Improving Ensemble Forecasting in the Global Forecast System by Better Accounting for Additional Sources of Uncertainty

In order for an ensemble forecast system to be effective as a forecasting tool, all sources of forecast uncertainty need to be accounted for. Historically, ensembles were generated by perturbations to the initial conditions, and the chaotic nature of the atmospheric models caused those small initial perturbations to grow over time. Although, this does produce an ensemble of forecasts, the results are typically under-dispersive, and result in a model forecast that is over confident.

The current operational Global Forecast System (GFS) ensemble run by the National Weather Service has a technique to address model uncertainty, but it is only effective in the winter-hemisphere's storm track. New methods of addressing uncertainty have been implemented into the GFS by NOAA/ESRL, and have been running in operations since January 2015 as part of the data assimilation cycle. More recently, a new method for addressing uncertainty in the lower boundary conditions (SST and land surface) has been added, and these additions being considered in a future upgrade to the GFS ensemble. The benefits of addressing model uncertainty during the forecasts will be presented.