While heavy snow events in the deep South are rare, on December 27, 1997 it was becoming apparent that a major snow storm might be developing for Mississippi for the evening of the 28th and morning of the 29th. Snow was first mentioned in the fourth period of the forecast issued by NWSFO Jackson at 410 AM on the 27th , based on data from the 0000 UTC December 27 numerical models. After forecasters reviewed data from the subsequent 1200 UTC model run, a Winter Storm Watch was issued at 345 PM for much of central and north Mississippi, valid from 0000 UTC to 1200 UTC on December 29. By the morning of December 28, all the models appeared to be pointing toward a major winter storm; therefore, the Winter Storm Watch was re-issued and expanded west and south to include all of the Jackson forecast area for Sunday night and Monday.
Accumulations of three to six inches were expected with local amounts to eight inches over much of the forecast area. Two to four inches of snow were expected elsewhere. By 4 PM on December 28 it still appeared the northeast corner of Mississippi would get four to six inches, so a Heavy Snow Warning was issued for that section of the forecast area for Sunday night and Monday. A Snow Advisory was issued for the remainder of the forecast area.
Forecasters felt they had done the best job possible in anticipating this significant event, but it turned out the northeast corner of Mississippi was the only area to receive any snowfall at all, and even there no significant accumulation was reported. The remainder of the forecast area receive from one-half to one inch of rain. This paper will examine the rationale behind the forecasts for heavy snow and discuss tools which might help forecasters better handle similar events in the future.
2. Synoptic Pattern
Throughout this event all three short- and medium-range models (Eta, Nested Grid [NGM], and Aviation [AVN]) were in agreement with the large scale pattern over the continental United States, and they were generally accepted. References to the models in this section will imply all three of these numerical models.
For three consecutive model runs (0000 UTC on December 27 to 0000 UTC on December 28) a digging 500 mb trough was forecast to plunge deep into the lower Mississippi Valley. By 0000 UTC on December 29 (Fig. 1), a 500 mb low (5370 m) was forecast to be centered near Shreveport, with strong and diffluent southwest flow aloft over the Jackson forecast area. Temperatures at 500 mb were forecast to be below -30 oC near the center of the low. The 850 mb -5 oC isotherm was forecast to be in the northwest Gulf of Mexico by 1200 UTC on December 29 (Fig. 2). Along with this, 1000-500 mb thickness values generally consistent with snow in the South (e.g., NWSTC 1993) were also forecast to extend well into the northern Gulf (Fig. 2). Mean relative humidity was forecast to increase as low level southerly flow tapped moisture from the Gulf of Mexico. Eta model data taken from PCGRIDDS and GARP showed total precipitation (water equivalent) amounts from 0.50 to 1.25 in for the event.
It appeared all the synoptic scale indicators for a significant snow event would be in place for the night of December 28-29. In the next section, we will show how the models performed, and what in hindsight may have helped forecasters recognize that significant snow might not occur.
3. Model Forecasts and Verification
Most parameters from the numerical models during this event seemed to favor significant snowfall after 0000 UTC on December 29. In general, thickness parameters appeared favorable for snow, as did the model forecasts of surface and upper air features.
Although all the 1200 UTC runs of the models on December 27 showed that most of the Jackson forecast area would see 1000-500 mb thickness values below 5400 m, the value typically cited as the first guess for a rain/snow line, after 36-hr (by 0000 UTC on the 29th) there was some disagreement among the models concerning the track of the surface low. The Eta forecast that a surface low would develop over southeast Louisiana around 0000 UTC on the 29th, and track into Alabama over the ensuing 12 hours. The AVN and NGM, however, showed a surface low track much farther to the north, through Arkansas and Tennessee.
All the models were forecasting surface temperatures across northern and central Mississippi to fall in to the mid to upper 30s after 36-hr, marginal for snowfall. However, an Eta solution for the surface pattern would have led to cold air advection in the lower levels, while an AVN/NGM solution would have yielded warm air advection in the lower levels. The Eta solution of low level cold air advection combined with the very cold 1000-500 mb thicknesses yielded a much higher potential for significant snow. The Eta solution was accepted, leading to the decision to issue a winter storm watch with this model run.
