Author's Note: The following report has not been subjected to the scientific peer review process.

1.  Introduction

While half the warned counties in the National Weather Service 
(NWS) Weather Forecast Office (WFO) Greenville-Spartanburg (GSP) 
county warning and forecast area (CWFA) verified in this event, 
a sizable part of the warned area did not receive the amount 
of winter precipitation, or even the precipitation type, that 
was expected.  This included much of the South Carolina 
Mountains and the southern and central North Carolina Mountains.  
It was also a missed forecast as significant icing did not 
develop in the Piedmont and parts of the Foothills of North 
Carolina.  Areas which verified were the higher elevations of 
the North Carolina Mountains, including the Tennessee border 
counties.  These counties primarily verified with a northwest 
flow snow event, as was expected.  The northern North Carolina 
Foothills verified, mainly with sleet, and parts of the Georgia 
Mountains verified, receiving significant snow early in the 
event, before warm advection changed the precipitation over 
to rain (Fig. 1).  The Upstate of South Carolina and the 
southern North Carolina Piedmont were not included in watches 
or warnings for this event.

Figure 1.  Map of snow, sleet and ice accumulations from the 
6-7 January 2002 winter storm.  

2.  Synoptic Overview

This was not an easy forecast by any stretch, with offices all 
along the East Coast verifying poorly on their watches and 
warnings.  Principally this was caused by a significant 
Quantitative Precipitation Forecast (QPF) error by the Eta 
model for points farther north.  The track of the heaviest 
QPF was expected much farther east than what verified, 
illustrative of the more westward track of the storm and hence 
warmer verification.  Some offices farther north saw snowfall 
totals of up to one foot where only light snow had been 
forecast just 12 to 24 hours before the event

The path of the low verified a little farther inland than the 
models had projected, though the errors were fairly small over 
the southern states.  Precipitation rapidly translated across 
the region, with moderate to heavy precipitation only lasting 
3 to 4 hours at most locations.  The 12 km Eta model exhibited 
a cold bias on select runs and exhibited a considerable amount 
of run-to-run variability.  The AVN model was more consistent 
than the Eta and if anything was a little warm.  The Canadian 
GEM model (both the global and regional versions), while perhaps 
the most consistent model in forecasting the general evolution 
of the event (it nailed the pattern days in advance), was also 
the model with the most significant cold bias. 

In the end, however, the primary contribution to the inaccurate 
forecast may have been forecaster misjudgment.  It was believed 
that the Eta soundings were a little too warm as dynamic cooling 
would subtract a degree or two Celsius.  In reality, what 
forecasters often call "dynamic cooling" is handled well by the 
Eta and should not be compensated for without a good reason.  
Also, the brief period of precipitation should have been 
considered, though this may not have been as big an issue as 
precipitation amounts were generally equal to or greater than 
what was projected.

The following sections will discuss the various aspects of this 
event touched on in the paragraphs above.  Some will be little 
more than a statement of purpose or intent at this time, while 
others will have a more significant amount of data. 

Before proceeding any further, it will be beneficial to first 
view this reflectivity loop from the event.

a. Run to Run Variations in the Eta Model and Erroneous Forecaster 
   Presumptions.

Across most of the central and southern North Carolina 
Mountains, where 4 to 8 inches of snow was forecast, a light 
glaze of freezing rain fell and temperatures warmed into the 
upper 30s.  Higher elevations had up to 2 inches of wet snow 
before the changeover.  The period of heaviest precipitation 
verified around 1300 UTC on 6 January in the mountains.  Bufkit 
soundings for the Asheville, North Carolina, airport (KAVL)
from the Eta model valid at 1300 UTC 6 January for model cycles
starting at 0000 UTC 5 January and ending with the 1200 UTC 
run on 6 January are shown in Figure 2.  There were timing 
differences in each run of the Eta.  The slightly different
timing of the warm nose at KAVL in each model run was captured 
by the Bufkit soundings at 1500 UTC (Fig. 3) and at 1700 UTC 
(Fig. 4).  The 0000 UTC run on 5 January was coldest Eta run 
of the group.  It was this run that the mid shift on 5 January 
used to issue the first Winter Storm Watch for the event.

