A climatology of single-banded snowfall in the central United States and the variability of processes at work in its formation are presented. Ninety-eight snowbands are identified in association with 66 cyclones over 5 yr spanning the winters from 2006/07 through 2010/11. An additional 38 cyclones featured nonbanded snowfall exceeding 4 in. (10.2 cm). Nearly twice as many bands were observed to the northeast of the surface low than to the northwest. Over each snowband’s life cycle, the median (mean) snowband lasted 4.0 (5.2) h, was 42 (45) km wide, 388 (428) km long, and had an aspect ratio of 10.2 (10.8). A common appearance exists for snowbands in different large-scale flow regimes and locations relative to the surface cyclone. The median snowband elongates during the first half of its life span, with its width remaining constant. During the second half of the median snowband’s life span, the length and width contract. Composite analysis of the synoptic and broad mesoscale environments that snowbands form in illustrates that the juxtaposition of the ingredients necessary for snowbands are similar no matter which quadrant of the surface low the band is located in, indicating that the synoptic-scale flow determines where these ingredients are organized with respect to the cyclone. The frequency of banded snowfall within each northern quadrant of the surface low, the typical snowband characteristics and their evolution, and the patterns that give rise to snowbands documented by this work can all prove useful to forecasters tasked with maintaining situational awareness in the presence of many solutions provided by ensemble numerical weather prediction.
Each year, an average of 105 snowstorms in the contiguous United States generates substantial economic costs and benefits (Adams et al. 2004). The snowfall in many of these storms takes the form of mesoscale snowbands. These snowbands are associated with heavy snowfall greater than 4 in. (10.2 cm) and snowfall rates in excess of 1 in. (2.5 cm) h−1. Forecast errors in the location of snowbands as small as 40 km (25 mi) can result in large errors in snowfall at individual locations, as shown in the hypothetical example in Fig. 1. The potential for error makes the issuance of accurate warnings for associated societal and economic impacts challenging. Strategies for forecasting banded snowfall (Novak et al. 2006) and examinations of numerical weather prediction of banded snowfall (Evans and Jurewicz 2009) highlight limits in the predictability of banded snowfall as forecast lead time increases. The climatology of banded snowfall and the variability of processes at work in its formation have received considerable study in the eastern United States (e.g., Novak et al. 2004, 2008, 2009, 2010; Novak and Colle 2012; Griffin et al. 2014; Ganetis and Colle 2015). In the central United States, the Profiling of Winter Storms (PLOWS) campaign observed the in-cloud processes responsible for banded snowfall associated with 14 cyclones (e.g., Keeler et al. 2016a,b; Plummer et al. 2014, 2015). Otherwise, with the exception of Byrd (1989), the majority of studies of banded snowfall completed over the central United States have been case studies (e.g., Martin 1998a,b; Moore et al. 2005; Grim et al. 2007; Han et al. 2007; Baxter et al. 2011; Rosenow et al. 2014, Rauber et al. 2014a,b). Thus, an examination of the climatology of banded snowfall and the variability of large-scale processes at work in its formation in the central United States is prudent.
The organization of moisture, lift, and instability in generating banded precipitation has been previously described by Nicosia and Grumm (1999), Jurewicz and Evans (2004), Novak et al. (2004), and Moore et al. (2005), among others. These studies describe a common set of physical processes involved in the production of banded snowfall, which often result from the interactions between large-scale conveyor belts. The concept of large-scale conveyor belts represents one paradigm for considering the organization of developing wintertime midlatitude cyclones. The cyclonically curving branch of the warm conveyor belt, known as the trough of warm air aloft (Martin 1998a,b), transports warm and moist air northward. Midlevel frontogenesis is produced where the rising air in the warm conveyor belt meets colder and drier air. This midlevel frontogenesis is associated with a mesoscale, transverse, ageostrophic circulation (Sawyer 1956). The stability of the atmosphere is reduced in the region of ascent and is often [though not always; e.g., Novak et al. (2009)] associated with air in the dry conveyor belt overrunning warmer, moister air in the warm conveyor belt, as shown by Grim et al. (2007) and Rauber et al. (2014a,b). When lift associated with frontogenesis is present in a saturated environment with saturated equivalent potential vorticity (SEPV) near zero, banded snowfall can result. The evolution of frontogenesis and SEPV in a saturated environment that produced banded snowfall in the central United States is well documented by Berndt and Graves (2009).
Results from the PLOWS campaign provide additional insights into precipitation banding by examining the interactions between large-scale flows and cloud-scale processes. The comma-head region of the cyclone contains small regions of enhanced radar reflectivity at the top of the clouds, known as convective generating cells. The dry conveyor belt contributes to destabilization that promotes the formation of these cells, which first form at the southern edge of the comma cloud and may continue forming farther north where the clouds are deeper (Rosenow et al. 2014). The instability is maintained via cloud-top radiative cooling (Keeler et al. 2016a,b). The convective generating cells produce a large number of ice crystals, which fall into underlying nimbostratus. Mesoscale deformation then acts on the ice crystals, merging them into bands (Plummer et al. 2014, 2015).
