Alan Gerard, Corey Mead, Stephen Miller, Russell Pfost, Peter Wolf
In spite of all of our sophisticated models, up-to-date computer methods, better communications, and exhaustive training, there are times in operational weather forecasting when everything seems to fail. The heavy snowstorm which surprised everyone across northeast Louisiana, central Mississippi, and extreme west Alabama was just one of those times. Light to moderate snow began falling shortly after midnight Sunday, 14 December, 1997, as a deep upper level storm system moved
The storm was not a disaster situation - indeed most residents of Mississippi seemed almost jubilant about the winter weather. Complaints about the terrible forecasting of the event were scarce. Since the storm occurred on Sunday, no children were in school and most regular office workers were at home and not traveling the roads. Even though the snow was heavy, it was a very wet snow with the surface temperature remaining near 32 degrees Fahrenheit. Thus, roads remained wet and still passable in spite of the intensity of the precipitation.
Even so, we at the National Weather Service Forecast Office in Jackson were a little frustrated with our performance. How could such a heavy event occur with little advance notice? How could we have not forecast such an event with our new, improved numerical models and our much better communications and gridded data? True, our forecasts had included a mention of snow and possible accumulations as early as the forecast package issued Thursday afternoon. But that was for the Friday and Friday night time period, and no snow actually occurred then. The only mention of snow we had in the forecast for Saturday night and Sunday was "flurries".
4.8 inches of snow fell officially at the Jackson International Airport, which was the 8th largest on record, and the largest so early in the winter (actually late fall). It was the second largest one day snowfall in the month of December (the largest December snow was 7.5 inches on 29 December, 1929).
This brief paper will try to address what happened and how we may be able to forecast such an event better in the future.
2. Model Performance General Remarks
The three NWS operational models, the Eta, Nested Grid Model (NGM), and aviation (AVN), all did poor jobs forecasting the depth of the 500 hPa low as well as the distribution of moisture with the system. This was surprising, especially in the case of the new Eta model, because the Eta is often touted as a much superior model due to closer grid spacing, the Eta vertical coordinate, and better physics packages. Actually, the Eta made the worst errors in the height of the 500 hPa low center, followed by the NGM and the AVN.
The following table shows relative height errors (+/- 3 m) for the center of the 500 hPa low from the three operational models at 12 UTC Saturday December 13 (48km Eta, 80 km NGM, and AVN) and the 15 UTC meso-eta (29 km resolution) verifying at 00 UTC and 12 UTC Sunday, 14 December, 1997 (the meso-eta was verified with the NGM's analysis):
|12/13 12Z F12||5485||5550||F09 5500||5505|
|12/14 00Z F00||5470||5455||NGM 5445||5445|
|12/13 12Z F24||5460||5555||F21 5505||5470|
|12/14 12Z F00||5420||5420||NGM 5395||5395|
Even more amazing is that if Slidell, Louisiana's, observed 500 hPa height at 5395 m for 12 UTC 14 December is taken as a close approximation to the actual minimum height for the center of the 500 hPa low (which is pretty close to the NGM's verifying analysis), the 24 hour forecast of the 500 hPa low center height is actually 160 m too high!
Moisture analysis was a second and more serious problem with the models. Both the Eta and NGM were consistently drier than the very wet AVN. While the AVN did verify better for the snowstorm event, it had been so wet and cold for the previous three days and did not verify well that Jackson forecasters believed that the Eta and NGM model runs were more accurate.
The 13 December, 1997, 15 UTC run of the meso-eta, in spite of its flaws in forecasting the intensity of the storm system, did a much better job with the moisture distribution. The 15 UTC meso-eta was the only model to correctly depict in advance an axis of moisture coming from the Atlantic off the east coast of Florida across Georgia and Alabama directly into central Mississippi. It also was the only model to show precipitation falling in central Mississippi, although the amounts were much less (generally about 0.15 inch water equivalent) than what actually occurred and the axis of precipitation was displaced too far southeast (fig. 2). This model run, which was not available in real time to Jackson forecasters for communications problems unknown, was the only reasonable model run from which a forecaster might have made a decision to issue a watch for Sunday morning with any respectable lead time. However, a watch issued from the 15 UTC meso-eta likely would have been in the wrong place and likely would not have specified 8 inches of snow.