Twelve hours later, when we received the 0000 UTC model runs on December28, this decision appeared to have been the correct one. With this run, all the models tended toward the previous Eta solution, showing a deepening surface low moving along the Gulf Coast between 0000 UTC and 1200 UTC December 29. The Eta showed a 1007 mb low developing over eastern Louisiana at 0000 UTC on the 29th. This low was forecast to deepen to 1002 mb as it moved to the vicinity of Gulfport, Mississippi, by 0600 UTC, then to southwest Alabama by 1200 UTC. Based on past events, this surface low track is favorable for snowfall across much of Mississippi. Combined again with 1000-500 mb thickness values forecast to be below 5400 m and 850 mb temperatures below 0 oC, significant snow was forecast across much of Mississippi with these model runs.
In the end, the more northerly track of the surface low suggested by the NGM and AVN model runs from 1200 UTC on December 27 verified closest to what actually occurred. In fact, even these models were too far south with the track of the surface low.
Around 0000 UTC on the 29th a 1004 mb surface low developed along the north central Arkansas/south central Missouri border. This low deepened to 1002 mb as it moved east- northeast to near Evansville, Indiana. Clearly, the 0000 UTC December 28 run of the models trended in the wrong direction when showing surface low development along the Gulf Coast. The 1200 UTC runs on the 28th did finally discern this more northerly trend, but even they were too far south, keeping the surface low track across northern Mississippi and Alabama.
With regard to forecasts of thermal fields, as can be seen in Table 1, the model forecasts of 1000-500 mb thickness actually verified too high, in spite of the farther north track of the surface low. However, the farther north track of the low had a great impact on low-level temperatures. As mentioned above, model forecasts for surface temperatures for the night of December 28-29 were generally in the mid to upper 30s. In reality, observed values were generally in the 40s, much too warm for any wintery precipitation.
4. Forecasts of Thickness and Freezing Level Heights
a. Thickness Forecasts
The track of the surface low well to the north of the track forecast by the numerical models contributed to more low-level warm air advection than anticipated, and hence warmer low-level temperatures, and more significantly - rain instead of snow. In spite of the poor model performance, other clues which might have given forecasters more insight into the precipitation type forecast can be obtained by a detailed post-event examination of the thickness and temperature profile forecasts from the models.
The mid-level thickness values, i.e., 1000-500 mb, 850-700 mb, and 1000-700 mb, were forecast by all the models to be favorable for snow over most of Mississippi from 0000 UTC December 29 on. NWSTC (1993) indicates that the first guess for the rain/snow line for these thickness layers are 5400 m, 1540 m, and 2840 m, respectively. In general, these thickness lines were forecast to be south of the forecast area at 0000 UTC on the 29th, with the airmass becoming progressively colder after this time. The actual verification of these thickness values showed the model forecasts for these thickness values were actually slightly too warm (see Table 1 for 1000-500 mb thickness forecast verification).
Hence, with the mid-level temperature profile forecast to be cold enough for snow, the main question would lie in the lower levels of the atmosphere. The thickness layer used to analyze the lower levels is 1000-850 mb, and the first guess rain/snow line for this layer is 1300 m (NWSTC 1993). 1000-850 mb thickness values were forecast to remain near or slightly above 1300 m throughout the event, ranging from 1305 m to 1320 m during the time most of the precipitation was forecast to occur. However, these values are still below the mainly rain value of 1325 m (NWSTC 1993), and combined with the very cold mid-level thicknesses and the potential for cooling due to melting and dynamic cooling provided by intense upward vertical motion, it was felt that snow would still be the predominant precipitation type. It should be noted, however, that a recently completed study (Pfost, et. al. 1998) indicates that significant snow (1 in or greater) at Jackson occurs with a mean 1000-850 mb thickness of 1291 m, and during the time of study (since 1950) did not occur with a 1000-850 mb thickness at or above 1305 m. Unfortunately, this information was not available to forecasters prior to the event.
b. Freezing Level Forecasts
In spite of the generally favorable thickness forecasts described above, detailed examination of the temperature profiles forecast by the models reveal that the depth of the low-level warm layer forecast by the model may have been too great to allow snow to fall as the predominant precipitation type. This may have been the case even if the models had been correct when showing a track farther south of the surface low. This can most clearly be seen in examining the forecast freezing level heights from the models.
