Forecast Models
Ensemble Forecasting
Isolines of convective available potential energy (CAPE) of 1000 J/kg valid at 0000 UTC 6 April 2006. Each line is from a different member of a model ensemble. (more about CAPE)
An ensemble is simply a group of model forecasts that are valid over the identical time period. These forecasts provide information on the different ways in which the atmosphere may evolve over the next few hours or the next few days and even out to the next few weeks. Ensembles are needed because we do not have enough information to accurately depict the present state of the atmosphere. Even with all the information we obtain from satellites, radars, weather balloons, surface instruments, and other data sources we are unable to provide a perfect three-dimensional picture of the atmosphere at any given time. This means that the information we use to start a numerical weather forecast model, called an initial condition, is imperfect. These imperfections grow once the model starts and lead to forecasts that do not agree with the actual evolution of the atmosphere.
An ensemble begins by taking our best initial picture of the atmosphere and perturbing it in ways that agree with what we know about the errors of our instruments and our ability to sample the atmosphere (e.g., there are fewer observations over the oceans than over land). Once we have a set of these initial condition perturbations, we start a separate model forecast from each initial condition. These different forecasts then represent different ways in which the atmosphere can evolve over the coming hours or days. By analyzing these different scenarios, we can determine the most likely evolution of the atmosphere and determine the odds that certain weather events will occur. Numerous studies have shown that ensembles are more accurate than providing a single forecast from the best initial condition, and we also know that ensembles provide more useful information to decision makers.
Ensembles presently are run operationally at the National Centers for Environmental Prediction for both short-range (0 to 3 days) and medium-range (0 to 14 days) forecasts. The short-range ensemble forecasts are used to help determine areas where severe weather is likely to occur. On the very short time scales, NSSL is beginning to study ensembles for very short-range (0 to 1 h) forecasts of severe weather events. These ensembles assimilate Doppler radar data into cloud-scale numerical models to provide improved predictions of thunderstorms and their associated severe weather. While still in a research mode, initial results suggest that it may be possible to use these forecasts in warning operations, leading to a shift from the present "warn on detection" strategy to a "warn on forecast" strategy that would provide longer lead times for severe weather events.
Model errors also contribute to our inability to make perfect forecasts from numerical models. NSSL scientists have been at the forefront of exploring the use of different models and different model parameterizations within short-range ensembles and data assimilation methods. Results indicate that ensemble forecasts are more skillful when model error is included explicitly as part of the ensemble, either through the use of different models in the ensemble or through the use of different parameterization schemes within the ensemble.
More information on ensembles can be found at http://meted.ucar.edu/nwp/pcu1/ensemble/frameset.htm which offers an online study resource. This course was designed by COMET.