Previous studies have examined the distribution of banded precipitation with respect to surface cyclones. Novak et al. (2004) described the variability of cold-season banding environments associated with northeast U.S. cyclones. Composites of 75 northeastern U.S. cyclones featuring banded precipitation and 13 cyclones that did not feature banding were created. Of all single precipitation bands, 81% were located in the northwestern quadrant of the surface low. Similar studies conducted over different areas demonstrate that the predominant band location is ahead of the warm front (northeast of the surface cyclone) in the Pacific Northwest (Houze et al. 1976) and in the United Kingdom (Browning 1985). The presence of precipitation banding northeast of the cyclone has been anecdotally noted in cases of weak cyclogenesis in the Great Plains and Great Lakes regions by Banacos (2003).
The large amount of variability in the location of precipitation banding demonstrated by the aforementioned studies invites further investigation into the distribution of single-banded snowfall in central U.S. cyclones. This paper presents a 5-yr climatology of single bands of snowfall in the northeast and northwest quadrants of central U.S. cyclones. In addition, composites of the large-scale environments associated with the snowbands are presented.
The present work is organized as follows. Section 3 describes the data used and explains the methods used to create the climatology and composites. Section 4 presents the results of the climatology of central U.S. banded snowfall. Section 5 introduces the composites of relevant atmospheric fields in the environment of banded snowfall, along with quantification of the variability within each composite. Finally, section 6 presents the conclusions of the work and discusses potential future work.
3. Datasets and method
a. Data sources
Snowfall analyses were created using the National Centers for Environmental Information (NCEI) Cooperative Summary of the Day (COOP) dataset. This dataset includes daily observations of new snowfall taken by trained observers. Snowfall was analyzed over the winter months (defined here as October–April) for 5 yr, consisting of the winters from 2006/07 to 2010/11. A discussion of the sources of error in COOP data can be found in Baxter et al. (2005).
Composites of National Weather Service (NWS) level III base reflectivity were used to identify snowbands. These radar composites (made from 0.5°-angle scans) have a 1-km spatial resolution and are available every 5 min, with reflectivity binned in 5-dBZ intervals. The radar composites were created by the Iowa Environmental Mesonet Program at Iowa State University. More information on these composites is available online (http://mesonet.agron.iastate.edu/docs/nexrad_composites/). No radar composites were missing during the period studied.
To create composites of atmospheric fields, the North American Regional Reanalysis (NARR) was used as a proxy for observed atmospheric fields. The NARR, with data available every 3 h, has a grid spacing of 32 km and 45 vertical layers to 100 hPa (Mesinger et al. 2006). NARR data were chosen for use because of their high temporal and spatial resolution relative to observations and also because of the consistency of the modeling system used throughout the study period. NARR data were acquired from NCEI.
Surface observations from the Automated Surface Observing System (ASOS) were examined every 3 h to discriminate areas where the precipitation type was snow. ASOS data were acquired from the Iowa Environmental Mesonet Program at Iowa State University. The National Centers for Environmental Prediction/Weather Prediction Center human-produced surface analyses were used to determine the location of mean sea level pressure minima and maxima, as well as frontal boundaries. The data were available every 3 h, with locations of features defined to the nearest 1°.
As a first step to identifying snowbands, NCEI COOP data were searched to find days with more than 4 in. (10.2 cm) of snowfall during the 1200 UTC–1200 UTC 24-h period. This threshold is similar to the 0.5 in. (1.3 cm) of liquid equivalent used by Novak et al. (2004) and was chosen in part because of the snow advisory criteria of 3–5 in. (7.6–12.7 cm) in 12 h used by many of the NWS forecasting offices in the central United States (National Weather Service 2010). Snowbands were identified if they featured a radar reflectivity of 25 dBZ or greater, a length of at least 250 km, and an aspect ratio (length to width) of 3:1. These qualities were assessed manually and were required to persist at least 2 h for the band to be recorded. Width (length) was defined as the distance along the short (long) axis of an area of 25-dBZ reflectivity. When any one of these criteria was not met, the lifetime of the snowband was considered to end. These criteria were chosen based upon their effectiveness in defining single bands in Novak et al. (2004). A reflectivity threshold of 30 dBZ was used by Novak et al. (2004). Our slightly lower threshold helps to account for the greater spacing in between radar sites in the central United States versus the northeast United States. In addition, many bands in the central United States are farther from a source of moisture when compared with northeast U.S. bands, resulting in lower radar reflectivity. The snowbands must have occurred in an area where snowfall was observed both in the COOP data and the ASOS data. No threshold number of observations was employed in order to characterize bands as consisting entirely of snow; thus, it is possible that portions of some bands contain nonsnow precipitation. Selected bands were restricted to the central United States, defined as covering the area from approximately east of the Rocky Mountains to west of the Appalachian Mountains. A band had to spend the majority of its duration within this domain to be included in the study, although the surface low associated with the band may have been outside of the domain.
