a. Snowfall data and event definition

Daily snowfall records for the period 1975–2000 served as the source of the daily snowfall data. We extracted these data from the National Climatic Data Center’s Cooperative Summary of the Day CD-ROM (NCDC 2002) for 121 cooperative observer (coop) stations in the southern Appalachian Mountains (Fig. 3). We defined a snowfall event as having occurred if at least one coop station in the region reported snow accumulation on a given date. To improve the temporal resolution of each event, we referenced hourly surface observation summaries from five nearby first-order stations (Fig. 3). From interpretation of these data, we were able to approximate the onset, maturation, and ending times as well as the duration of reported snowfall across the region. Assessment of the maturation time involved determining the hour in which the spatial extent of snowfall was greatest across the network of first-order stations. An event remained active if precipitation was reported during a 6-h period. When precipitation was no longer reported at any of the five first-order stations for more than 6 h, we defined the event as having ended at the hour precipitation was last reported.

Because of the large areal extent and significant topographic diversity within the southern Appalachians, the study area was divided into 14 snow regions to facilitate intraregional comparisons. Coop stations were grouped together into snow regions based on similarities in snowfall patterns, elevation, and topography (Fig. 4). In most cases these regions correspond to zone groupings that the National Weather Service (NWS) uses for forecast products. However, it is important to note that Snow Region 14 is not contiguous, but rather consists of all locations in the study area with elevations greater than 1220 m (4000 ft). These areas are largely limited to the mountains of eastern Tennessee and western North Carolina. Recent work (e.g., Perry 2006) has shown that higher-elevation stations (e.g., Mount Mitchell, North Carolina) average as much as 75 cm (1700%) more NWFS on an annual basis than nearby valley stations (e.g., Asheville, North Carolina), suggesting dramatic differences in snowfall climatologies. Pronounced differences are also evident between northwest and southeast slopes and between southern and northern areas, thereby necessitating a high degree of complexity in the number and delineation of snow regions. Mean and maximum event snowfall totals for those stations reporting snow were calculated for each snow region. Although useful for analyzing the general spatial patterns across the southern Appalachians, it is important to recognize that considerable variability of snowfall still occurs within snow regions.

b. Identification of NWFS events

We used gridded (2.5° × 2.5° latitude–longitude mesh), twice-daily synoptic fields that were extracted from CDs containing the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis dataset (Kalnay et al. 1996) to obtain the u and υ components of 850-hPa wind direction and vertical velocity ω. These fields were spatially interpolated to the center of each snowfall event. Using the 0000 and 1200 UTC gridded synoptic fields, we undertook a temporal interpolation to estimate field values during the event maturation time. We employed an inverse distance technique to carry out all spatial and temporal interpolations. Using these spatially and temporally interpolated data, we identified 432 NWFS events during this period on the basis of northwesterly (270°–360°) 850-hPa wind direction at maturation hour. The 850-hPa level, found at approximately 1450 m, is a good indication of the mean wind direction between 1000 and 2000 m, where much of the orographic enhancement in association with NWFS likely occurs.

We further narrowed this sample of 432 NWFS events by identifying only those events with 700-hPa synoptic-scale sinking motions (e.g., vertical velocity > 0) with durations between 2 and 3 days. Events with rising motions were eliminated because a portion of the precipitation may have been tied to synoptic features (e.g., approaching troughs), thus confounding the evaluation of the importance of air trajectories on snowfall totals. Exceptionally short and long events were also eliminated from the sample in order to control for the influences of event duration (i.e., exceptionally heavy snowfall totals resulting from an extended period of snowfall). We also excluded all events in the months of April and May because of the complicating effects of diurnal instability, which can result in locally intense snowshowers. The remaining 191 NWFS events served as the sample for the backward air-trajectory analysis.

c. Trajectory analysis

The National Oceanic and Atmospheric Administration’s Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) tool (information available online at http://www.arl.noaa.gov/ready/hysplit4.html) was utilized to calculate 72-h backward air trajectories for air parcels at 1450 m (4757 ft), or approximately the 850-hPa level. The HYSPLIT tool uses u and υ components of the horizontal wind, temperature, height, and pressure at different levels of the atmosphere to calculate the backward trajectories (Draxler and Hess 1998). We used the weighted spatial mean or centroid of each event and the event maturation time as the starting point and time for the backward air-trajectory analysis. Although finer-scaled data could be assessed using the HYSPLIT tool, NCEP–NCAR reanalysis data (2.5° × 2.5 latitude–longitude) were used because of their availability over the entire study period. As shown in Fig. 5, the trajectories computed from these data compared favorably to those calculated from higher-resolution final analysis (FNL; 191 km) and Eta Data Assimilation System (EDAS; 80 km) data.