Figure 2.  Eta model Bufkit soundings for KAVL, valid at 
1300 UTC 6 January 2002.  The soundings are taken from the 
0000 UTC 5 January model cycle (upper left), 1200 UTC 5 January 
cycle (upper right), 0000 UTC 6 January cycle (lower left), and 
the 1200 UTC 6 January cycle (lower right).  Click on each 
image to enlarge.

Figure 3.  As in Figure 2, except for 1500 UTC 6 January.  
Click on each image to enlarge.

Figure 4.  As in Figure 2, except for 1700 UTC 6 January.  
Click on each image to enlarge.

The Air Resources Lab (ARL) has an online archive of both the 
EDAS and GDAS data.  We are still looking into whether or not 
we can compare the ARL archive of EDAS and GDAS data since the 
GDAS analysis sounding from Asheville at 1200 UTC was much 
warmer than the EDAS analysis sounding from 1200 UTC.  At 
1200 UTC 6 January the Asheville airport was reporting freezing 
rain and a temperature of 32 degrees F.  A location a few miles 
northeast of the airport, at an elevation of 4,320 feet 
reported a temperature of 36 deg F at 1330 UTC.  If we can 
rely on the GDAS data, this would be a very interesting case 
of the Eta model verifying too cold for the onset of the event.  
However, until we learn how these soundings are mapped, it is 
best to rely on the Bufkit soundings for point specific model 
sounding data.

The Bufkit precipitation algorithm appears to run too warm, as
suggested by the Partial Thickness Universal Nomogram plots at
1700 UTC 6 January (Fig. 5).  The larger red dots indicate the 
most recent hourly thickness plot, thus it should be fairly 
easy to extrapolate back through the four or five hours of 
significant precipitation using the thickness images.  Even 
though the 0000 UTC run on 6 January was cooler than the 
1200 UTC run on 5 January, the thicknesses were still leaning 
a little toward a mix rather than all snow.  Based on the cooler 
GEM model, and a belief that "dynamic cooling" should help hold 
the warm nose at bay during the critical 1200-1500 UTC time 
frame, the cool side of the Eta solution was favored when the 
mid shift made the decision to warn on 6 January.  In fact, 
the low tracked a little farther west, bringing the warm nose 
across all of the mountains by midmorning, and verifying a 
warmer solution during the most intense precipitation than 
suggested by the 0000 UTC Eta run on 6 January.

Figure 5.  Partial Thickness Universal Nomogram plots from Eta 
model Bufkit soundings at KAVL, valid at 1700 UTC 6 January 2002.  
The nomograms are taken from the 0000 UTC 5 January model cycle 
(upper left), 1200 UTC 5 January cycle (upper right), 0000 UTC 
6 January cycle (lower left), and the 1200 UTC 6 January cycle 
(lower right).  Click on each image to enlarge.

The 850 mb height and temperature fields from the last four
runs of the Eta and AVN are quite interesting (Figs. 6 and 7).  
Between the 1200 UTC run on 5 January and the 0000 UTC run on 
6 January, 850 mb temperatures cooled nearly 4 degrees C over 
the North Carolina Mountains, and there were significant 
temperature fluctuations in all four runs shown.

Figure 6.  Eta model 850 mb geopotential height and temperature 
(deg C) valid 1200 UTC 6 January 2002.  Model run time is below 
each image.

However, the temperature fluctuations in the AVN solution were 
considerably smaller.  Notice that the 0 deg C line stays nearly 
over the Tennessee - North Carolina border for each of the last 
three runs herein (those being 0000 UTC 5 January, 1200 UTC 
5 January, and 0000 UTC 6 January, respectively).  While the 
last three runs, valid at 1200 UTC on 6 January, were very 
consistent, they also appeared to verify reasonably well in the 
mountains, though perhaps a little too warm.  Unfortunately, 
a more detailed analysis of the AVN is not possible at this 
time due to lack of data.  Still, these gross fields are still 
very interesting.  Is it possible that the 12 km Eta is subject 
to more run-to-run variability than pervious versions of the 
model?  An event which happened less than a week before also 
exhibited run-to-run differences, though those differences 
were not as important since that event was all snow.

Figure 7.  As in Figure 6, except for the AVN model.