The width, length, and aspect ratio of the bands were recorded every 30 min. The geographic location of the band was recorded every 3 h, with the band being defined by a left, right, and center point. The location of the surface low at the same times was recorded. In cases where multiple surface lows were analyzed, the low closest to the band was used, provided that the low maintained continuity throughout the band’s duration. Events in which the band did not appear to be clearly associated with a surface cyclone were not included. The presence of 850-hPa warm-air advection downstream from the trough associated with the surface cyclone in the NARR data was used to aid in making this determination. Based on these measurements, the location of the snowband with respect to the cyclone was recorded as either northeast or northwest of the surface cyclone. Any banding to the southeast (in the warm sector) or southwest (along or behind a cold front) of the cyclone was not included in this study, as bands in these quadrants were almost exclusively rainbands or were transient. The location of the snowband with respect to the cyclone was determined according to the quadrant that the band spent the greatest amount of time in during its life cycle. If the band spent an equal amount of time in both the northwest and northeast quadrants, the band was recorded as being in “both” quadrants. Case examples of bands in each quadrant of the surface low can be seen in Fig. 2. Events that featured at least 4 in. of snow but no banding were recorded as “nonbanded” events, and the location of the nonbanded snowfall relative to the surface low was recorded. The number of nonbanded events recorded is sensitive to the dimensional criteria employed to identify bands, though these criteria have been employed in previous studies (e.g., Novak et al. 2004).
This paper focuses on single snowbands that form in the northeast or northwest quadrant of the surface low and further divides these snowbands by midlevel flow regime. Generally, snow can form in 500-hPa flow regimes that feature a trough advancing from either the northwest or the southwest. These flow regimes are associated with the climatological cyclone tracks that impact the central United States (Zishka and Smith 1980). Thus, the movement of the 500-hPa trough in the NARR during the cyclone’s life cycle was recorded. In contrast to Novak (2002) and Novak et al. (2004), the climatology of banded snowfall presented here is not a complete examination of the types of banded structures that can form in the central United States.
Five composites of atmospheric fields were created: northwest and northeast bands in southwesterly flow, northeast bands in northwesterly flow, and nonbanded events in both flow regimes (Table 1). The initial analysis time (t = 0) was defined as the 3-hourly time closest to band onset. For nonbanded composites, the initial analysis time is the approximate midpoint of the event. To average fields in a system-relative sense, a subset grid of the NARR (comprising 189 × 213 points with 32-km spacing) was centered on the location of each surface cyclone at each time. The use of the surface cyclone’s location to calculate system-relative composites was also employed by Novak et al. (2004). If the subset grid used for compositing extended beyond the native NARR grid boundaries as a result of recentering the grid to the location of the surface low, the grid values at the edge of the domain were duplicated to all points beyond the edge. These duplicated regions were well removed from the cyclone center, and relatively small in size, and thus did not impact the analysis. Composite fields are plotted with their center points at the average locations of all surface cyclones that compose the composite.
Frontogenesis was composited to evaluate the presence and magnitude of mesoscale forcing for ascent, using Petterssen’s (1956) formulation. SEPV was composited to evaluate the presence and magnitude of instability, using
(Martin et al. 1992; Moore and Lambert 1993), where is the three-dimensional vorticity vector, is the three-dimensional gradient operator, and is the saturated equivalent potential temperature with respect to water. The influence of saturation with respect to ice on SEPV is generally not included (Schultz and Schumacher 1999), as ice has been shown to have minimal influence on SEPV in comparison with that of liquid (Rivas Soriano and García Díez 1997; Reuter and Beaubien 1996). Regions where SEPV < 0 and air is saturated indicate areas of instability in either the slantwise, or vertical, directions. Additionally, air in this region may be inertially unstable, regardless if saturation is present. The full wind is used in the calculation of SEPV, following Jurewicz and Evans (2004). Frontogenesis and SEPV were calculated for each individual event and then averaged to create composites. The level of maximum frontogenesis and SEPV reduction in the vertical plane varies for each event, which will result in reduced magnitudes of these fields in the composites versus the individual events. The plan-view composites in the present study display frontogenesis averaged over the 650–750-hPa layer, and SEPV averaged over the 550–650-hPa layer. The use of layer-averaged quantities rather than individual levels likely contributes to more spatially consistent signatures, though the layer-averaged frontogenesis maxima may appear broader and weaker as a result of the slope of the frontal surface with height.
Many cyclones were associated with multiple bands, and some of these bands overlapped in time. In this study, each band was recorded individually and included in the composites. Therefore, in instances of multiple bands within a single cyclone, the same cyclone may be included in composites of both northwest and northeast bands, or different times of the cyclone may be included in the same composite. When multiple bands were present in more than one quadrant of the cyclone, Novak et al. (2004) also chose to include these bands in more than one composite class. There were no instances in which multiple bands had the same initial analysis time (t = 0), so no exact duplication of fields was present in the composites.
Over the 5-yr period, 104 events with greater than 4 in. (10.2 cm) of snowfall were identified. Of these, 38 events (37%) did not contain bands, while 66 events (63%) featured a total of 98 bands (Table 1). The number of nonbanded events is more than double that found by Novak et al. (2004), who determined that 13 (15%) of their 88 events exhibited nonbanded precipitation. The majority of bands (77) were found in southwest flow at 500 hPa. Nearly twice as many bands were observed to the northeast of the surface low (54, 55% of all bands) compared with those to the northwest of the surface low (29, 30% of all bands). This is in contrast to cold-season banded precipitation associated with northeastern U.S. cyclones, where 81% of bands exhibited most of their length in the northwest quadrant (Novak et al. 2004). Only two bands in the northwest quadrant were observed in northwest flow, suggesting that these are comparatively rare events, possibly related to the fact that the large-scale flow exercises controls on the characteristics of low-level fronts in midlatitude cyclones (Schultz et al. 1998). For nonbanded events, 19 had snowfall to the northeast of the surface low, 9 to the northwest, and 10 in both quadrants.