A visual analysis of the computed trajectory allowed us to classify each 6-h segment by counting the number of segments contained within each quadrangle of our trajectory classification grid (Fig. 6). If the plotted air trajectory associated with each snowfall event resided in a respective quadrangle for the specified minimum number of hours (Table 2), we classified the trajectory according to the quadrangle number. The Great Lakes (grid 3) are roughly divided into the western lakes of Superior and Michigan and the eastern lakes of Huron and Erie in order to facilitate intraregional comparisons. The Lake Ontario region is not explicitly delineated in this study because of the relatively infrequent occurrence of north-northeasterly flow from this region. Because the western and eastern Great Lakes subregions (grids 3.1 and 3.2) are the same spatial dimensions, it is possible to directly compare the frequency, snowfall totals, and synoptic characteristics of air trajectories. The relatively course spatial resolution of the NCEP–NCAR reanalysis data used to compute the backward air trajectories also supports using a more generalized classification grid, and not focusing on the precise land–water boundary.

The hourly latitude, longitude, and height (hPa) coordinates associated with each event backward air trajectory were used to develop composite trajectories for each region in the trajectory classification scheme. Additional composite trajectories were developed for T − 12 (12 h prior to event maturation) and T + 12 (12 h following event maturation). For each event with a GLC (≥6 h of the trajectory in trajectory grids 3.1 or 3.2), we consulted the Great Lakes Ice Atlas DVD (Assel 2003) and noted the percent ice cover for the western and eastern Great Lakes. Composite mean and maximum snowfall totals for GLC events in the top quartile of percent ice cover (≥25%) were compared with those events in the bottom three quartiles (<25%) to evaluate the importance of ice cover. Last, we used NCEP–NCAR reanalysis data (Kalnay et al. 1996) to develop composite plots of the 1000-hPa height for each trajectory class. For each group we calculated composite mean and maximum snowfall totals, composite backward air trajectories for each region, and also values of various synoptic fields. The purpose of this analysis was to determine if part of the enhanced snowfall could be associated with antecedent air trajectories across the lakes. For those events with a GLC, we compared snowfall totals between the North Carolina high country (region 8) and the Appalachian Plateau of West Virginia (region 13) to identify events in which snowfall may have been enhanced in one of these regions due to a GLC.

d. Sounding and synoptic fields

To provide details on vertical thermal and moisture profiles, an analysis of vertical soundings for Huntington, West Virginia (HTS) (Fig. 6), was undertaken. Unfortunately, upper-air data were only available through 1995 at this site, thereby limiting the sample to 155 of the original 191 events. These soundings provided much greater detail regarding the vertical profiles of temperature and moisture in association with NWFS events. We chose HTS because it was the closest sounding station upwind of the study area and therefore provides the best assessment of the vertical profiles during periods of northwest flow. Table 1 summarizes the sounding variables that were calculated. We defined the moist layer as that portion of the lower troposphere with mean relative humidity greater than 80%. The mean temperature of the moist layer was calculated by averaging the temperatures at the lower and upper heights of the moist layer. The presence or absence of moderately low temperatures (−14° to −17°C) in the moist layer was also noted as ice crystal growth is maximized in the dendritic range of −14° to −17°C (Ryan et al. 1976; Pruppacher and Klett 1997; Fukuta and Takahashi 1999); furthermore, heavy snowfall is often associated with moist-layer temperatures within the dendritic temperature range (Auer and White 1982). Additionally, forecasters at the Greer, South Carolina, NWS Office have found that heavy NWFS events are more likely to occur in the North Carolina mountains when moist-layer temperatures coincide with the dendritic temperature range (Lee 2005). Values for relative humidity, temperature, and height were taken from the University of Wyoming archived text soundings (online at http://weather.uwyo.edu/upperair/sounding.html).

