.69] | 18%[9%, 34%] | 24% | 33% | +0.59[+0.45, +0.68]vs 12.45 (n=33) | +0.67n=18 | 2026-01-04 → 2026-02-06 |
| 365d · all · 00Z · nearest | 33 | 5.13[3.80, 6.43] | 4.63 | −3.96[−6.30, −1.17] | 6.38[4.71, 7.66] | 15%[7%, 31%] | 24% | 33% | +0.59[+0.45, +0.67]vs 12.45 (n=33) | +0.68n=18 | 2026-01-04 → 2026-02-06 |
| 365d · all · 12Z · bilinear | 34 | 4.98[3.67, 6.52] | 4.82 | −3.66[−6.07, −0.69] | 6.62[4.94, 8.29] | 15%[6%, 30%] | 32% | 47% | +0.60[+0.51, +0.67]vs 12.47 (n=34) | +0.69n=19 | 2026-01-04 → 2026-02-06 |
| 365d · all · 12Z · nearest | 34 | 5.00[3.72, 6.53] | 4.66 | −3.73[−6.14, −0.88] | 6.59[4.92, 8.25] | 15%[6%, 30%] | 26% | 44% | +0.60[+0.52, +0.66]vs 12.47 (n=34) | +0.70n=19 | 2026-01-04 → 2026-02-06 |
| instantaneous temperature, 18Z t2_18z | |||||||||||
| 365d · all · 00Z · bilinear | 33 | 3.21[2.49, 3.95] | 3.30 | −1.50[−3.40, 0.55] | 3.98[3.12, 4.77] | 21%[11%, 38%] | 42% | 55% | +0.72[+0.61, +0.79]vs 11.55 (n=33) | +0.69n=18 | 2026-01-04 → 2026-02-06 |
| 365d · all · 00Z · nearest | 33 | 3.18[2.51, 3.89] | 3.17 | −1.54[−3.35, 0.47] | 3.94[3.14, 4.71] | 18%[9%, 34%] | 42% | 58% | +0.72[+0.62, +0.79]vs 11.55 (n=33) | +0.70n=18 | 2026-01-04 → 2026-02-06 |
| 365d · all · 12Z · bilinear | 34 | 3.42[2.70, 4.13] | 3.61 | −1.33[−3.10, 0.96] | 4.13[3.34, 4.81] | 21%[10%, 37%] | 44% | 50% | +0.71[+0.60, +0.78]vs 11.59 (n=34) | +0.68n=19 | 2026-01-04 → 2026-02-06 |
| 365d · all · 12Z · nearest | 34 | 3.37[2.70, 4.05] | 3.44 | −1.35[−3.06, 0.86] | 4.06[3.35, 4.71] | 18%[8%, 34%] | 44% | 50% | +0.71[+0.61, +0.78]vs 11.59 (n=34) | +0.70n=19 | 2026-01-04 → 2026-02-06 |
| sampled daily maximum tmax_s | |||||||||||
| 365d · all · 00Z · bilinear | 33 | 3.57[2.60, 4.57] | 2.84 | −3.05[−4.38, −1.72] | 4.56[3.19, 5.66] | 18%[9%, 34%] | 39% | 48% | +0.68[+0.55, +0.76]vs 11.09 (n=33) | +0.76n=18 | 2026-01-04 → 2026-02-06 |
| 365d · all · 00Z · nearest | 33 | 3.59[2.65, 4.55] | 2.68 | −3.11[−4.38, −1.78] | 4.57[3.28, 5.63] | 21%[11%, 38%] | 36% | 48% | +0.68[+0.56, +0.76]vs 11.09 (n=33) | +0.77n=18 | 2026-01-04 → 2026-02-06 |
| 365d · all · 12Z · bilinear | 34 | 3.81[2.57, 5.04] | 3.28 | −2.95[−4.52, −1.42] | 4.80[3.24, 6.08] | 21%[10%, 37%] | 38% | 38% | +0.66[+0.54, +0.76]vs 11.35 (n=34) | +0.74n=19 | 2026-01-04 → 2026-02-06 |
| 365d · all · 12Z · nearest | 34 | 3.81[2.66, 4.98] | 3.17 | −2.99[−4.49, −1.52] | 4.76[3.29, 6.03] | 21%[10%, 37%] | 35% | 38% | +0.66[+0.55, +0.76]vs 11.35 (n=34) | +0.75n=19 | 2026-01-04 → 2026-02-06 |
| sampled daily minimum tmin_s | |||||||||||
| 365d · all · 00Z · bilinear | 33 | 4.76[3.17, 6.34] | 5.51 | −3.16[−5.42, −1.16] | 6.03[3.97, 7.61] | 21%[11%, 38%] | 27% | 36% | +0.56[+0.37, +0.71]vs 10.91 (n=33) | +0.51n=18 | 2026-01-04 → 2026-02-06 |
| 365d · all · 00Z · nearest | 33 | 4.79[3.24, 6.34] | 5.37 | −3.23[−5.39, −1.27] | 6.02[4.06, 7.55] | 15%[7%, 31%] | 27% | 36% | +0.56[+0.37, +0.71]vs 10.91 (n=33) | +0.52n=18 | 2026-01-04 → 2026-02-06 |
