Model Ensembles

An ensemble of model forecasts is: more than one model forecast made for the same forecast time and location.


Time-Lagged Ensembles: Model runs begun at different initial times.

Example: GFS forecasts initialized every six hours. When a big storm is brewing, we look at each new model run carefully to see how the model forecast for the storm event has changed.

Example: Forecasts get worse as lead time increases.


Perturbation Ensembles: Model runs begin at the same initial time, but with slightly different initial conditions (perturbations). The basic idea is, there is error in the initial condition. Making multiple runs with slightly different initial conditions provides a sense of how the forecast might be different should the initial condition be in error.

Examples:

A set of perturbation forecasts and initial conditions.

A common way to view the ensemble forecasts: Spaghetti Diagrams

A way to look at the ensemble forecasts of precip for DC

Current GEFS Ensembles (for a selected interesting time)

NCEP NCO Model Page

NCEP EMC GEFS Page


Multi-Model Ensembles: Model forecasts for an event made by different models. Each model has characteristic, and different, biases.

Examples:

Different model forecasts of a hurricane track

Hurricane track forecast from the perturbation ensembles from three different models.

For a forecast of a single event (eg, Hurricane Sandy), there are many models that are run around the world, usually initialized every six hours, and for many models there are perturbation model runs made. So there are hunreds or even thousands of model runs made for a specific forecast time!