In retrospect, the situation was handled the best by the 
1200 UTC Eta model run from 5 January.  The previous runs of 
the Eta had featured a stronger, slower, and more cutoff 500 mb 
short wave than the AVN or GEM models.  This resulted in a 
low farther to the south and a colder airmass.  The 1200 UTC 
run on 5 January came more into line with the other two models 
in showing a more rapidly translating wave.  Compare the Eta 
model's surface low position at 1200 UTC 6 January to the 
analyzed position to see that the Eta was pretty much right 
on the money with this run.  However, the day shift on 
5 January strongly considered the colder Canadian model and 
the previous, colder run of the Eta, when the decision was made 
to upgrade the warnings in the mountains.

As we were unsure of how the Eta model handled cooling due 
to strong upward vertical velocities, there was a tendency 
throughout the forecast to assess a small amount of cooling 
which it was assumed the model would not generate.  The model 
should take into account cooling by expansion as a stable 
layer is lifted (Dr. G. Lackmann, NC State University, 
personal communication).  The only process which could result 
in unmodeled cooling would be creation of a stable layer by 
melting of snow.  This layer would be more subject to cooling 
than the Eta would project due to shallower lapse rates. 
Considering that strong low level warm advection should 
completely overwhelm cooling from melting, it is best not to 
assess any kind of "dynamic cooling" fudging to the Eta data 
unless it could be proven that vertical velocities were in 
fact stronger than predicted.

Additionally, it would have been helpful if a rudimentary set 
of data from other events which affected the CWFA in the past, 
exhibiting similar characteristics to this one, was available. 
In particular, a look at the "surprise" snow event in 
Asheville in January 1998 would have been of benefit.  It 
appears that the Eta actually did a good job with this event, 
though we have only anecdotal evidence of this.

b. Freezing rain verification in the Piedmont and Foothills

There has been talk about the Eta 2-meter temperatures being 
too cold, and warming of the boundary layer by "warm" clouds. 
Both of these effects were slightly in evidence that morning. 
However, the differences were not great as can be seen in 
Table 1.

A more basic mistake was assuming that a weak in-situ wedge 
would provide enough support to overcome latent heat released 
by freezing rain.  It was not.  In fact, later in-situ wedge 
events which occurred on 19 January 2002 and 6 February 2002, 
which had higher surface pressures, also failed to result in 
damaging ice accumulations.  While ice did form on trees at 
least as far south as Greer, South Carolina, there was only one 
pocket of ice accumulation that even might have been close to 
winter storm criteria, in Catawba County, North Carolina.  It 
is simply very difficult for in-situ events to generate the 
necessary cold advection to overcome the latent heat released 
by freezing rain.  Coupled with the Eta's tendency to run 1 to 
1.5 degrees too cool with surface temperatures this year, this 
is something that forecasters need to take into account when 
forecasting damaging ice accumulations.

c. Other models - the Canadian GEM and SEF

It is with some trepidation that I supply these maps from the 
0000 UTC run on 6 January from the GEM (Fig. 8).  Since we do 
not have any archived data from the 0000 UTC or 1200 UTC runs 
on 5 January we can only speculate based upon what we remember 
the GEM doing for those runs.  However, while we have a great 
deal of faith in the skill of the GEM at forecasting the overall 
evolution of synoptic scale patterns (it proved to be right 
when the Eta was wrong in a couple of the more significant 
winter weather busts in the GSP CWFA over the past two years), 
it does exhibit a cold bias most of the time.  Below are the 
1200 UTC and 1800 UTC surface fields from the 0000 UTC 6 January 
run of the Canadian GEM.  Compare how much farther south the 
surface low is to that of the verified features.  This southward 
bias is what resulted in a colder airmass.  We need to 
investigate whether or not the cold bias might simply be the 
GEM's developing systems a little too far south at our latitude.