| 365d · all · 12Z · bilinear | 34 | 4.78[3.25, 6.46] | 5.42 | −3.09[−5.08, −0.87] | 6.37[4.21, 8.21] | 15%[6%, 30%] | 35% | 53% | +0.56[+0.39, +0.71]vs 10.97 (n=34) | +0.53n=19 | 2026-01-04 → 2026-02-06 |
| 365d · all · 12Z · nearest | 34 | 4.78[3.30, 6.45] | 5.27 | −3.15[−5.06, −1.00] | 6.36[4.25, 8.18] | 15%[6%, 30%] | 35% | 53% | +0.56[+0.39, +0.71]vs 10.97 (n=34) | +0.55n=19 | 2026-01-04 → 2026-02-06 |
+ model too warm − model too cold bias interval includes zero ★ lowest MAE no Holm-corrected comparison is available in this build, so no model is marked as distinguishable from the leadern < 30 greyed and unranked CI is a 95 % moving-block bootstrap interval on the group's own days, and reads — when it could not be computed skill reads — when fewer than 10 days are common to the model and the baseline: the ratio of two means over a handful of shared days is not a number worth printing
Against the NWS daily extremes (secondary)
This comparison scores four 6-hourly forecast samples against the true daily extreme, so it includes a sampling penalty. The size of that penalty depends on each model's own diurnal amplitude, which is one of the things this site measures, so these numbers are published for operational relevance and are never used for ranking.
| Window · init · method | n | MAE °F | Bias °F | RMSE °F | ±1 °F | ±3 °F | Period |
|---|---|---|---|---|---|---|---|
| daily maximum vs NWS CLI tmax_cli | |||||||
| 365d · all · 00Z · bilinear | 33 | 5.38[4.20, 6.82] | −5.05[−6.76, −3.59] | 6.46 | 12%[5%, 27%] | 36% | 2026-01-04 → 2026-02-06 |
| 365d · all · 00Z · nearest | 33 | 5.43[4.26, 6.82] | −5.11[−6.71, −3.68] | 6.48 | 9%[3%, 24%] | 33% | 2026-01-04 → 2026-02-06 |
| 365d · all · 12Z · bilinear | 34 | 5.33[3.89, 6.92] | −5.07[−6.76, −3.43] | 6.52 | 15%[6%, 30%] | 29% | 2026-01-04 → 2026-02-06 |
| 365d · all · 12Z · nearest | 34 | 5.34[3.98, 6.85] | −5.11[−6.73, −3.54] | 6.50 | 15%[6%, 30%] | 29% | 2026-01-04 → 2026-02-06 |
| daily minimum vs NWS CLI tmin_cli | |||||||
| 365d · all · 00Z · bilinear | 33 | 5.30[3.94, 6.66] | −0.46[−3.23, 1.92] | 6.79 | 9%[3%, 24%] | 39% | 2026-01-04 → 2026-02-06 |
| 365d · all · 00Z · nearest | 33 | 5.27[3.93, 6.65] | −0.53[−3.22, 1.78] | 6.74 | 9%[3%, 24%] | 36% | 2026-01-04 → 2026-02-06 |
| 365d · all · 12Z · bilinear | 34 | 5.35[3.99, 7.02] | −0.36[−2.72, 2.04] | 6.99 | 21%[10%, 37%] | 35% | 2026-01-04 → 2026-02-06 |
| 365d · all · 12Z · nearest | 34 | 5.31[3.97, 6.96] | −0.42[−2.72, 1.91] | 6.93 | 18%[8%, 34%] | 38% | 2026-01-04 → 2026-02-06 |
Daily error, last 90 days
Forecast − observed, °F, 00Z bilinear instantaneous temperature. The dashed line is zero; the thick line is a centred 7-day mean, drawn only where the whole window is present. Gaps are days with no score, never interpolation.