The majority of bands were observed in the north-central United States and eastern Great Lakes, with few bands in the southern United States (Fig. 3a). Surface lows associated with northwest-quadrant bands were more common in the southeastern portion of the domain versus northeast-quadrant bands (Fig. 3b). The snowbands in Fig. 3a are shown relative to their parent cyclones in Fig. 3b via the polar plots in Fig. 4. Most bands in the northwest quadrant are oriented from southwest to northeast, likely because of the dominance of southwesterly flow at 500 hPa associated with these bands, though this is not the only control on band orientation. In contrast, bands in the northeast quadrant exhibit a variety of orientations, as both northwest and southwest flow regimes occurred with these bands. In southwest flow at 500 hPa, northeast bands were typically oriented from southwest to northeast, as short-wave troughs trend toward a negative tilt. In northwest flow at 500 hPa, northeast bands were typically oriented from northwest to southeast, as short-wave troughs follow the well-known Alberta clipper storm track (Thomas and Martin 2007). Most northwest-quadrant bands (66%) and bands spanning both quadrants (87%) were within 600 km of the cyclone center, with distances measured from the midpoint of the band. The median distance (standard deviation) between the band and surface low was 409 km (224 km) for northwest bands and 339 km (214 km) for bands in both quadrants. In contrast, 50% of northeast bands were more than 600 km away, with a few bands as far away as over 1000 km. The median distance (standard deviation) was 608 km (242 km).
As observations of bandwidth and length were collected every 30 min from the radar data, this information was used to compile distributions of snowband characteristics. The width, length, and associated aspect ratio were averaged over each band’s life cycle, and the duration of each band was recorded. The distribution of these characteristics from each band is plotted in Fig. 5, with data stratified by band location. Focusing on all snowbands and using averages taken over each snowband’s life cycle, the median (mean) snowband lasted 4.0 (5.2) h, was 42 (45) km wide, 388 (428) km long, and had an aspect ratio of 10.2 (10.8). Considerable positive skewness exists in all four variables, in particular the length and time, as some bands were much longer in length and duration. Additionally, lower bounds were prescribed on the duration, length, and aspect ratio variables when bands were identified, contributing to the observed skewed distributions. Few previous studies have evaluated band characteristics. In a study of lightning within banded snowfall, Market and Becker (2009) found a median length of 168 km and a median width of 35 km over 16 bands. Their width is similar to that found here (42 km), while the length is much smaller owing to lower threshold values in the characteristics used to define a band.
In the present study, the medians of length, width, aspect ratio, and duration for different categories of band (e.g., northeast-quadrant band in southwest flow; see Table 1) did not vary by more than 13% from those for all bands, as evidenced through comparing the medians for northeast-quadrant bands, northwest-quadrant bands, and all bands in Fig. 5. This suggests that, when band characteristics are averaged over each band’s duration, a common appearance exists for bands, despite where the bands might appear relative to the surface cyclone, or what flow regime they might exist in. Further, this suggests that the mesoscale ingredients that result in banded snowfall [as described in previous studies; e.g., Novak et al. (2004) and Rauber et al. (2014b)] are similar across band categories, but the synoptic-scale flow determines where these ingredients are organized to produce bands with respect to the cyclone. The standard deviations of band characteristics for different band categories exhibited greater percent differences relative to the standard deviations for all bands, but only one value was over 20% (the standard deviation of the aspect ratio exhibited a 43% difference for bands that spanned both the northeast and northwest quadrants of the surface low; not shown).
The distributions that describe band characteristics were calculated using values averaged over each band’s life span. Snowbands are known to grow and decay as the moist processes responsible for them evolve (Novak et al. 2009). Can statistical changes in the length and width of snowbands in the central United States be used to describe a characteristic snowband evolution? To examine this, the width, length, and aspect ratio were recorded at the beginning, end, and midpoint of the periods that each band met the criteria described in section 3b. The duration of these periods will vary according to the duration of the band, the distribution of which is shown in Fig. 5d. Next, the percent change in these characteristics was measured over the first half of each band (from the beginning to the midpoint of the band’s lifetime) and the second half of each band (from midpoint to end). The median percentage change for all bands in the first half of their life is 0% for width, 14.8% for length, and 16.5% for aspect ratio (Fig. 6). This indicates that the median band is lengthening, with its width held constant, thus increasing its aspect ratio. The median percentage changes for all bands in the second half of their life are −7.9% for width, −9.7% for length, and −2.9% for aspect ratio. Thus, the median band is contracting in both width and length, resulting in a very small change in aspect ratio. The interquartile range of the percent change in band length over the first half of the bands is greater than 0% and does not overlap with that of the second half, which is less than 0%. Therefore, an increase in length through the midpoint of a band’s life cycle is followed by a decrease in length for the middle 50% of the distribution of the bands.