Figure 8.  GEM model 12-hour forecast of sea level pressure isobars, 
1000-500 mb thickness, and 12-hour precipitation valid at 1200 UTC 
(upper left) and 1800 UTC (upper right) 6 January 2002.  Surface 
observations plot with sea level pressure and fronts analysis at 
1200 UTC (lower left) and 1800 UTC (lower right) 6 January 2002. 
Click on each image to enlarge.

d. The Short Range Ensemble Forecast (SREF)

Could the Eta ensemble have been a useful tool?  We looked at 
the ensembles, but what they provided was inconclusive.  For 
borderline events where we need to understand in great detail 
how the model is handling the vertical structure of the 
atmosphere, the SREF ensembles are essentially useless as the 
data provided is of too coarse a resolution to be of any good. 
In addition the SREF is a set of models run with a far coarser 
resolution than the operational Eta and AVN, and with a 
different physics package.  This seems more like comparing 
apples to oranges rather than an apple to lots of apples.  The 
ensembles are a useful, if not necessary tool, which could 
greatly help reduce forecast busts, but only if we are provided 
with real-time, high resolution data through AWIPS.

e. Rapid translation/coupled jet

Most of the CWFA was free of significant precipitation by 
1700 UTC on 6 January.  Contrast this with a time section of 
Eta precipitation at Asheville from the 0000 UTC run on 
6 January.  A weak deformation zone turned out to be 
nonexistent as the upper forcing associated with the storm 
raced to the north (as seen by the 9 mb/3 hr pressure falls 
on the surface maps below).

Finally, this was a rapidly moving system.  Precipitation took 
a while to reach the surface.  When all was said and done, the 
duration of precipitation was only 3 to 4 hours, all of it 
associated with lower and mid-tropospheric isentropic upglide. 
There was no backward wrapping deformation zone precipitation 
to speak of.  Again, reference the loop of radar data for the 
event to see how short a period of time it precipitated over 
the area.  In our defense, the color curve used here is a 
little "cooler" that the one we employ in the office.

The fact that the system tracked farther inland than expected 
may also have had to do with the impressive coupled upper jet 
associated with it.  There was tremendous upper divergence and 
attendant upward vertical velocities, associated with this 
feature.  Perhaps the low level low developed farther into the 
cold air as a result of this forcing.  Obviously the track 
errors, which were fairly small over the southeast, became
 much larger as the low deepened and moved north.

3.  Other work to be done

One thing that would be of some benefit would be to take the 
0000 UTC soundings from 6 January and compare the observed 
thickness to the Eta initialization.  Because the 0000 UTC 
6 January run came in much colder than the previous run, it 
would be interesting to see if other upper air soundings from
around the region (GSO, RNK, FFC, BNA and CHS) were appreciably 
different from the Eta. 

4.  Summary and conclusion 

The 6-7 January 2002 winter storm was a good learning experience 
for the staff at WFO GSP.  Taken at face value, the Eta model 
would have supported a forecast of mixed precipitation over the 
mountains of the western Carolinas and northeast Georgia, and 
perhaps some warning criteria icing over the North Carolina 
Foothills. 

However, forecasters erroneously applied a small amount of 
"dynamic cooling" to Eta model soundings, allowing them to 
issue watches and warnings for a considerably larger area.  In 
reality, the Eta already takes such cooling into effect, at least 
in situations dominated by strong low level warm advection.  
Forecasters relied on the Canadian SEF and GEM models during 
the storm as well. While the GEM does a good job in forecasting 
the evolution of an event, both the GEM and SEF frequently have 
a large cold bias over the region, making it tricky to determine 
the proper precipitation type.

The Eta exhibited rather significant temperature fluctuations 
in the runs leading up to the event, more so than the AVN or 
Canadian models.  A short range ensemble forecast may be 
helpful in these cases, though the data currently available 
on the Internet is difficult to use and takes time away from 
the rest of the already busy forecast process. Having this 
data in AWIPS may prove useful.

Finally, damaging accumulations of ice were forecast over much 
of the northwest Piedmont and the Foothills of North Carolina. 
While a mix of sleet, snow and freezing rain verified the 
Foothills, the Piedmont saw only light accumulations of ice. 
It is difficult for in-situ damming events to generate damaging 
ice accumulations.  And, as the Eta runs a little cold, it may 
be in the best interest of the forecasters at GSP to hold off 
in issuing ice storm warning on events that appear "borderline" 
per the Eta low level temperature fields.

Acknowledgements

The images in Figures 6 and 7 were obtained from the NWS State 
College (PA) Lagged Average Forecast Page.  NWS GSP Senior 
Forecaster Harry Gerapetritis provided the temperature data 
for Table 1.  Patrick Moore converted the web page to the 
standard NWS template.