Daily forecast minus observed error in °F for GraphCast (GFS init) at KMSP, lead day 3, 00Z bilinear t2: no scored days yet.
Source: CastCheck · NWS ASOS routine METAR · CSV of these values. Hover a point for its date and value.
Error distribution
All scored days for 00Z bilinear instantaneous temperature, 20 equal bins. The dashed marks are the 50th and 90th percentiles of the absolute error, mirrored on both sides.
Source: CastCheck · bins are equal width over the plotted range. Hover a bar for its bin and count.
The same series as a table (0 days, newest first) · download CSV
| Climatological day | Error °F |
|---|
Bin counts as a table
| From °F | To °F | Days |
|---|---|---|
| −50.00 | −45.00 | 0 |
| −45.00 | −40.00 | 0 |
| −40.00 | −35.00 | 0 |
| −35.00 | −30.00 | 0 |
| −30.00 | −25.00 | 1 |
| −25.00 | −20.00 | 2 |
| −20.00 | −15.00 | 4 |
| −15.00 | −10.00 | 21 |
| −10.00 | −5.00 | 96 |
| −5.00 | +0.00 | 156 |
| +0.00 | +5.00 | 80 |
| +5.00 | +10.00 | 10 |
| +10.00 | +15.00 | 2 |
| +15.00 | +20.00 | 0 |
| +20.00 | +25.00 | 0 |
| +25.00 | +30.00 | 0 |
| +30.00 | +35.00 | 0 |
| +35.00 | +40.00 | 0 |
| +40.00 | +45.00 | 0 |
| +45.00 | +50.00 | 0 |
Month by month
00Z bilinear instantaneous temperature, newest first.
| Month | n | MAE °F | Bias °F |
|---|---|---|---|
| 2026-04 | 24 | 3.34 | -3.34 |
| 2026-03 | 124 | 5.52 | -4.63 |
| 2026-02 | 112 | 3.97 | -2.01 |
| 2026-01 | 112 | 4.54 | -2.80 |
Availability
The scored period behind each model at this station, on a shared scale.
| Model | Scored period | Days | Coverage |
|---|---|---|---|
| GraphCast (GFS init) | 2026-01-04 → 2026-02-06 | 33 | |
| GraphCast (GFS init) | 2026-01-04 → 2026-02-06 | 33 | |
| GraphCast (GFS init) | 2026-01-04 → 2026-02-06 | 34 | |
| GraphCast (GFS init) | 2026-01-04 → 2026-02-06 | 34 |
Cite this page
The address is stable; the numbers behind it are not. Always cite with an access date and the data-through date.
CastCheck, KMSP graphcast_gfs lead 3, methodology v0.3.1, accessed 2026-08-31, https://castcheck.zifanzhang.com/station/KMSP/model/graphcast_gfs/lead/3/
CastCheck (2026). Station-level verification of raw weather-model 2 m temperature forecasts: KMSP (Minneapolis-St Paul), model graphcast_gfs, lead day 3. Methodology version 0.3.1, schema version 0.3, data through 2026-08-30. Accessed 2026-08-31. https://castcheck.zifanzhang.com/station/KMSP/model/graphcast_gfs/lead/3/
2026-08-31 (data through 2026-08-30; page built )