Table 1 shows the model forecasts for freezing level height for 0000 UTC December 29, by which time significant precipitation should have been in progress, ranged from 2800 ft AGL to 4900 ft AGL at Jackson. Similar values were forecast for Tupelo (TUP). As is discussed by NWSTC (1993), the possibility of snow being the predominant precipitation type is 50 percent when the freezing level is at 920 ft AGL, and on average a freezing level height of 1200 ft AGL will assure that rain will be the predominant precipitation type.
While the possibility that the models were incorrect in their low-level temperature forecast would certainly have to be taken into account with all the other favorable model forecast parameters for snow, a detailed examination of freezing level heights showing such high freezing levels might have given forecasters some hesitation with regard to the event being one with snow as the predominant precipitation type. This is especially true given that these freezing level heights, for the most part, were forecast with nearly saturated conditions, meaning that evaporative cooling would not play a significant role in lowering freezing level heights. Cooling due to melting does not generally effect large changes in the thermal profiles, and would likely have been insufficient to lower freezing level heights enough to make snow the predominant precipitation type (NWSTC 1998).
c. Forecasting Tools Relating to Freezing Level Heights
The above discussion indicates the importance of freezing level height data in forecasting rain versus snow events; however, model forecasts for heights of the freezing level cannot always be easily determined in the operational environment. Model soundings, which in general are the best tool for forecasting precipitation type, are usually viewed using software such as GARP or NTRANS, programs which do not give cursor output for temperature and height as SHARP (Hart and Korotky 1993) does. This makes estimating the exact forecast height for freezing level difficult. The software package BUFKIT (Mahoney 1998) does give values for freezing level height, and this program may be the best way to view data regarding forecasts for the height of the freezing level. This program was not being used operationally in Jackson at the time of this event, however.
Forecast freezing level heights can also be calculated through the use of GEMPAK programs. Figure 3 shows the output from a script (Appendices 1 and 2) which was recently written at Jackson. The GEMPAK script in Appendix 1 uses the GEMPAK-like program GDTINT (Barker 1998) to create a freezing level height forecast grid through vertical interpolation. This grid can then be viewed in GARP in a planar mode by using the field description file (FDF) in Appendix 2, with contours drawn every 500 ft. Such a forecast product should prove invaluable in determining the potential for snow versus rain when the low-level temperature structure is forecast to be marginal. As can be seen in Fig. 3, for this case the Eta forecast freezing level heights at 0000 UTC on the 29th to range from 2000 ft over far northeast Mississippi to near 5500 ft over south-central Mississippi, values too high to support any frozen precipitation.
The 950 mb level data in gridded model output provides another method for making a rough estimate of the forecast height of the freezing level. NWSTC (1993) indicates that snow is the predominant precipitation type 50 percent of the time when the freezing level is 35 mb AGL, and the probability decreases to virtually nil when the freezing level reaches approximately 50 mb AGL. Since station pressure averages around 1000 mb at Jackson, it seems clear that snow will be very unlikely there if the 950 mb temperature is above 0 oC. Hence, forecasts of 950 mb temperature viewed in planar mode should provide a first guess at determining if the depth of the warm layer in the lower levels would preclude snowfall. These conclusions, however, would not apply when station pressure was significantly lower than normal (below 985 mb). In this particular case, the 950 mb temperature was forecast by the models to be above 0oC over all but northeast Mississippi throughout the anticipated precipitation event.