The model runs from 14 December, 1997, 00 UTC, were much better with the moisture distribution around the storm system. The storm relative isentropic vertical motion was also impressive on the 00 UTC model runs, as is described in detail later. However, snow was already beginning to fall across the Jackson CWA by the time the midnight shift arrived Sunday morning, and little time was therefore available for analysis. Forecasters at Jackson issued a Winter Weather Advisory at 3 AM Sunday, followed by a Heavy Snow Warning at 7 AM Sunday. Fortunately, most of the snow fell after the Heavy Snow Warning was issued.
3. Isentropic Potential Vorticity Considerations
Potential vorticity is defined as the absolute vorticity divided by a static stability parameter.
Potential vorticity is found in abundance in the stratosphere and is created through stratospheric warming associated with the ozone layer (FDC course book, 1996). Ageostrophic circulations associated with upper level jet streaks are a means by which this stratospheric potential vorticity can enter into the troposphere. These "stratospheric intrusions" are often seen as warm pockets on 200 and 300 hPa charts in the vicinity of intense upper level storm systems.
In 1940, Rossby determined that potential vorticity, when evaluated on isentropic surfaces, was conserved for frictionless, adiabatic flow. From the equation above, it can be seen that when stratospheric air descends along an isentrope into a less stable environment, the absolute vorticity of this air parcel must increase to conserve the potential vorticity. Numerous studies have attempted to trace upper tropospheric pockets of potential vorticity and link them to cyclogenesis at the surface. As an example, Barnes and Colman (1994) investigated the possible role of isentropic potential vorticity (IPV) on the rapid development of a storm system over Colorado that was not well handled by short term numerical models.
Since the Mississippi snowfall event began at roughly 0600 UTC on 14 December, the 1200 UTC 13 December and 0000 UTC 14 December runs of the Eta and AVN were used to analyze IPV and its potential effect on the Gulf of Mexico storm system:
a. 1200 UTC 13 December Run
The 1200 UTC 13 December run of the Eta gave no indications of high IPV air in the vicinity of the upper low that was forecast to move across the northern Gulf Coast. Conversely, the 18-hr forecast of the AVN, valid 0600 UTC on the 14th, showed a significant pocket of warm air over southern Louisiana being advected into southern and central Mississippi. To verify
whether or not this warm pocket was a source of potential vorticity, a cross section was constructed through this feature, as denoted by the heavy black line (fig. 3). Indeed, IPV values in excess of 6 Potential Vorticity Units (1 PVU = 10-6 m2 s-1 K kg-1 ) were found extruding into the troposphere in the vicinity of the upper level low (fig. 4). IPV values of 1.5 PVUs or greater usually are associated with stratospheric air (Bluestein, 1993). Values of this magnitude were found as low as 650 hPa at this time. Model-derived potential vorticity advection fields and trajectory analysis (not shown), suggested that this "high IPV" air was being advected northeastward into central Mississippi. This would signal the deepening of the entire storm system through increasing vertical velocities and the associated decreased static stability.
b. 0000 UTC 14 December Run
By the 0000 UTC 14 December run, the 6-hour forecast of the Eta had slowed the progression of the upper level low considerably and developed a pocket of stratospheric air over southern Louisiana. This was a significant change from the previous run and was in good agreement with the 18-hour forecast from the 1200 UTC run of the AVN on the 13th. Meanwhile, the 6-hour forecast of the AVN, valid 0600 UTC 14 December, continued the trend of strong 250 hPa warm advection over southern and central Mississippi.
The Eta appeared to lose the stratospheric air on the 12-hour forecast, as it cooled temperatures considerably to the southeast and east of the 250 hPa low center. The AVN, however, continued to maintain the integrity of this feature on its 12-hour forecast with temperatures in the stratospheric intrusion warming 2 degrees C. As it turned out, the 00-hour forecast of 1200 UTC run on the 14th for both the Eta and AVN, showed that the trend of AVN was the correct solution.