Table 1 lists 15 synoptic fields extracted from the NCEP–NCAR reanalysis dataset that will be incorporated into the analysis. Recent work (Perry and Konrad 2004) suggests that many of these synoptic fields are tied to heavy NWFS in the southern Appalachians and, thus, may be tied to certain air trajectories. In particular, high values of relative humidity in the lower troposphere are critical, as are lower heights from 1000 to 500 hPa. Modest, as opposed to strong, cold advection also appears to be tied to heavy NWFS. The substantial pressure rises and subsidence associated with strong cold advection may work to counteract available low-level lift and decrease snowfall totals. Upper-level divergence has also been linked to heavier NWFS (Perry and Konrad 2004). Differential cyclonic vorticity advection below 500 hPa is known to help raise the height of the capping inversion in situations of lake-effect snow in the Great Lakes (Niziol 1989; Lackmann 2001), and a vorticity maximum also may play a role in creating an environment more conducive for heavy NWFS in the southern Appalachians. The sounding and synoptic fields together provide a comprehensive assessment of the vertical and synoptic environments associated with different air trajectories.

a. Trajectory classification

Table 2 summarizes the results of the trajectory classification. The most common air trajectory associated with NWFS in the southern Appalachians is trajectory class 3.1 (30.9%), followed closely by class 2 (26.7%). Interestingly, almost half (47.1%) of the 191 NWFS events analyzed in this study exhibit a GLC (trajectory class 3) and approximately three-quarters (73.8%) of the events display a northwesterly trajectory (trajectory classes 2 or 3). Besides subclasses 3.2 and 3.3 of trajectory class 3, the most infrequent trajectories are trajectory classes 1 and 4, which are characterized by a more westerly trajectory, as opposed to northwesterly, at event maturation. The remaining events did not meet any of the trajectory classification requirements and were therefore grouped together in trajectory class 5. The temporal pattern of trajectory classes by month (Fig. 7) indicates significant variability within and among the classes. Most of the trajectory class 1 events were observed during the coldest months of the winter (i.e., December, January, and February), when westerly trajectories are associated with air sufficiently cold for snowfall to occur. Trajectory class 3.1, on the other hand, displayed a relatively equal monthly distribution, as the other trajectory classes were largely limited to the months of December, January, February, and March. Interestingly, the overwhelming majority of events in the month of November were associated with GLC trajectories. The lake–air temperature contrast, which is greatest in the late autumn and early winter when the sensible and latent heat fluxes are maximized (Niziol et al. 1995), may contribute somewhat to the observed monthly patterns for trajectory class 3.1. The very low number of February events for trajectory class 2 is noteworthy, as is the spike of events for trajectory class 3.3 in the same month. Both of these anomalies likely result from small sample sizes rather than monthly variations in the synoptic environment.

b. Composite trajectory plots

Northwesterly trajectories, or trajectory classes 2 and 3, will be discussed first, followed by the more westerly trajectories of trajectory classes 1 and 4. Composite trajectory plots (Fig. 8) for trajectory class 2 show a backward air trajectory that extends to the northwest, but well to the west of the western Great Lakes. Subsidence is noted in the vicinity of the Ohio River Valley, with the air parcels descending from 800 hPa to slightly below 850 hPa, before ascending slightly as a result of orographic forcing. Trajectory class 3.1 is very similar to class 2, but in this case part of the trajectory at event maturation (T0) and T + 12 crosses the extreme southern portion of Lake Michigan. Subsidence is also evident from 800 to 850 hPa over the 72-h period, although it is not quite as abrupt in the vicinity of the Ohio Valley as with class 2. Air parcels in association with trajectory class 3.2 spend a considerable amount of time in the vicinity of the Great Lakes, with the backward air trajectory centered over the eastern Great Lakes at T0. Air parcels are also near the surface for an extended period of time, with modest lift occurring as a result of the topography. A veering of the trajectory to a more northerly component is evident between T − 12 and T0 for all three composite trajectories, with less consistency noted between T0 and T + 12, particularly in trajectory class 3.2, where backing is noted. The position of the composite cyclone is related to the orientation of the air trajectories (Fig. 8). Cyclones positioned farther south (e.g., off the mid-Atlantic coast) favor a more northerly low-level airflow with a GLC while cyclones positioned farther north off the New England coast favor a more westerly trajectory that is positioned west of the Great Lakes.

