Titan U glossary

Ensemble Forecast

An ensemble forecast is a set of model forecasts valid for the same time, each started from slightly different initial conditions or run with different versions of a model. Instead of one answer, it shows a range of plausible outcomes and how closely they agree.

Storm Forecasting Models

Why one run is not enough

The atmosphere is never measured perfectly, and small errors in a model's starting data grow over time. Models also cannot represent processes smaller than their grid or capture every physical process.

An ensemble samples those uncertainties by nudging, or perturbing, the starting data and sometimes by swapping in different physics packages.

Spread, mean and probability

Spread is how far the members fan out from each other. Wide spread points to lower predictability, while tight clustering suggests more confidence in that part of the forecast.

The ensemble mean averages the members, which filters out small, uncertain features and gives a smoother picture. A probability counts how many members meet a threshold, such as an inch of rain, and divides by the total number of members.

Examples as of October 2026

NOAA's Global Ensemble Forecast System (GEFS) has 31 members, one unperturbed control and 30 perturbed runs, at about 25 km. It runs four times a day out to 16 days, with the 00 UTC run extended to 35 days, and a June 2026 bug-fix upgrade brought it to version 12.3.20.

For storm-scale detail, the High Resolution Ensemble Forecast (HREF) combines several models at about 3 km. HREF and the Short-Range Ensemble Forecast (SREF) are scheduled to retire on 10/14/26, replaced by the Rapid Refresh Forecast System and its ensemble.

What an ensemble cannot do

Every member inherits its model's grid, so a 25-km ensemble cannot show individual thunderstorms no matter how many members it has. An 80 percent ensemble probability means 80 percent of members agree, and it needs a forecaster's judgment before it becomes a real-world chance.

Keep going

How Ensemble Forecast connects

Builds on 1

Forecast Uncertainty
Ensembles exist to sample forecast uncertainty from imperfect starting data and models, so that idea comes first.

Leads to 4

Probabilistic Winter Precipitation Forecast
The PWPF starts from a WPC forecast and uses a multi-model ensemble to supply the spread of possible snow and ice amounts around it.
ECMWF
When forecasters cite ECMWF they usually mean its main model and its 51-member ensemble, so understanding ensemble forecasts comes first.
Quantitative Precipitation Forecast
Probabilistic QPF adds uncertainty to a single amount by using the spread of an ensemble of model forecasts.
Winter Storm Severity Index
The probabilistic WSSI runs the same impact algorithms across an ensemble of forecasts to show the chance of reaching each impact level.

Includes 2

Ensemble Spread
Spread is one of the core outputs of an ensemble forecast, alongside the mean and probabilities.
High-Resolution Ensemble Forecast
The HREF is an ensemble forecast built from existing convection-allowing runs rather than one model nudged many times.

Contrasts with 3

Convection-Allowing Model
More ensemble members do not add storm detail, since each member keeps its model's grid, while a convection-allowing model resolves storms directly.
Probabilistic Outlook
An ensemble probability counts agreeing members, while an outlook probability is a forecaster's chance of severe weather near a point.
National Blend of Models
The National Blend is a calibrated blend of many models rather than a single ensemble, though like an ensemble mean it smooths small, uncertain features.