The 241 largest snowstorms over the eastern two-thirds of the United States during 1950–2000 exhibited considerable temporal variability ranging from 1 storm in three winters to 10 storms in 1993/94. The peak decadal frequency was 55 storms (1950s), and the minimum was 45 storms (1970s and 1980s). The east–north-central region experienced the greatest number of large snowstorms (148) followed by the west–north-central (136) and central (133) regions. Regional trends were different. Assessment of surface cyclone tracks associated with the large snowstorms identified three primary tracks: one was located from the leeward side of the south-central Rocky Mountains east-northeast toward the Great Lakes; a second was from the lower Mississippi River basin northeastward toward the Great Lakes; and a third was along the coastal mid-Atlantic region northeast toward Maine. Temporal differences in the frequency of certain surface cyclone tracks were related to the decadal trends in regional large snowstorm occurrence. The minimum surface pressure associated with these storms ranged from 959 to 1013 hPa with more than 67% of all storms having a minimum surface pressure between 980 and 999 hPa. The average orthogonal distance from the storm track to the heavy snow region was 201 km. The average rate of cyclone movement ranged from less than 483 to more than 1930 km day−1, with more than 57% of storms moving between 805 and 1287 km day−1.
It is well known that low pressure systems during the cold season that move across the central and/or eastern United States are key factors causing snowstorms. A newly developed dataset defined the dimensions of extra large damaging snowstorms during 1950–2000 in the eastern two-thirds of the nation, and these data provided an opportunity to assess the spatial and temporal behavior of the surface lows associated with these major storms. Such analysis allowed assessment of two key issues: 1) the primary storm tracks that cause such storms, and 2) how these low pressure systems have fluctuated over time.
Snowstorms create many social and economic impacts across a wide range of weather-sensitive sectors including insurance, transportation, energy, ranching, health, and government services (Changnon and Changnon 1978, 2006a; Branick 1997; Kocin and Uccellini 2005; Changnon et al. 2006). Changnon and Changnon (2006b) found that insured property losses related to U.S. snowstorm catastrophes during the period 1949–2001 totaled $21.6 billion (in 2001 dollars). Another study (Changnon and Changnon 2007) found that nearly 2300 snowstorms, those causing 15.2 cm of snowfall in a 1- or 2-day period at three or more adjacent weather stations, occurred during the 1950–2000 period in the region east of the U.S. Rocky Mountains. A strong relationship (r value of +0.86) between the size of the storm and the related amount of insured property loss indicated that storm size was the most important characteristic in determining economic impacts. Thus, an objective of this study was to assess the surface low pressure systems that caused the 241 largest and most costly snowstorms, those producing heavy snow (>15.2 cm of snowfall) over 258 000 km2 (100 000 mi2) or more area in the central and eastern United States.
Other studies have examined and described snowstorms and related weather conditions associated with major snowstorms and developed related climatologies (Goree and Younkin 1966; Changnon 1969, Beckman 1987, Suckling 1991; Branick 1997; Mote et al. 1997; Zielinski 2002; Kocin and Uccellini 2005). However, a limitation of these studies is that they either examined snowstorms impacting a state or region, or examined snowstorm characteristics for a short period (generally less than 20 yr), or both. Thus, this comprehensive 51-yr study of large snowstorms occurring east of the Rockies is unique in length of data and region assessed. Schwartz and Schmidlin (2002) conducted a similar long-term (1959–2000) national study; however, their research focused on blizzards, a particular type of winter storm with a different set of criteria. This study used a newly developed snowstorm database (Changnon et al. 2006) to describe “large” snowstorms and assess surface cyclone characteristics associated with these storms. Results from this research may be useful to a wide range of atmospheric scientists who are interested in temporal and spatial variability issues and atmospheric conditions associated with climate extremes.
2. Data and methods
An intensive study identified approximately 1150 first-order and cooperative National Weather Service (NWS) stations located east of the Rockies with good quality snowfall records during the 51-yr period 1950–2000 (Changnon et al. 2006). Although these stations were not equally spaced, on average there was one for each 1000 km2. For each site, a list of dates through the 50 yr when the 1- or 2-day snowfall amounts >15.2 cm was developed. This snowfall threshold was identified in prior research by Changnon (1969) as causing significant social and economic impacts. Branick (1997) suggested that snowstorms producing >15.2 cm of snowfall were “heavy” and represented a category of snowstorms. Snowstorms in the western United States were not analyzed owing to the influence of topography on snowfall distribution and the fact that the majority of NWS reporting stations were located at lower elevations where less snowfall occurs (Changnon et al. 1991).