Trajectory class 1 (Fig. 9) extends west from the study area in association with strong subsidence and backing of winds as air parcels descend from 650 to 850 hPa over the 72-h period. The composite 1000-hPa height field reveals a strong anticyclone located to the west, with a departing surface cyclone to the east. Several of the 72-h backward air trajectories extend as far west as the Pacific Ocean, suggesting that moisture for many of these events may be of Pacific origin. A veering of the flow from southwest to northwest is evident from the temporal trajectory analysis, consistent with the temporal pattern for trajectory class 2. Trajectory class 4 (Fig. 9) occurs in situations of weak southwest flow at T − 12 that becomes northwest near event maturation, and then increases and becomes a more northwesterly trajectory as the event approaches T + 12. The vertical trajectory plot clearly shows that air parcels are lifted from the surface in this case, which stands in contrast to the pronounced subsidence evident in trajectory class 2. The composite 1000-hPa height pattern indicates a cyclone to the east and southeast off the North Carolina–South Carolina coast, which is evidently far enough away to preclude synoptic-scale lift, but close enough to result in northwest flow at 850 hPa. It is important to recognize that the more westerly trajectories characterized by classes 1 and 4 are infrequent, together representing just 16.3% of the sample.

c. Composite snowfall totals

Snowfall totals exhibit substantial variability among the different trajectory classes and across the study area (Fig. 10). The high country (region 8), central and northern plateau (regions 12 and 13, respectively), and highest elevations (region 14) received the greatest amounts of snowfall across the different trajectories. The lower-elevation regions, and particularly those on the leeward slopes [southern, central, and northern foothills (regions 3, 9, and 11, respectively)], received no measurable snowfall across most of the trajectories. These results are consistent with the climatology of orographic precipitation and those reported by Perry and Konrad (2006), in which the greatest snowfall totals for NWFS events occur along the windward slopes and at higher elevations. The lowest composite mean snowfall totals are found in association with trajectory classes 1 and 2 across all snow regions, whereas the highest snowfall totals generally are found with trajectory classes 3.2, 4, and, in some cases, 5. The low composite snowfall totals in association with trajectory class 1, in particular, may be tied to the strong subsidence along a westerly trajectory. Trajectory class 4 stands out as producing the highest snowfall totals across all but one of the snow regions, suggesting the presence of more abundant moisture and/or higher relative humidity.

The southern, central, and northern foothills (regions 3, 9, and 11, respectively) exhibit the lowest composite mean snowfall totals across most of the trajectory classes, although the totals are substantially higher for trajectory classes 4 and 5. These results should be viewed with caution, however, as synoptic-scale lift in association with a coastal low and broad upper trough apparently extended just far enough to the west to bring significant snowfall to these areas in a small number of events (3 out of the 16 in the sample). Interestingly, some of the lowest composite mean snowfall totals across the different trajectory classes are also found in the New River Valley of southwestern Virginia (region 10). Even though the mean elevation is relatively high (686 m), this snow region is heavily shadowed in periods of northwest flow by the higher terrain of the northern plateau (region 13) to the west and northwest. Therefore, it is rare for accumulating snowfall to occur in the absence of synoptic-scale ascent in periods of low-level northwest flow. A general latitudinal gradient in the composite mean snowfall, with the exception of the New River Valley (region 10), is evident across the different trajectory classes as northern snow regions at the same elevation and exposure display higher totals [e.g., southern plateau (region 6) versus central plateau (region 12)]. This latitudinal gradient likely continues northward to the central Appalachian Mountains of northern West Virginia, western Maryland, and Pennsylvania.

d. Composite sounding and synoptic fields

Results from the sounding analysis and the NCEP–NCAR reanalysis data provide some additional insight into the observed snowfall patterns by trajectory class (Table 3). The lower snowfall totals observed in trajectory classes 1 and 2 in particular are associated with a “thinner” moist layer, lower values of relative humidity and moisture at 850 hPa, and higher values of cold advection. Previous work (e.g., Auer and White 1982) has demonstrated the importance of a deep moist layer in producing heavy orographic snowfall in the western United States. Perry and Konrad (2004) also found that the heavy NWFS events in the southern Appalachians were tied to higher values of lower-tropospheric relative humidity, so the results of this study are consistent with previous work. The higher values of 850-hPa cold advection for trajectory classes 1 and 2 also imply stronger subsidence and, therefore, a less favorable synoptic environment for snowfall in the lower troposphere. The 500-hPa vorticity, though still positive (cyclonic), is lower in trajectory classes 1 and 2, while the remaining trajectory classes are tied to higher absolute values of 500-hPa cyclonic vorticity.