Figure 7 illustrates how the life cycle of a hypothetical band would evolve if viewed on radar, using the median length and width at the beginning, midpoint, and end of the life cycle of the band. It is clear that the band behaves according to the percent changes previously described. As the distributions of the percent change in band characteristics show (Fig. 6), variability in band evolution does exist, but it appears that from the 98 bands studied, a signal describing the evolution of a median or “typical” band does exist. It must be noted that all of the statistics describing the bands and their evolution are relative to the choice of 25 dBZ to define the band. Bands associated with higher values of reflectivity will exhibit different dimensions and may evolve differently.
Novak (2002) demonstrates that bands in the northeast quadrant can transition to the northwest quadrant after the formation of the occluded front in northeastern U.S. cyclones, as depicted by Schultz and Vaughan (2011). Of the 66 events that contained banding, only nine events featured this type of transition. Most events (44) only had one band that remained in the same quadrant for its duration. Of the remaining 13 events, 2 had bands in both quadrants at roughly the same time, 7 had multiple bands in the same quadrant, and 4 had bands first in the northwest quadrant moving into the northeast quadrant. These results suggest that while some events do feature banding that transitions cyclonically with respect to the surface low, significant variability exists. Identifying a relationship between occlusion and a transition of bands across the northern quadrants would require a rigorous analysis of the evolution of frontal structures and is, thus, beyond the scope of this study.
The initial analysis time (t = 0) for composites was defined as the 3-hourly time closest to band onset. Only composites at the initial analysis time will be shown, as this paper represents the initial investigation into the environments associated with snowfall banding for a number of different locations of banding and flow regimes. Composites for the following categories in Table 1 will be shown: northwest-quadrant band in southwest flow (27 events), northeast-quadrant band in southwest flow (37 events), nonbanded snow in southwest flow (25 events), northeast-quadrant band in northwest flow (17 events), and nonbanded snow in northwest flow (13 events). As the composites are centered on the surface low, the composites are positioned on the map relative to the average latitude–longitude of the surface low in the events that make up the composite, The average latitude–longitude pairs that define the two ends and midpoint of the band are used to define the composite band location in the figures that follow. A cross section is taken through each composite, oriented perpendicular to the composite band location. In nonbanded composites, this cross section is taken where one might reasonably expect snowfall to be occurring, as diagnosed by the juxtaposition of frontogenesis and reduced SEPV in the plan-view diagrams.
a. Events in southwesterly 500-hPa flow
The composite for northwest-quadrant bands in southwest flow (27 events) is similar to what has been previously shown in numerous case studies (e.g., Moore et al. 2005; Berndt and Graves 2009). A 1004-hPa low is located in the inflection point of a 500-hPa wave, downstream from a neutrally tilted trough and in the left-exit region of a 300-hPa jet streak (Fig. 8a). Another jet streak is downstream from the trough, farther north in zonal flow. The surface low is in an area where both the cross-stream (jet streak) and along-stream (upper wave) ageostrophic circulations play a role in enhancing the vertical ascent (Keyser et al. 1989; Loughe et al. 1995). A maximum of 650–750-hPa layer-averaged (midlevel) frontogenesis is observed in the northwest quadrant of the surface low, in the vicinity of the location of the average band (Fig. 8b). An area of 550–650-hPa (midlevel) SEPV less than zero is seen to the southeast of the band. A cross section reveals a maximum in upward vertical motion of −7 μb s−1 directly above the composite band location (Fig. 8c). This maximum is located where relative humidity with respect to ice is near 90%, and the SEPV is near zero. Since the θe lapse rate is near zero, SEPV near zero indicates that the atmosphere is near neutral stability with respect to vertical and slantwise motion (not shown). Forcing for upward vertical motion is provided by low-to-midlevel frontogenesis, which slopes upward in the northwest direction. The frontogenesis exceeding 7 K (100 km)−1 (3 h)−1 that extends from the surface to near 700 hPa is likely the result of very strong frontogenesis present at a variety of different locations in each event. In summary, the composited fields shown clearly indicate that moisture, lift, and instability organized in a way that is well recognized to be associated with banded snowfall.
The composite for northeast-quadrant bands in southwest flow (37 events) shows the same organization of moisture, lift, and instability known to produce banded snowfall, but in a different location relative to the surface low. The strength of the surface low is the same as that in the northwest band composite: 1004 hPa (Fig. 9a). The 500-hPa trough is slightly less amplified compared with the northwest band composite, and the ridge downstream from the trough is considerably less amplified. These less amplified heights are associated with comparatively less prominent jet streaks at 300 hPa. The 700-hPa heights depict confluent flow in the northeast quadrant of the surface low, leading to the midlevel frontogenesis seen along the cold side of the composite snowband (Fig. 9b), similar to the anecdotal conceptual model presented by Banacos (2003). Schultz et al. (1998) analyzed a cyclone moving into confluent, low-amplitude flow, and found it to be associated with a strong zonally oriented warm front, a result generalized by idealized simulations. Some midlevel frontogenesis is seen in the northwest quadrant of the surface low, but it lacks the linear spatial pattern known to be associated with banded snowfall. Midlevel SEPV less than zero is present along the air flowing toward the composite snowband. The composite cross section (Fig. 9c) shows many of the same features seen in the composite cross section for northwest bands (Fig. 8c). Vertical motion and relative humidity with respect to ice are less than those in the northwest composite, and the frontogenesis is less spatially continuous.
