Ensembles and spaghetti plots, explained

What an ensemble is

Weather forecasting starts with an estimate of the current atmosphere, and that estimate is never perfect. ECMWF explains that errors in the starting state and approximations in the model both grow with time. Instead of pretending the starting point is exact, forecast centers run the model many times with small, deliberate changes to the start. Each run is called a member.

NOAA’s Weather Prediction Center puts it simply: an ensemble is two or more forecasts verifying at the same time. If the members stay close together, the atmosphere is in a setup where small errors do not matter much. If they split apart, small differences are growing fast, and the forecast is less certain. ECMWF describes ensemble spread as a measure of predictability and forecast confidence.

There are also multi-model ensembles, which combine members from different models. WPC notes that a multi-model ensemble may verify better than an ensemble from one model, and that the GFS ensemble mean began to beat the single operational GFS run for 500 mb heights at about day 3.5.

Four ways to look at an ensemble

NOAA’s Climate Prediction Center publishes spaghetti plots of ensemble members for days 3 to 7, with the ensemble mean drawn as a dashed line, and mean maps for later periods. The mean is the middle of the fan. The spaghetti shows the fan.

How to use spread to judge confidence

1. Look for clusters. If most members bunch together, that is the main scenario.
2. Look for two clusters. A split into two groups often means two different outcomes, such as a storm that either turns up the coast or goes out to sea. Plan for both until it resolves.
3. Find the outliers. A single member far from the rest is possible, but it is a minority, not a forecast.
4. Watch how spread changes between runs. Tightening spread over successive runs is a better sign than a single tight run.
5. Remember that spread is not truth. A tight cluster can still be wrong if every member shares the same flaw.

This is the grown-up version of asking how many ensemble members support the wild run. If the answer is one out of fifty, the wild run is entertainment. If the answer is twenty out of fifty, it is a scenario.

Tropical storms

For hurricanes, the spaghetti plot is a plot of tracks, one line per member or per model. Two cautions help. First, it shows where the center of the storm may go, not where the damage will be. The National Hurricane Center says its cone shows the probable track of the center only and that tropical cyclone effects extend well beyond it. Second, on average the NHC official forecast has smaller errors than any individual model, because forecasters weigh all of the guidance. Use the spaghetti to see the range of possibilities, and the official forecast for decisions.

Ensemble tracks that bunch up near the coast with the same turn give more confidence about where the storm will be. Tracks that fan out, with some recurving out to sea and some crossing land, mean the forecast is waiting on the steering pattern, which you can check on a 500 mb map.

Winter storms

For a nor’easter, a few miles of track can move the rain-snow line across a city. Spread in the track therefore translates into spread in snowfall. NWS State College notes that the heaviest snow in a nor’easter falls north and west of the low’s track and that timing precipitation changes is very difficult when the track is uncertain. A plume at your location helps because it translates the track spread into the totals that matter: some members will show a foot, some a dusting, and the count tells you which is more common.

Pair a spaghetti plot of storm tracks with a plume at your location. The first shows where the storm may go. The second shows what that does to your snow or rain. Our 40/70 benchmark guide shows how forecasters use a fixed reference point for the track.

Ensembles in 4070

• Plumes: ensemble members and recent-run sets at a U.S. point for temperature, precipitation and snow, with the models grouped so you can look at one at a time.
• Storm Tracks: developing lows through the GEFS and ECMWF ensembles and the four most recent runs of several models, with agreement, spread and central pressure. The paths show model support, not a probability or impacts.
• Ensemble storm tracks for tropical systems: GEFS, ECMWF ENS and AIFS ENS member tracks and an experimental consensus. These are model runs, and the National Hurricane Center remains the official source.
• Model Explorer includes GEFS spaghetti contours, GEFS mean and spread, and ECMWF EPS mean and spread maps.

All of these are part of 4070 Pro, with a seven-day free trial. See the model pages for GEFS, ECMWF EPS and the most accurate model guide.

Ensemble spread shows model uncertainty. It is not a probability of rain, snow or damage at your address. Official National Weather Service forecasts, watches and warnings remain the authority.