Several of the sounding and synoptic variables in association with the higher snowfall totals of trajectory class 3.2 are quite distinctive from the other trajectory classes. Temperatures were in the dendritic temperature range for only approximately one-third (38%) of the events, and the mean temperature of the moist layer was also considerably greater than the other trajectory classes. In addition, the 850-hPa mixing ratio was exceptionally high (2.66 g kg⁻¹), as were 500-hPa relative humidity, precipitable water, 1000–500-hPa mean relative humidity, and 850-hPa θ_e. The warmth of the moist layer is particularly striking, since previous work has demonstrated that ice crystal growth is most rapid and efficient in the dendritic temperature range (Ryan et al. 1976; Pruppacher and Klett 1997; Fukuta and Takahashi 1999); furthermore, Auer and White (1982) have tied heavy snowfall to moist-layer temperatures within the dendritic growth range. These results indicate that heavier NWFS can occur—albeit infrequently—with moist-layer temperatures outside of the dendritic growth range. However, a deep moist layer with exceptionally high values of lower-tropospheric moisture appears to be needed to make up for the reduced precipitation efficiency.

Trajectory class 4 is also associated with a considerably thicker moist layer, much higher values of humidity and moisture at 850 and 500 hPa, and lower values of cold advection when compared with the remaining trajectory classes. Interestingly, however, the 850-hPa wind speed for trajectory class 4 is more than 4 m s⁻¹ below the speeds for the other classes. This reduction in 850-hPa wind speed is associated with a less pronounced pressure gradient, apparently due to a weaker surface anticyclone to the west (Fig. 9). The weak flow is evident in the composite trajectory plots (Fig. 8), with the 72-h backward trajectories extending only slightly west of the Mississippi River. In this trajectory class, which once again occurs relatively infrequently (n = 16, or 8.4% of the sample) compared with the others, abundant and deep low-level moisture is lifted orographically, resulting in heavier snowfall. Therefore, significant NWFS can also result across the region when high values of relative humidity and moisture throughout the lower troposphere are lifted, albeit weakly compared with other NWFS events.

e. Influence of the Great Lakes

Comparison of the northwesterly trajectories of class 2 (no GLC) with classes 3.1, 3.2, and 3.3 (GLC) reveals that trajectories with a GLC are associated with higher composite mean and maximum snowfall totals across windward slopes and higher elevations of the study area, with the influence the greatest in the high country and northern plateau (regions 8 and 13, respectively) (Fig. 11). In both of these regions, the composite mean and maximum snowfall totals are significantly greater (p < 0.05) for events with a GLC, with the composite maximum snowfall 1.8–2.1 cm greater (150%–135%). Modest increases are noted in several of the other snow regions, particularly those with portions along the windward slopes. At the lower elevations and along the leeward slopes, very little if any snowfall occurs with either trajectory, highlighting the importance of topographic controls on the spatial patterns of snowfall for events with a northwesterly trajectory.

The increase in snowfall totals in portions of the study area is tied to higher values of moisture and humidity, particularly in the low levels (Table 4). The difference is most pronounced and statistically significant (p < 0.05) with the sounding variables of thickness of the moist layer and lapse rate above the moist layer. The synoptic variables of precipitable water, 850-hPa relative humidity, 1000–500-hPa mean relative humidity, and 850-hPa mixing ratio also display significant (p < 0.05) differences, suggesting a more moist and saturated lower troposphere for events with a GLC. Significant differences in 500-hPa relative vorticity are also evident. Events without a GLC typically are accompanied by lower values of cyclonic vorticity, whereas events with a GLC display higher values of cyclonic vorticity. In GLC events, a vorticity maximum at 500 hPa rotating through the base of the 500-hPa trough may help to raise the height of the low-level capping inversion, as suggested by Lackmann (2001) and Niziol (1989). Our sample, however, was restricted to events in which subsidence was observed at 700 hPa (i.e., no approaching troughs to provide synoptic-scale ascent). Interestingly, no significant differences in 700-hPa vertical velocity, 850-hPa temperature, or percent of events in the dendritic temperature range were noted.

The composite plots of 850-hPa relative humidity and 1000–500-hPa mean relative humidity for trajectory classes 2 and 3 (Fig. 12) further highlight the significant differences in lower-tropospheric relative humidity between the two trajectory classes. In both classes, the highest values of relative humidity are generally found to the northeast, but a distinct ridge of higher values extends southwest toward the Ohio Valley in association with trajectory class 3. The greatest mean differences between the two fields for both trajectory classes are located over western North and South Carolina, as higher values of lower-tropospheric relative humidity in these areas in particular are tied to trajectories with a GLC. These composite plots also show that lower-tropospheric relative humidity is higher across a large area downwind of the Great Lakes for trajectories with a GLC.

