Raw model output on the native 0.25° grid — no MOS, no bias correction, no post-processing. Scores understate operational forecast quality. Read the fairness statement →

CastCheck methodology v0.2 · data through 2026-08-30

KATL — Atlanta Hartsfield

33.6403, -84.4269 · 313 m· standard offset UTC-5 h · truth product CLIATL

These pages are always bilinear interpolation, the headline method. The nearest-node variant, and every other window/initialization/interpolation/variable combination, is published on each permanent link below — follow any number.

Daily maximum temperature — MAE °F

the last 90 days, 12Z, bilinear. MAE with the bias and n underneath; every cell links to its permanent page. The sparkline shows the same model across lead days 1–9 on a scale shared within this table.
Modeld0d1d2d3d4d5d6d7d8d9lead 1–9
ECMWF AIFS Singleaifs_single 4.33-4.33 · n=29 4.38-4.38 · n=28 4.34-4.34 · n=27 4.49-4.49 · n=26 4.40-4.40 · n=24 4.79-4.70 · n=24 5.22-4.97 · n=23 5.22-4.97 · n=22 5.63-5.63 · n=21 ECMWF AIFS Single MAE by lead day 1 to 9, tmax
Aurora (GFS init)aurora_gfs 6.56-6.56 · n=2 7.31-7.31 · n=1 Aurora (GFS init) MAE by lead day 1 to 9, tmax
Aurora (IFS init)aurora_ifs 6.57-6.57 · n=2 7.61-7.61 · n=1 Aurora (IFS init) MAE by lead day 1 to 9, tmax
FourCastNet v2 (GFS init)fourcastnet_gfs 2.33-2.33 · n=2 4.07-4.07 · n=1 FourCastNet v2 (GFS init) MAE by lead day 1 to 9, tmax
FourCastNet v2 (IFS init)fourcastnet_ifs 2.04-2.04 · n=2 3.84-3.84 · n=1 FourCastNet v2 (IFS init) MAE by lead day 1 to 9, tmax
NCEP GFSgfs 2.19+0.27 · n=29 3.16-0.92 · n=28 2.94-0.45 · n=27 3.03+0.11 · n=26 3.11-0.29 · n=25 4.97-2.20 · n=24 2.74-0.66 · n=23 4.43-2.04 · n=22 5.27-1.46 · n=21 NCEP GFS MAE by lead day 1 to 9, tmax
GraphCast (GFS init)graphcast_gfs 4.97-4.97 · n=2 6.04-6.04 · n=1 GraphCast (GFS init) MAE by lead day 1 to 9, tmax
GraphCast (IFS init)graphcast_ifs 6.61-6.61 · n=27 6.40-6.40 · n=26 5.95-5.95 · n=25 5.47-5.47 · n=25 5.55-5.55 · n=25 5.36-5.15 · n=24 5.51-5.32 · n=23 5.05-4.76 · n=22 5.69-4.92 · n=21 GraphCast (IFS init) MAE by lead day 1 to 9, tmax
ECMWF IFS HRESifs_hres 2.85-2.82 · n=24 2.76-2.40 · n=22 3.33-2.71 · n=22 3.24-1.86 · n=21 2.69-0.84 · n=20 3.18-0.68 · n=19 3.66-1.59 · n=17 4.20-1.71 · n=17 4.06-1.51 · n=15 ECMWF IFS HRES MAE by lead day 1 to 9, tmax
Pangu-Weather (GFS init)pangu_gfs 2.71-2.71 · n=2 3.82-3.82 · n=1 Pangu-Weather (GFS init) MAE by lead day 1 to 9, tmax
Pangu-Weather (IFS init)pangu_ifs 2.19-2.19 · n=2 3.45-3.45 · n=1 Pangu-Weather (IFS init) MAE by lead day 1 to 9, tmax
Persistence (baseline)persistence 2.59-0.06 · n=90 3.28-0.21 · n=90 3.52-0.30 · n=90 3.87-0.36 · n=90 4.12-0.39 · n=90 4.16-0.49 · n=90 4.18-0.60 · n=90 4.03-0.74 · n=90 4.26-0.86 · n=90 Persistence (baseline) MAE by lead day 1 to 9, tmax

Up to 1 of the scored days in this table carry a QC flag on the observation (a CLI value that disagreed with the hourly observations, or a fallback source). They are kept in the scores and counted here rather than dropped — see methodology.

