Spaghetti models, explained
A spaghetti model plot overlays the projected tracks of many computer forecast runs on one map. Each line is one model run's or one ensemble member's estimate of where a storm's center will go. Tight clustering suggests agreement; wide spread signals uncertainty. The National Hurricane Center's official forecast is separate, and is the authority.
What the lines actually are
A spaghetti plot is a map with many storm tracks drawn on it at once. Each line is one run of one forecast model, usually a physics simulation started from the best available snapshot of the atmosphere. The lines are not a forecast; they are what the models produced, side by side.
They come from two kinds of source, and the distinction matters. Some are deterministic runs: one high-resolution run per cycle, one track. Others are ensemble members — the same model run many times at coarser resolution, from slightly perturbed initial conditions and physics, to sample what the analysis cannot pin down. The ECMWF ensemble carries 51 members, a control plus 50 perturbed. The GEFS carries 31.
That asymmetry is why counting lines fails. Thirty clustered ensemble members look like overwhelming agreement when they are really one model expressing itself thirty times. The Model Lab keeps deterministic tracks and ensemble spread visually separate.
Clustering means confidence. Spread means uncertainty.
When the lines bunch together, the models are resolving the same steering pattern — usually a well-defined subtropical ridge or a clear trough. Small differences in the starting analysis do not change the outcome. That agreement is real information.
When the lines fan out, the steering is ambiguous. The usual culprit is a weakness in the ridge the storm may or may not turn into, or a trough running half a day late. Shift it six hours and the storm recurves out to sea instead of continuing west. That is chaos, not model failure.
Spread is a measurement, not a defect. A wide fan at day five is the honest answer. On the environment view you can often see why the models disagree: a shear axis or the edge of a ridge sitting where the tracks split.
The official forecast is not the average of the lines
The official National Hurricane Center track is written by human forecasters. They see the same guidance you do and more — aircraft reconnaissance, satellite fixes, scatterometer passes — plus their own read on which models have handled this storm well. The result is not the mean of the lines.
Consensus aids — averages of selected models — are strong guidance, often better than any single member of the average. But NHC also discounts a run that initialized the center wrongly, weights models by recent performance, and keeps continuity so the forecast does not lurch on one bad cycle.
The cone is not model spread either. Its circles are sized from NHC's official track errors over the past five years — two-thirds of those errors fall inside — and the center stays within the whole cone roughly 60 to 70 percent of the time. It says nothing about where the hazards are: wind, surge, and rain routinely reach well beyond it. The live map mirrors the official cone, track, and watches and warnings. Act on those. NHC and your local emergency management are the authority.
Which model is which
Global models simulate the entire atmosphere and are the reference for steering. Among the physics-based globals, ECMWF's has the strongest long-run record on tropical cyclone tracks. The GFS is NOAA's global model, run four times a day out to sixteen days. The UKMET and Canadian CMC are the other globals forecasters watch; NAVGEM, the Navy's model, carries less weight.
The ensembles come from those same globals: GEFS is the GFS run many times over, the ECMWF ensemble the European equivalent. No single member deserves attention. The shape of the cloud does.
Newer lines come from AI models, trained on decades of past weather instead of solving the physics each cycle. Their track performance has been strong enough that NHC now weighs them alongside the physics models; intensity is less settled. One line is still one run.
Hurricane models are a different class. They run at much finer resolution on a moving nest and couple to the ocean, so the storm can cool the water beneath itself. HAFS replaced NOAA's long-serving HWRF and HMON and runs two configurations on active storms. Read them for intensity and structure more than for a five-day track. The historic archive shows how past storms played out.
The classic misreads
The most common mistake is treating a lone outlier as a scenario worth planning around. Sometimes it is. More often it is a run that initialized badly — the model put the center in the wrong place, or never built a coherent vortex, and everything after inherits the error. Over open ocean, with few observations, analyses can be poor.
Some plots also carry lines that are not true forecasts. CLIPER is a climatology-and-persistence baseline that real guidance has to beat to show any skill; the BAM family are simple trajectory schemes that push a storm along with the surrounding flow. Read the labels before reacting to a line.
The last misread is intensity. Track skill has improved enormously over three decades; intensity skill barely moved until about 2010 and has gained ground since, though rapid intensification remains among the hardest problems in the field. A spaghetti plot shows center positions and nothing else — not peak winds, not the size of the wind field, not how far surge reaches. Saved places gives you a threat card for one address, not a line on a map.
When the runs update
Models run on fixed cycles tied to Coordinated Universal Time: 00Z, 06Z, 12Z, 18Z. The GFS, UKMET, and NAVGEM run four times a day; ECMWF and CMC issue their full-length runs at 00Z and 12Z. Output arrives a few hours after the cycle time.
One cycle is a data point. Two or three consecutive cycles agreeing is a trend. For Atlantic storms, NHC issues full advisories at 5 AM, 11 AM, 5 PM, and 11 PM Eastern, with intermediate advisories every three hours in between once watches or warnings are in effect.
The Model Lab draws these tracks side by side, ensemble spread alongside the individual runs. Use it to see the shape of the uncertainty. Then take your decisions from NHC and your local emergency managers — official links are on the resources page.
Eyewall presents live public NOAA / National Hurricane Center data. It is not a warning authority — always follow the NHC and your local emergency management.
Open the live hurricane map