It should be noted that all of the above conclusions concerning the forecast freezing level apply only if the lower levels of the atmosphere are near saturation. If not, evaporative cooling can play a significant role in lowering the freezing level height once precipitation develops. In these cases, the height of the wet bulb zero (WBZ) can be used to gauge where the freezing level height will be after precipitation induced evaporative cooling. Again, BUFKIT (Mahoney 1998) can be used to view model forecasts for height of the WBZ.
Forecasting winter precipitation is always a difficult challenge. The challenge is especially tough in the deep South where such events are relatively rare, meaning that the forecast has an even greater importance to the public and emergency managers who use forecasts regarding significant winter events. Hence, when a "busted" forecast for a significant snow storm occurs, a post-analysis is useful to help analyze what went wrong and hopefully improve forecast techniques for similar events in the future.
The post-analysis of this "non-event" yields two major points. First, the models had a very difficult time in forecasting surface cyclogenesis with this event. Successive model runs actually trended in the wrong direction by showing a surface low track farther south along the Gulf Coast, while the actual track of the low was across the middle Mississippi Valley. The models came into better agreement with each successive run, showing the track closer to the coast, but anticipating such a large model error would have been quite difficult. Clearly, had the models advertised the track of the low farther to the north, the potential for warmer temperatures in the boundary layer would have been more readily apparent, likely leading to a less aggressive forecast with regard to snow.
The second major conclusion is that model forecasts of the freezing level height seemed to indicate that the very lowest levels of the atmosphere would be too warm to support an all-snow event, even assuming diabatic cooling processes such as cooling due to melting or dynamic cooling. However, forecast products and programs capable of viewing the model forecasts of freezing level were quite limited during the time of this event, thus making it difficult to arrive at such a conclusion. Implementation of programs such as BUFKIT and macros for GARP should make such data much easier to view in the future, and hopefully AWIPS will provide an even easier means of viewing these data. This should greatly enhance the forecaster's ability to analyze the low level temperature structure when forecasting rain versus snow events.
Barker, T., 1997. GDTINT documentation. Available via FTP from the SOO/SAC home page.
Mahoney, E., 1998. BUFKIT98 documentation. Available via the World Wide WEB at http://www.wbuf.noaa.gov/bufkit/bufkit97.htm
Hart, J.A. and W. Korotky, 1991. The SHARP workstation users guide. NWS, NOAA, U.S. Department of Commerce, 30 pp.
National Weather Service Training Center, 1993. Winter weather: precipitation type. National Weather Service, Kansas City, MO, 29 pp.
_____, 1998. An introduction to winter precipitation nowcasting. CD-ROM training package.
Pfost, R.L., A.E. Gerard, S.A. Miller, E. Cannon and R. Guyton, 1998. A heavy precipitation climatology for the Jackson, MS, forecast area. Unpublished study.
Table 1. Table showing forecast 1000-500 mb thickness, 1000-850 mb thickness, 850 mb temperature, and freezing level height, from various model runs for Jackson, MS at 0000 UTC 29 December 1997. Bottom row is the observed values for 0000 UTC 29 December. Thickness values are in meters, temperatures are in degrees Celsius, and freezing level heights are in feet above ground level.
Figure 1. 0000 UTC 28 December 1997 Eta forecast of 500 mb heights in dm (solid), temperatures in degrees Celsius (dashed), and winds valid 0000 UTC 29 December.
Figure 2. 0000 UTC 28 December 1997 Eta forecast of 850 mb temperatures in degrees Celsius (solid) and 1000-500 mb thickness in dm (dashed) valid 1200 UTC 29 December.
Figure 3. Example of planar plot of freezing level heights generated by GEMPAK script and viewed through GARP field description file as developed at NWSFO Jackson.
Appendix 1. Script which creates the freezing level height forecast grid using GEMPAK.
Appendix 2. Field description file (FDF) which allows display of the freezing level height forecasts in GARP.