+ model too warm · − model too cold · bias interval includes zero · ★ lowest MAE · = not distinguishable from the leader · ▼ significantly worse · n < 30 greyed and unranked · CI is a 95 % moving-block bootstrap interval.

Daily minimum temperature — MAE °F

the last 90 days, 12Z, bilinear. MAE with the bias and n underneath; every cell links to its permanent page. The sparkline shows the same model across lead days 1–9 on a scale shared within this table.
Modeld0d1d2d3d4d5d6d7d8d9lead 1–9
ECMWF AIFS Singleaifs_single 1.46-0.32 · n=29 1.68-0.35 · n=28 1.83-0.36 · n=27 1.90-0.35 · n=26 1.96-0.11 · n=24 1.79-0.15 · n=24 1.84+0.00 · n=23 2.44-0.45 · n=22 2.30-0.13 · n=21 ECMWF AIFS Single MAE by lead day 1 to 9, tmin
Aurora (GFS init)aurora_gfs 0.96-0.96 · n=2 0.51-0.51 · n=1 Aurora (GFS init) MAE by lead day 1 to 9, tmin
Aurora (IFS init)aurora_ifs 1.73-1.73 · n=2 1.20-1.20 · n=1 Aurora (IFS init) MAE by lead day 1 to 9, tmin
FourCastNet v2 (GFS init)fourcastnet_gfs 1.02-1.02 · n=2 0.04+0.04 · n=1 FourCastNet v2 (GFS init) MAE by lead day 1 to 9, tmin
FourCastNet v2 (IFS init)fourcastnet_ifs 0.99-0.87 · n=2 0.44+0.44 · n=1 FourCastNet v2 (IFS init) MAE by lead day 1 to 9, tmin
NCEP GFSgfs 2.17+1.58 · n=29 1.91+0.92 · n=28 2.22+1.39 · n=27 2.66+1.92 · n=26 2.92+1.85 · n=25 2.91+1.63 · n=24 2.33+0.92 · n=23 2.46+1.16 · n=22 2.96+1.60 · n=21 NCEP GFS MAE by lead day 1 to 9, tmin
GraphCast (GFS init)graphcast_gfs 0.97-0.97 · n=2 0.53-0.53 · n=1 GraphCast (GFS init) MAE by lead day 1 to 9, tmin
GraphCast (IFS init)graphcast_ifs 1.69-0.84 · n=27 1.72-0.78 · n=26 1.83-0.68 · n=25 1.85-0.35 · n=25 1.83-0.13 · n=25 1.87-0.04 · n=24 2.49-0.19 · n=23 2.75+0.40 · n=22 2.71+0.78 · n=21 GraphCast (IFS init) MAE by lead day 1 to 9, tmin
ECMWF IFS HRESifs_hres 1.58-0.29 · n=24 1.85-0.09 · n=22 1.95+0.24 · n=22 1.90+0.80 · n=21 2.68+0.97 · n=20 2.29+1.44 · n=19 1.82+1.18 · n=17 2.53+1.71 · n=17 3.48+0.17 · n=15 ECMWF IFS HRES MAE by lead day 1 to 9, tmin
Pangu-Weather (GFS init)pangu_gfs 1.38-1.38 · n=2 0.23-0.23 · n=1 Pangu-Weather (GFS init) MAE by lead day 1 to 9, tmin
Pangu-Weather (IFS init)pangu_ifs 2.94-2.94 · n=2 1.21-1.21 · n=1 Pangu-Weather (IFS init) MAE by lead day 1 to 9, tmin
Persistence (baseline)persistence 1.93-0.09 · n=90 2.43-0.19 · n=90 2.76-0.22 · n=90 3.11-0.22 · n=90 3.17-0.23 · n=90 3.40-0.27 · n=90 3.53-0.29 · n=90 3.46-0.34 · n=90 3.56-0.38 · n=90 Persistence (baseline) MAE by lead day 1 to 9, tmin

Up to 1 of the scored days in this table carry a QC flag on the observation (a CLI value that disagreed with the hourly observations, or a fallback source). They are kept in the scores and counted here rather than dropped — see methodology.

Month-by-month slices, model availability and the observed-truth table live on the canonical page: they do not depend on the window, initialization or interpolation selected here.

All scores as JSON scores_latest.csv daily_errors.csv.gz data & licences