In the nonbanded snow in the southwesterly flow composite (25 events), a weaker 1008-hPa surface low is seen relative to the previous two composites, despite the presence of a slightly more amplified wave at 500 hPa (Fig. 10a). Jet streaks at 300 hPa are more distant from the surface low than either of the previous composites, suggesting a reduced potential for jet coupling. Midlevel frontogenesis is lower in magnitude and more spatially variable than that of the previous composites, and the air mass with midlevel SEPV less than zero remains well south of the surface low (Fig. 10b). In the nonbanded composite cross section, vertical motion is present above shallow frontogenesis below 800 hPa, in the presence of relative humidity with respect to ice exceeding 85% (Fig. 10c). SEPV values in the area of maximum vertical motion are near 0.25 PVU (1 PVU = 10−6 K kg−1 m2 s−1). Taken together, the values of moisture and lift in the composite cross section suggest the potential for snowfall greater than 4 in. (10.2 cm, as occurred with each event in the composite), but the higher stability and weaker frontogenesis preclude development of a band. The ascent of −4 μb s−1 in the nonbanded composite is similar to that in the northeast band composite (Fig. 9c). Lift associated with snowbands is known to be smaller in scale and more intense than that of nonbanded snow, and it is therefore likely that compositing resulted in significant smearing associated with variability in the location of banding relative to the surface low (see Fig. 4), which diminished the magnitude of the lift to a greater extent in the banded composite versus the nonbanded case. Additionally, the NARR’s 32-km grid spacing likely does not permit adequate resolution of the true frontogenetical circulation (Persson and Warner 1993). Case examples of northeast bands in southwest flow will be presented in section 5c. However, the magnitudes of these processes and their juxtaposition are distinctly different from those seen in the previous two banded composites. For example, the layer of reduced SEPV near 600 hPa is displaced from the maximum midlevel frontogenesis near 750 hPa, indicating that the frontogenetical circulation is vertically displaced from the least stable air. As reduced stability results in a contraction in the spatial scale of frontogenetically forced ascent (Emanuel 1985), the lack of reduced stability in this region promotes a broader updraft.
b. Events in northwesterly 500-hPa flow
The composite for northeast-quadrant bands in northwest flow (17 events) again shows the same organization of moisture, lift, and instability known to produce banded snowfall, although the 500-hPa trough moves from northwest to southeast in the events that compose the composite. The composited 500-hPa trough and associated surface low (Fig. 11a) are along the southern edge of the distribution of 177 Alberta storm tracks identified by Thomas and Martin (2007). A weak amplitude wave is present at 500 hPa, within a tight gradient of geopotential heights that is associated with a 45–50 m s−1 jet streak at 300 hPa. At 700 hPa, the wave has slightly greater amplitude, with midlevel frontogenesis on the cold side of the band that is considerably stronger than that seen in previous banded composites (Fig. 11b). The less stable air mass depicted by SEPV below zero is to the southwest of the band. The composite cross section shows vertical motion of −4 μb s−1 along the warm side of strong vertically oriented frontogenesis, in the presence of relative humidity with respect to ice exceeding 90% and SEPV between 0.1 and 0.25 PVU (Fig. 11c). The tighter geopotential height gradient at 500 and 700 hPa is associated with a stronger thermal gradient, and thus stronger midlevel frontogenesis in comparison with the other banded composites, in part because the value of Petterssen’s (1956) frontogenesis is directly proportional to the strength of the thermal gradient. Compared with the composite cross section for northeast-quadrant bands in southwest flow (Fig. 9c), here the stability is slightly less reduced above the band, with SEPV ranging from 0.1 to 0.25 PVU versus 0 to 0.1 PVU. As previously discussed, this composite features relatively stronger frontogenesis in comparison with the other banded composites, paired with slightly more stable air. This suggests that the stronger frontogenetical circulation is able to enhance the lift in the presence of slightly more stable air and, thus, produce a snowband. The composite for northeast-quadrant bands in northwest flow again demonstrates that moisture, lift, and instability can be focused to generate mesoscale banded snowfall in a variety of locations relative to the surface cyclone, and in a variety of flow regimes at 500 hPa.
