Archived snapshot from the 06Z Fri 9 Oct 2026 model cycle. Forecasts have changed since. See the latest analysis.
NHC model cycle 06Z Fri 9 Oct 2026 · north-central Gulf of Mexico
Physics-based, statistical and AI models are all forecasting the same Gulf hurricane. This page shows where they agree, where they split, and whether the split is unusual. It uses NOAA's public model file and three seasons of verification.
Nearly every model agrees on the shape. Isaias strengthens for about 12 hours, then weakens fast after landfall. NHC's track is offshore at +12 h and inland by +24 h. NHC's discussion gives "a high-end category 2 intensity at landfall." The models disagree on two numbers: how strong the storm is at the coast, and how fast it falls apart over land.
This cycle the early aids and NHC's official forecast all start from 90 kt at 06Z. NHC's peak of 105 kt at +12 h is well above every aid at that time: the highest are HAFS-A and LGEM, at 93 kt. NHC's discussion explains the reasoning: the storm "could continue to intensify for the next 12 hours or so" over very warm water before "a large increase in shear" weakens it, and "it would only take about a 6h delay in the onset of the shear to bring a stronger hurricane than expected onshore."
Isaias sat nearly stationary in the Bay of Campeche for about two days, first as a weak disturbance and then as a depression. It then turned east-northeast and intensified rapidly, from 35 kt at 06Z Wednesday to 70 kt at 06Z Thursday. That is 35 kt in 24 hours. In NHC's operational best track it reached 85 kt at 00Z Friday and 90 kt at 06Z Friday. NHC has revised earlier points since they were first posted (12Z Thursday from 70 to 75 kt, 18Z Thursday from 85 to 80 kt); the scores below use the latest best track. Early on, the models disagreed about both the turn and the intensification. Some moved the system east quickly and strengthened it. Others kept it slow, weak and farther south. Those forecasts can now be checked against NHC's best track.
The 06Z Wednesday runs, made 48 h before Friday's 06Z position, show who handled the turn. DeepMind's raw run, the corrected consensus (HCCA) and NHC's official forecast all placed the center 19 nmi from where it was. Scored against the latest best track, raw DeepMind has been the closest aid with its own track at two days for the three runs before that as well (12Z and 18Z Tuesday, 00Z Wednesday: 6, 13 and 21 nmi). The raw HAFS-A and HAFS-B runs from 06Z Wednesday missed by 152 and 139 nmi. Single runs vary, which is why the averages below matter more. Over every forecast so far that all the aids below share, the interpolated DeepMind aid has the smallest average track error, 16 nmi, and NHC is next at 17 nmi. NHC's 4 AM discussion describes its track forecast as "very near the Google DeepMind guidance, which has been very accurate for Isaias."
The same raw 06Z Wednesday DeepMind run forecast 90 kt for 06Z Friday, matching the best track exactly, as did the corrected consensus. The raw HWRF run was next, at 87 kt. The raw HAFS-A and HAFS-B runs had 78 and 76 kt, and the Navy's COAMPS-TC 96 kt. Across the shared forecasts, the smallest intensity errors belong to the corrected consensus (HCCA), DeepMind, NHC and the neural-network consensus (NNIC), between 3.7 and 5.5 kt.
Nearly every aid was too slow to intensify Isaias. Every aid in the table except COAMPS-TC has a low bias. The global models ran the lowest, as expected at their resolution, and the statistical LGEM was 14 kt low. Two days out, the raw GFS and its ensemble mean put the center 77 and 134 nmi from where it was. COAMPS-TC had the largest track error of the regional models.
| Aid | Type | Intensity error | Intensity bias | Track error |
|---|---|---|---|---|
| HCCA | Consensus | 3.7 kt | −2.7 | 26 nmi |
| GDMI | AI (DeepMind) | 4.8 kt | −3.0 | 16 nmi |
| OFCL | NHC official | 5.0 kt | −2.9 | 17 nmi |
| NNIC | Machine learning | 5.5 kt | −4.5 | — |
| CTCI | Regional physics | 6.7 kt | +5.8 | 91 nmi |
| HWFI | Regional physics | 7.6 kt | −6.9 | 48 nmi |
| HMNI | Regional physics | 7.9 kt | −6.8 | 57 nmi |
| SHIP / DSHP | Statistical | 8.7 kt | −8.4 | — |
| HFAI | Regional physics | 10.0 kt | −9.0 | 56 nmi |
| HFBI | Regional physics | 10.2 kt | −6.9 | 54 nmi |
| AVNI | Global physics | 13.8 kt | −13.8 | 50 nmi |
| LGEM | Statistical | 14.7 kt | −14.2 | — |
| AEMI | Global physics | 23.0 kt | −23.0 | 74 nmi |
Mean absolute errors at +12 to +48 h, over the 24 forecast cycle and lead time pairs that every listed aid shares (a homogeneous sample), scored against NHC's operational best track through 06Z Friday. Corrected 9 Oct: when first published, this table still showed the 00Z cycle's values (20 pairs); the text quoted the 06Z values shown here. Statistical and ML aids borrow another model's track, so their track errors are not listed. This covers two to three days of one storm against a preliminary best track. It shows which models handled this storm's genesis and rapid intensification, not which models are better overall.