IPV is commonly considered to be a potential tool for forecasting the deepening rate of surface cyclones. However, it may also be used to gauge the deepening potential of the entire storm system, from the surface to the upper troposphere. Forecasters who accepted the AVN solutions for these consecutive model runs and understood the potential synoptic role of IPV may have correctly anticipated continued deepening of the storm system. Although these concepts do not address the amount of moisture available to the system, they do directly diagnose vertical motion and static stability potential.
4. Isentropic Considerations
a. 1200 UTC 13 December Eta Run
The 1200 UTC 13 December run of the Eta model provided few clues to assist in forecasting the impending snow event, even when viewed in an isentropic framework. On the 290K and 295K surfaces, some upward motion was forecast over central and southern Mississippi and northeast Louisiana at 0600 UTC and 1200 UTC 14 December. This vertical motion was more impressive when viewed in a system-relative sense, using the model forecast for system motion of 270 degrees at 15 knots. Using this motion, an axis of 7 to 10 microbars s-1 of upward motion was forecast over southern Mississippi at 0600 UTC and 1200 UTC 14 December. However, this area of vertical motion was forecast to move quickly to the east by 1800 UTC, and was also placed south of the area where the precipitation actually developed.
Even with the upward motion forecast by the 1200 UTC 13 December Eta run, the amount of moisture forecast by the model did not appear nearly sufficient to support significant snow. Condensation pressure deficits during the period of maximum upward motion were forecast to be between 100 and 180 hPa over the area concern. These values are very dry, and would generally be insufficient to support cloud development even in the presence of upward motion. Mixing ratio values on the 295K surface, which the model forecast to be around 750 hPa over the area of concern (the critical surface to examine for snowfall forecasting based on the Garcia technique (1994)), were forecast to be less than 2 g kg-1 between 0600 and 1800 UTC 14 December. The combination of the low mixing ratios and upward motion for only a fairly short period of time would imply very little in the way of snowfall accumulation based on the Garcia technique.
b. 0000 UTC 14 December Eta Run
The 0000 UTC 14 December Eta run indicated a totally different scenario for Mississippi and Louisiana than did the previous run. When using the forecast system motion of 270 degrees at 20 knots, one obtained a well defined axis of system-relative upward vertical motion on the 290K surface from about Natchez to near Jackson, then east to north of Meridian (Fig. 5). The axis was slightly further south on the 295K surface. The upward motion was forecast to persist by the Eta from 0600 to around 1800 UTC, and be strongest between 0600 and 1200 UTC with values of up to 12 microbars s-1 on the 295K surface. The area of upward motion forecast by the model correlated quite well to the position of the observed snowfall, particularly on the 290K surface.
In contrast to the 1200 UTC 13 December run, this run of the Eta indicated that sufficient moisture would be present to support significant snowfall during the time in question. In fact, the Eta forecast that an axis of condensation pressure deficits of 10 hPa or less on the 290K and 295K surfaces would be collocated with the upward motion axis described above through the event. This forecast of near saturation at these levels was a total reversal of the very dry airmass forecast by the previous run of the Eta, and when present with upward motion these low condensation pressure deficit values indicated that precipitation development was likely.
The Garcia technique when used with this run also indicated a much greater snowfall potential than did the previous run. The 0000 UTC run of the Eta showed that the critical surface to examine for moisture per the Garcia technique was again the 295K surface. The average mixing ratio on this surface during the 0600 UTC to 1800 UTC time period using the technique outlined by Garcia was 4 g kg-1. This mixing ratio value in the presence of strong, persistent upward motion yielded a forecast maximum snowfall accumulation during the 0600 UTC to 1800 UTC time period of 8 inches. This matches the observed snowfall very well, and demonstrates that the 0000 UTC 14 December run of the Eta gave a much clearer indication of the heavy snowfall potential during the morning of 14 December than the 1200 UTC 13 December run did.
