A look at deterministic model behavior for the 2015 month of BIG snows in the Northeast

A tranquil winter in the Northeast made an abrupt turn with an incredible month long period of winter storminess beginning in late January 2015. Snowfall from a series of storms, some in themselves record breaking, crushed records for multiday periods and for the month of February as a whole for a number of locations in New England. The onslaught began calmly enough with a modest Nor'Easter on 23-24 January that was quickly followed by a raging and record-setting blizzard on 26-28 January. At least three other significant storms followed through the middle of February, creating an impressive snowpack across much of New England.

At NOAA/ESRL/GSD we have been running a new global model for several years, called the FIM, for Flow-following finite-volume Icosahedral Model. There are several unique aspects of the FIM, including the icosahedral horizontal grid (mostly uniform across the globe), an isentropic-sigma hybrid vertical coordinate, and finite-volume horizontal transport. Current and potential future roles of the FIM include its use as a climate model, forecasting for hurricanes, and perhaps becoming a part of the North American Ensemble Forecast System (NAEFS). Several versions of the FIM are run at GSD, with varying physics and at different horizontal grid resolutions (~15 km, 30 km, and 60 km). In addition to quantitative verification of the FIMs and comparisons in this regard with other global models, an ongoing effort involves a more qualitative look at how the FIMs perform for various significant weather events, and how the FIM forecasts compare to other operational global models.

For the period of massive snows in New England five separate storms were archived, with comparative forecasts from the deterministic global models including the ECMWF, UKMO, NAVGEM and operational GFS. A number of interesting aspects of model behavior appeared to occur with consistency for the various storms, and these will be discussed in this paper. In particular, the longer range (days 8-10 in advance) forecasts often gave a good indication of an upcoming event. Into the middle and even at times somewhat shorter (about 3 days out) range the forecasts often drifted to much less of a threat, or no threat at all (sometimes together amongst the models), typically sending the threatening storm out to sea well south of New England. The model forecasts then improved (although there were often some smaller-scale and important forecast issues; NYC snowfall for the second storm) within about 3 days of the storm. This work will compare the various model forecasts for these events. We will examine the initial conditions for the varying forecast times to determine if some pattern was present for successful versus unsuccessful forecasts. This includes tracing back the upper level feature that eventually produces the storm, so that comparisons can be made in how the features evolved for each event and whether the track of the important feature across the globe was important to the forecast quality.