At +12 h, near the forecast peak and still over water, the regional physics models span 87–93 kt, with HAFS-A highest. The statistical LGEM is also at 93 kt, and the global physics models are at 90–92 kt. Every aid is between 85 and 93 kt, a narrow range. NHC's 105 kt sits 12 kt above it, for the reasons NHC gives above.
By +24 h, with every aid's center inland, the regional models have dropped to 43–54 kt and the interpolated DeepMind aid to 51 kt. The Canadian ensemble mean (82 kt), the UK global model (71 kt) and the GEFS mean (72 kt) are much higher. NHC is at 65 kt.
At +24 h SHIPS, a statistical model with no land in it by design, holds 72 kt. Its land-aware version, Decay-SHIPS, is at 61 kt. The 11 kt between those two is the land effect as that statistical scheme models it, which makes the pair a useful control. By +36 h the DeepMind aid is at 17 kt, its last forecast point, and the regional models at 18–30 kt. The UK model and Canadian ensemble mean are still at 55 and 66 kt.
This cycle the spread is not mainly track spread. At +24 h every plotted aid has the center within 30.6–31.4°N and 87.1–88.2°W, yet their intensities range from 43 to 82 kt. Much of the disagreement is in how each model weakens a hurricane over land.
The DeepMind line on most spaghetti plots is GDMI, an interpolated aid. The raw model output (GDMN) arrives too late for the forecast cycle it would serve. NHC therefore takes the previous run and adjusts it to the storm's intensity at the start time.
For Isaias, the raw 00Z DeepMind run forecast 87 kt for 06Z, against 90 kt in the best track, a close start. The aids were adjusted to 90 kt. After the adjustment, the 06Z aid runs 3 kt above the raw run at 0 h, 9 kt below it at +12 h and 5 kt above it at +24 h, small differences in both directions. The raw run peaks at 94 kt at +12 h, below NHC's 105 kt. In the three cycles before this one the raw run started well above the observed intensity, by 7 to 26 kt against the latest best track, and the adjustment pulled the interpolated aid down. This cycle it barely changed it.
"AI is less aggressive" can be partly an artifact of this adjustment, as it was in earlier cycles. Compare raw and interpolated output before drawing conclusions about any model class.
Intensity forecasts over land depend on when and where the center crosses the coast. The tracks below cluster tightly toward the Alabama and western Florida Panhandle coast. At +24 h every plotted aid has the center inland, between 30.6°N and 31.4°N and between 87.1°W and 88.2°W. NHC's discussion says its forecast track is "very near the Google DeepMind guidance." Small east–west and timing differences remain, but this cycle they explain less of the intensity spread than how each model handles land.
Not among the core models. The split comes from the models outside that core.
We took eight aids that ran in every season from 2023 to 2025: HAFS-A, HAFS-B, HWRF, HMON, COAMPS-TC, GFS, Decay-SHIPS and LGEM. For every Atlantic forecast cycle that started at hurricane strength, we computed their spread, the standard deviation across models. By that measure, Isaias's disagreement is ordinary at +24 and +48 h (57th and 38th percentiles). At +72 h it is low, in the 2nd percentile, but only six of the eight core models reach +72 h, and all have the storm far inland and weak. Spread in knots shrinks over land (see the Fig. 5 caption).
Isaias stands out once the other models come back in. Adding the UK global model, the Canadian ensemble mean and SHIPS raises the +48 h spread from 7.5 kt to 13.9 kt. (The DeepMind aid is normally in this comparison too, but it has no +48 h value this cycle.) That figure uses a different set of models, so it can't be compared to the histograms. It does show where the disagreement comes from: models that treat land very differently.
NOAA's global ensemble (GEFS) runs 31 versions of the GFS, each with slightly perturbed starting conditions and model physics. At this resolution its intensities are much too weak. At 00Z the members started between 47 and 65 kt against an observed 85 kt, so read the plume for timing, not strength. On the 06Z clock, 18 of the 31 members weaken fastest between +12 and +24 h, and 30 of the 31 lose at least 15 kt in that window. It spans 18Z Friday to 06Z Saturday (1 PM Friday to 1 AM Saturday CDT), consistent with NHC's forecast landfall tonight or early Saturday. That is independent evidence that the timing of the move inland is fairly well constrained.
The European (ECMWF) model and ensemble are not in NOAA's public model file, so they are not shown. Google's model is itself an ensemble, but the public file carries one forecast per cycle.
Each version of this page, one per NHC model cycle, is committed to a public repository before landfall, and the commit history timestamps it. After the storm, each aid will be scored against NHC's best track from +12 to +72 h, on the questions this storm raises:
One storm can't settle which model is better. NHC's annual verification reports are where models get judged. This page is a worked example of reading the guidance.
This page is updated as new NHC model cycles arrive. Each earlier version is preserved unchanged.