6. Satellite Considerations
From the World Wide Web (WWW) page maintained by the Cooperative Institute for Meteorological Satellite Studies (CIMSS), University of Wisconsin - Madison (here) the heavy snow band apparently was associated with a deformation zone that formed just to the north and northwest of the cold cloud shield "pivot point" associated with the nearly barotropic (vertically stacked) Gulf closed low. Since deformation zones are normally associated with frontogenesis at some level (Moore, 1998), an in depth study of the frontogenetical aspects of this system would be a good topic for a future paper.
It's tempting to brush this event off as a failure of the models, as shown previously. However, as professional operational meteorologists, it is our job to maintain a weather watch over the defined area of responsibility whether the models are good or not. Snow, especially heavy snow, in the south is relatively rare and residents of our region anxiously anticipate each occurrence. Because winter weather forecasts in the south are always subject to intense scrutiny whether they are right or wrong, we must strive to meet the challenge and do an even better job with them than more routine situations.
Looking back, synoptic scale forecasters at Jackson had few real clues from the models that might have resulted in a better forecast for the snow event in the fourth, third, or second periods. Only older, cliche-type, rules like "There's often a 'surprise' with a closed 500 hPa low" led Jackson forecasters to even keep a mention of snow flurries in the forecast at all. Closer synoptic scale attention to IPV concepts, satellite depiction of moisture distribution and it's agreement with model forecasts, and storm relative isentropic vertical motion will help with future cases like this, but it's still doubtful that a watch for this event would have ever been issued with the model guidance available. A first look by the 'R' shift forecaster at the better model runs at 00 UTC on 14 December could have provided the information necessary for a watch for Sunday, but a watch issued at 10 or 11 PM CST on Saturday night would not have the desired lead time or media distribution as one issued Saturday afternoon.
Mesoscale forecasters, however, had an abundance of information that could have been used if time was available to forecast the amount of snow. The meso-eta model run at 15 UTC on 13 December was the first model run to reasonably depict the vertical motion and moisture fields, and meso-eta model soundings each hour are routinely available at NWSFO Jackson. Satellite imagery in time lapse distinctly showed the deformation zone, and the WSR-88D showed obvious banding structures that led forecasters finally to issue the Heavy Snow Warning and 7 AM CST Sunday, before most of the snow fell.
To sum up the entire experience in a few sentences:
Barnes, S. L., and B. R. Colman, 1993: Quasigeostrophic Diagnosis of Cyclogenesis Associated with a Cutoff Extratropical Cyclone--the Christmas 1987 Storm, Mon. Wea. Rev., 121,1613-1634.
Bluestein, H., 1993: Synoptic-Dynamic Meteorology in Midlatitudes, Volume II, Observations and Theory of Weather Systems, Oxford University Press, pg. 182.
Garcia, Jr., C., 1994: Forecasting Snowfall Using Mixing Ratios on an Isentropic Surface, U.S. Dept. of Commerce, National Oceanic and Atmospheric Administration, National Weather Service, NOAA Technical Memorandum NWS CR-105.
Moore, J. T., 1998: Personal Communication.
Rossby, C.-G., 1940: Planetary Flow Patterns in the Atmosphere, Quart. J. Roy. Meteor. Soc., 66, Suppl., 68-87.
U.S. Dept. of Commerce, National Oceanic and Atmospheric Administration, National Weather Service Training Center Forecaster's Development Course booklet, 1996: Potential Vorticity.
CAPTIONS FOR FIGURES
Fig. 1 Visible Satellite imagery 1545 UTC 15 December 1997
Fig. 2 Average 1000-500 hPa Relative Humidity from the Eta (dotted), NGM (heavy dash-dot), and AVN (thin solid) models for 12 UTC 14 December
Fig. 3 250 hPa heights and temperatures for 06 UTC 14 December 1997. Heavy black line shows cross section for Fig. 4
Fig. 4 Cross section of IPV for 06 UTC 14 December 1997. STJ = subtropical jet PFJ=polar jet
Fig. 5 Storm relative upward vertical motion (microbars/s) on the 290K isentropic surface (heavy black lines) with pressure (thin black lines) and storm relative winds (assuming storm motion vector of 27020)