In the nonbanded snow in northwesterly flow composite (13 events), the 500-hPa geopotential height gradient (Fig. 12a) is less than that seen in the previous banded composite. The 300-hPa isotachs are also much weaker, and the mean sea level pressure is only 1016 hPa. There is no maximum in midlevel frontogenesis in close proximity to the surface low, and SEPV values less than zero are well south of the surface low (Fig. 12b). A composite cross section taken to the northeast of the surface low reveals a broad area of weak upward motion in air with relative humidity with respect to ice from 75% to 80% (Fig. 12c). As in the nonbanded composite cross section in southwest flow (Fig. 10), the frontogenesis maximum was concentrated below 800 hPa, and the less stable air with SEPV values less than 0.1 PVU was present much higher than the maximum in frontogenesis, at 500–600 hPa. While the composite does suggest the presence of moisture and lift that lead to snowfall exceeding 4 in. (10.2 cm) in the individual events, it is clear that the strength and juxtaposition of these ingredients did not favor banded snowfall.
c. Representativeness of composite fields
As discussed in section 3b, a composite is simply an average of atmospheric states, in this case centered about the location of the surface low. Some a priori constraints were placed on the atmospheric states that went into each composite, in particular, the direction of movement of the 500-hPa trough and the location of the snowband with respect to the surface low. These constraints promote the likelihood that the composites are representative of the individual events that compose them, but further quantitative analysis can assess how representative the composites are.
The linear spatial correlation coefficient between the individual events and the composite fields was calculated. For more information on the linear spatial correlation coefficient, the reader is invited to see Moore et al. (2003). Larger (smaller) correlations indicate greater (lesser) agreement between the pattern in the composite field and each individual event. Fields that feature smaller correlations warrant further investigation on an event-by-event basis. As each composite is comprised of N events (documented at the beginning of this section), correlations for each field are best viewed as a distribution of correlations across events, as shown in Fig. 13, for each of the categories for which composites were created.
The analysis of the composite fields in section 5 demonstrates that the composited fields for each category exhibit the same organization of moisture, lift, and instability that results in banded snow [as described in this study and others; e.g., Novak et al. (2004)], but these ingredients are focused at different locations relative to the cyclone and are contained within different flow regimes. All of the fields in the composite plan views have similar correlations, no matter which of the composites is examined, which indicates that each composite is roughly equivalently representative of its individual events. This suggests that the variabilities in the patterns of the individual events for each category are similar. With the exception of 650–750-hPa frontogenesis, the median correlations are above 0.5 for each field in each composite (discussed below). These correlations, when paired with the fact that the composites depict physical processes known to be associated with banded snowfall (or lack thereof), suggest that the composites are reasonably representative of the individual events that compose them.
The median correlations for 500- and 700-hPa geopotential heights were all over 0.9, as might be expected because of the presence of climatological variation in these fields and the fact that the direction of trough movement was prescribed. Median correlations of midlevel SEPV, 300-hPa isotachs, 650-hPa relative humidity with respect to ice, and mean sea level pressure all ranged from 0.5 to 0.7. The patterns in these fields thus exhibit more case-to-case variability than the geopotential height fields. For the moisture and instability fields, this is likely due to the fact that in cases of cyclogenesis, maxima in moisture and instability are concentrated along the warm conveyor belt (Madonna et al. 2014; Browning 1997), a smaller-scale feature than a wave in the 500- or 700-hPa heights. The relatively lower correlations in the 300-hPa isotachs may be associated with variability in the presence, location, and strength of jet maxima distant from the banded snowfall and associated cyclone. In many composite categories, the mean sea level pressure exhibits larger variability in the correlations of individual events with the composite. This is likely because snowfall (banded or nonbanded) can occur in association with surface cyclones of varying strength, as long as moisture, lift, and instability are able to organize to create precipitation. For example, Moore et al. (2005) analyze a case of heavy banded snowfall in association with a surface cyclone of only 1009 hPa.
Median correlations for midlevel frontogenesis were the lowest of all fields, ranging from 0.2 to 0.3 for all categories. In their analysis of correlations between individual elevated thunderstorm events and an associated composite, Moore et al. (2003) noted that more variability was present in parameters that involve horizontal derivatives of basic variables. Frontogenesis represents mesoscale forcing for ascent that promotes banded snowfall, and given the large variability in band location with respect to the cyclone center seen in Fig. 4, lower correlations are not surprising. Also, mesoscale fields have more spatial variability than larger-scale fields, which are observed to have higher correlations. In individual events, the band may be associated with frontogenesis at higher or lower elevations than the 650–750-hPa average chosen for plan-view display, and the spatial extent of the maximum frontogenesis will vary in association with the spatial dimensions of the band (the variability of which was described in section 3b). For these reasons, the frontogenesis in the individual events is highly variable. The variability of frontogenesis in the vertical is illustrated via a comparison of two bands found to the northeast of the surface low in southwest flow (Fig. 14). Both bands are embedded in synoptic patterns that are relatively similar, though one band is closer to the surface low (Fig. 14a) than the other (Fig. 14b). Since the warm conveyor belt rises as it flows northward over colder air, the maximum frontogenesis in the column might be expected at higher levels for bands farther away from the surface low. Contrary to this, the band closer to the surface low features a deep layer of sloping frontogenesis, associated with a maximum in ascent near 600 hPa. The band farther from the surface low features shallow frontogenesis associated with a maximum in ascent near 700 hPa. These cases illustrate that there exists no simple relationship between the distance of the band from the surface low and the location of maximum frontogenesis in the vertical, since the slopes of the frontal zones also vary for each event. Finally, examination of the cases that featured the lowest correlations of frontogenesis showed that the other fields had correlations similar to those previously described. This is evidence that these events were still similar to the composites when other fields are used, and thus these events are not wholly unlike their composites.
