Model Evaluation Group (MEG)
About the Model Evaluation Group
The Environmental Modeling Center's Model Evaluation Group (MEG) formed in 2012 with the goal of bolstering statistical evaluation of NCEP model forecasts with evaluation from a synoptic, sensible-weather perspective. Presentations are given weekly, covering topics that include case studies of high-impact weather events, the performance of parallel models, model biases, and other model behaviors requiring documentation or fixes. Presentations are also occasionally given by interested members of the NWS and academic communities.
MEG meetings are held most Thursdays at 11:30 a.m. eastern time. Contact a group member to be added to a list for email announcements and dial-in information.

Model feedback is always appreciated and can be directed to any group member.

The MEG is:

Geoff Manikin (
Logan Dawson (
Alicia Bentley (

Presentation Examples
Past MEG presentations are available to those with a email address.
A few MEG presentations have been organized into recaps to give a more concise overview of model performance:

Download the MEG Hurricane Harvey recap
Download the MEG Hurricane Irma recap
Download the MEG Hurricane Maria recap

Forecasting the onset and dissipation of marine stratus is a major challenge along the California coast. Above is an example of how two high-resolution NCEP model total cloud fields (percent) from the NAM nest (middle) and HRRR (right) depicted early-morning stratus observed in the Salinas Valley (left). In this case, the HRRR failed to extend clouds into the valley while the NAM nest successfully produced valley clouds. Low-level stratus is very sensitive to boundary layer processes; small differences in the HRRR's boundary layer evolution likely led to its restricted total cloud field.

A multi-day severe outbreak in January 2017 concluded with a large high risk issued by the Storm Prediction Center over southern Georgia and northern Florida on 22 January (top right). Despite large probabilities for siginificant tornado formation, the day was less active than 21 January (convective outlook and reports top left). While severe parameters human forecasters rely on reached extreme values over Florida and supported the issuance of a high risk, HRRR forecasts were less indicative of a 22 January Florida tornado outbreak. In both the 04Z 22 initialization (bottom left) and 17Z 22 initialization (bottom right), the 12-h maximum updraft helicity fields, a field commonly used to assess severe threat in a model, ending 16Z 22 and 05Z 23, respectively, restricted the largest values to areas in Georgia where tornadoes were observed (red markers). Over Florida, values were more modest. These HRRR forecasts are an example of a model success in which the model may have identified factors limiting severe potential that the human forecasters could not detect.

A snowstorm on 14 March 2017 produced feet of snow across New England as it tracked across the eastern tip of Long Island and eastern Massachusetts (stage IV liquid equivalent in fill with the low track overlaid, right). The Global Ensemble Forecast System (GEFS) is frequently used by forecasters to assess the range of storm possibilities. The two left images show probability of GEFS 48-h accumulated precipitation exceeding one inch in grayscale with GEFS member low tracks overlaid. In this case, the 12Z 11 March cycle of the GEFS contained the verified track within its range of solutions, but the 00Z 12 March cycle did not. An observed event lying outside an ensemble's range of solutions indicates that the ensemble is underdispersive. Underdispersiveness is a frequent issue in the GEFS, especially in short-range forecasts.

In the medium range, when Hurricane Matthew was still south of Cuba, its track was uncertain. Specifically, it was unclear how close Matthew would come to Florida. Four global ensemble systems are shown here: the GEFS (top left), European (top right), Canadian (bottom left), and UK (bottom right). In this 00Z 03 October 2016 cycle, the observed track of Matthew was outside the range of GEFS solutions. The observed track was also nearly outside the range of Canadian solutions. The European ensemble better captured the possibility of a Matthew track near the coast of Florida, and the UK ensemble was most accurate with its ensemble mean slightly east of the observed track. Ultimately, Matthew tracked far enough off the Florida coast for Florida to avoid widespread significant damage.