Interestingly, although higher values of lower-tropospheric moisture and humidity across the southern Appalachians occur in association with trajectory class 3, significant (at p < 0.05) increases in composite mean and maximum snowfall only occur in the high country and northern plateau (regions 8 and 13). The mean elevation in these two snow regions is considerably higher than the others, with the exception of the highest elevations (snow region 14 or areas >1220 m). Therefore, the increase in snowfall totals appears to be somewhat elevation dependent, suggesting the important role orographic processes play in extracting lower-tropospheric moisture. The greater regional topographic relief along the windward slopes of the high country and northern plateau translates into a longer period of orographic lifting. In addition, in a conditionally unstable lower troposphere, the orographic ascent along the windward higher-elevation slopes may provide sufficient forcing for convection to develop, further extracting lower-tropospheric moisture. Even though an increase in the composite mean and maximum snowfall totals is evident for events with a GLC in the highest elevations (region 14), it is not as great as that seen in the high country and northern plateau (regions 8 and 13, respectively). This may be related to the more efficient scavenging of low-level moisture at the highest elevations in trajectory class 2 (e.g., no GLC) when compared with stations at slightly lower elevations. No significant topographic features are present upwind of the study area in association with northwesterly trajectories, so it is safe to assume that upstream moisture robbing snow associated with topographic forcing is not connected to the differences in snowfall totals.

Additional findings related to the influence of the Great Lakes indicate that no significant difference (p < 0.05) exists in the composite mean and maximum snowfall totals between events with percent ice cover in the top quartile (≥25%) and those in the bottom three quartiles (<25%). This finding may be explained by the very small number of events with a GLC in which the percent ice cover was greater than 75% (n = 3 or 3.3%) or even 50% (n = 10 or 11.1%). In these cases, much of the ice cover existed on Lake Superior and the northern portions of Lake Michigan, leading to only a small effect on the sensible and latent heat fluxes for trajectories crossing the southern end of Lake Michigan. Although extensive ice cover greater than 75% greatly limits the sensible and latent heat fluxes, the small sample size (n = 3) limits our ability to draw definitive conclusions.

f. Comparative air trajectories

The spatial patterns of snowfall across the southern Appalachians in association with different trajectories are best illustrated by comparing air trajectories for two samples of events: (a) ones in which the highest totals occurred in the high country (region 8) and (b) ones in which the highest totals occurred in the northern plateau (region 13). These regions were chosen for comparative purposes because of the significant increase (p < 0.05) in the composite mean and maximum snowfall totals for events with a northwesterly trajectory and a GLC. When considerably more snow fell in the high country (top quartile of the snow region, 8:13 ratio), air trajectories for this region appeared to coincide approximately with the long axis of Lake Michigan (Fig. 13), whereas farther north, air trajectories were less favorable for lake interaction (e.g., across lower Michigan). Snowfall was observed along the windward slopes and higher elevations throughout the study area, with the greatest amounts centered over the high country. For events in which the highest totals occurred in the northern plateau (bottom quartile of the snow region, 8:13 ratio), air trajectories were most favorable for lake interaction in that respective snow region (Fig. 13). To the south, only trace amounts of snow fell in the higher elevations—with no snowfall occurring elsewhere—in association with air trajectories well west of the western Great Lakes.

Measures of lower-tropospheric relative humidity represent some of the most significant differences in synoptic fields between the bottom quartile and top quartile of the ratio of mean snowfall between the two snow regions (Table 5). These same fields also showed the greatest differences between trajectory class 2 (no GLC) and trajectory classes 3.1, 3.2, and 3.3 (GLC), and are consistent with a moistening of the lower troposphere as a result of the latent heat fluxes from the lake surfaces. Air trajectories, however, certainly are not the only factor influencing the spatial patterns of snowfall in these cases, as indicated by the significant differences in 1000- and 500-hPa heights, 500-hPa vorticity, and 850-hPa thermal advection. In particular, NWFS events in the bottom quartile (e.g., no snow in the high country or region 8) of events with a GLC are tied to higher 500- and 1000-hPa heights, and lower values of 500-hPa vorticity. In these events, not only is the lower troposphere much drier, but the 500-hPa trough is displaced farther to the north. In addition, cold advection at 850 hPa is significantly greater, suggesting stronger subsidence in the lower troposphere. Therefore, the characteristics of the air mass being advected into the high country (region 8) are not conducive for snowfall development, and may be partly explained by an air trajectory that is less favorable for interaction with the southern portion of Lake Michigan.