Differences between the composites of northwest bands in southwest flow (Fig. 8) and northeast bands in southwest flow (Fig. 9) were discussed in section 5a. While subjective differences in the large-scale fields were noted, correlations between these composites show a large degree of similarity in the 300-hPa isotachs (0.88), 700- and 500-hPa heights (0.98), and mean sea level pressure (MSLP; 0.82). Yet the individual events that compose each of these composites are more different from each other than the comparison of composites suggest. Correlations for these fields between the northwest band in southwest flow events and the northeast band in the southwest flow composite are less than those between the two composites: 31% less for 300-hPa isotachs, 5% less for 700- and 500-hPa heights, and 37% less for MSLP. Values are similar when the northeast band in southwest flow events is compared with the northwest band in the southwest flow composite. This result further suggests that there are differences in the large-scale patterns associated with banding in northwest versus northeast quadrants of the surface low.
Over the five years spanning the winters from 2006/07 through 2010/11, 98 snowbands were identified, in association with 66 surface cyclones. Nearly twice as many snowbands were found in the northeast quadrant of the surface low (55% of all bands) compared with the northwest quadrant of the surface low (29% of all bands). The presence of snowbands in the northeast quadrant of the surface low has been noted anecdotally by Banacos (2003). Additional studies recognize the presence of precipitation banding ahead of the warm front in different geographical areas (e.g., Houze et al. 1976; Browning 1985; Market et al. 2002). Novak et al. (2004) found that most cold-season banded precipitation in the northeastern United States occurred in the northwestern quadrant of the surface cyclone. When placed within context with these prior findings, the present study further demonstrates that band location with respect to surface features varies geographically. Additionally, out of a total of 104 surface cyclones, 37% exhibited greater than 4 in. (10.2 cm) of snowfall that was nonbanded.
The large number of snowbands identified and analyzed in this study presents a unique opportunity to statistically define the dimensions and duration of a typical snowband in the central United States. Based on averages taken over each snowband’s life cycle, the median (mean) snowband lasted 4.0 (5.2) h, was 42 (45) km wide, and 388 (428) km long, with an aspect ratio of 10.2 (10.8). The medians of these quantities for each category of band (see Table 1) did not vary by more than 13% from those of all bands, which indicates that no matter the flow regime or location of the band with respect to the surface cyclone, bands have similar dimensions and durations when statistics are computed over many bands. Distributions of band characteristics were positively skewed, evidence that individual events can be very different from the median band. Examination of the median snowband dimensions at their beginning, approximate midpoint, and end show that the median band is elongated during the first half of its life span, while its width remained constant. During the second half of the median band’s life span, both the length and width contracted. The evolution of the median snowband described in this study is consistent with observations from the PLOWS campaign that banding is the result of deformation acting on falling ice crystals (e.g., Keeler et al. 2016a,b; Plummer et al. 2014, 2015).
Composites of snowbands show the organization of moisture, lift, and instability known to produce banded snowfall, as documented in several studies (e.g., Nicosia and Grumm 1999; Novak et al. 2004). No matter which quadrant of the surface low the band is located in (northwest or northeast), the juxtapositions of the ingredients necessary for banded snowfall are similar. Banded snowfall can exist in either northern quadrant of the surface low, provided the environment is moist and midlevel frontogenesis coexists with reduced midlevel SEPV. The large-scale patterns that promote these conditions in either northern quadrant contained differences, which were specific to the flow regime at 500 hPa. As snowbands in the northeast quadrant have received less attention in the literature, schematics that summarize key composite features are shown for northeast bands in northwest flow (Fig. 15a) and northeast bands in southwest flow (Fig. 15b). The individual events that compose the composites are reasonably well correlated with the composites, suggesting that forecasters can look for the patterns in the composites within numerical model output to help diagnose regions favorable for banded snowfall.
The frequency of banded snowfall within each northern quadrant of the surface low, the typical snowband characteristics and their evolution, and the patterns that give rise to banded snow can all prove useful to forecasters tasked with maintaining situational awareness in the presence of a number of solutions provided by ensemble numerical weather prediction. Future work could analyze the evolution of the environments that give rise to banded snowfall, as well as how banded snowfall evolves during the occlusion process, thus providing forecasters with a more complete understanding of the changes in processes that lead to case-to-case variability in the presence of bands in either northern quadrant of the surface low. Additionally, understanding the environmental changes that give rise to the evolution of the median snowband’s length and width (Fig. 7) would help forecasters anticipate how snowband dimensions might change in response to model-predicted changes in atmospheric processes.
Funding for this work was provided by a subaward with the University Corporation for Atmospheric Research (UCAR), under Cooperative Agreement NA06NWS4670013 with the National Oceanic and Atmospheric Administration (NOAA), U.S. Department of Commerce (DoC). The statements, findings, conclusions, and recommendations are those of the authors and do not necessarily reflect the views of NOAA, DoC, or UCAR. We thank Chuck Graves of Saint Louis University for creating and sharing the code used to make the composites, Daryl Herzmann of Iowa State University for providing composite radar imagery, and David Novak of NOAA/NWS/NCEP/WPC for providing coded frontal bulletins. We thank the three anonymous reviewers, whose suggestions helped to improve the paper.