The initial snowstorm criteria used in this study involved having three or more adjacent stations with >15.2 cm of snowfall in the same 1–2-day period. Using this criteria, an earlier study (Changnon and Changnon 2007) identified and described 2305 central and eastern U.S. snowstorms, determining the size, shape, length, and other characteristics. Storm size was found to be well related to snowstorm-related insured property losses. Since large storms produced more insured losses (Changnon and Changnon 2006b), this study focused on the 241 largest storms, those with heavy snow over 258 000 km2 (100 000 mi2) or more area, and the top 10% of all snowstorms (Changnon and Changnon 2007). Regions outside the U.S. boundary (e.g., in Canada) or over the Great Lakes and Atlantic Ocean, which might have experienced >15.2 cm of snowfall, were not counted toward the total storm area and this biased the results of this research. For example, if a snowstorm occurred over parts of the U.S. and Canada, however, the contiguous area with heavy snowfall (>15.2 cm) in the United States was less than 258 000 km2, it would not be counted as a large snowstorm in this study. An example of a large storm occurring 14–17 March 1960 is shown in Fig. 1.
Cyclone tracks related to each of these large snowstorms, such as the 14–17 March 1960 event, were plotted (Fig. 1). Because daily snowfall observations occurred at various times of the day, the cyclone lifetime was considered to be the period from the start of the large snowstorm (e.g., 14 March 1960) to the day after the end of the large snowstorm (e.g., 18 March 1960). The surface cyclone data were obtained from the National Oceanic and Atmospheric Administration (NOAA) Daily Weather Map Series (NOAA 2007). A grid of 5° latitude by 5° longitude numbered boxes (Fig. 2) was superimposed over each map with the surface cyclone track, and boxes in which the surface cyclone moved, even if in just a small amount of the box, were identified and counted (multiple entries of the same cyclone into a box were excluded). Although this a rather low resolution (i.e., 5° by 5° boxes) for counting all synoptic features, the strong cyclones associated with these large snowstorms (in central and eastern U.S. areas with little variation in elevation or topography) were easily identified in the daily weather maps. Furthermore, we had the advantage of looking directly on the maps, so that the uncertainties associated with tracking cyclones was smaller (Blender and Schubert 2000; Zolina and Gulev 2002; Jung et al. 2006). For example the 14–17 March 1960 cyclone tracked through boxes numbered 4, 3, 8, 13, 18, 17, 22, and 21. This “box” technique for determining cyclone frequency was used by Hosler and Gamage (1956), Klein (1957, 1958), Dickson and Namias (1976), Whittaker and Horn (1981), and Olson et al. (1984). Total cyclone counts per box were determined for each winter and for the five decades (1950s–90s). The decadal cyclone counts were compared to the regional number of large snowstorms. Other surface cyclone information related to these large snowstorms including proximity of the storm track to the edge of the heavy snow (>15.2 cm) area, the lowest storm pressure, and movement were obtained from the daily weather maps and characterized. Because storm pressure was only available once per day using the daily weather maps, the lowest storm pressure is biased and most likely was underestimated in this study for many storms.
This manual approach was selected over other more objective approaches using numerical storm-tracking algorithms (Konig et al. 1993; Serreze 1995; Sinclair 1997; Blender et al. 1997; Sickmoeller et al. 2000; Simmonds and Keay 2000; Gulev et al. 2001; Hoskins and Hodges 2002; Rudeva and Gulev 2007) because this study related surface cyclone tracks to the location of specific large snowstorms events (identified through an arduous examination of point heavy snowfall values across the study region). Previous research (Gulev et al. 2001) showed that semimanual tracking using the National Centers for Environmental Prediction (NCEP) reanalysis provided reasonable results over North America.
a. Large snowstorm spatial and temporal characteristics
The time and space distributions of the large snowstorms (i.e., storms producing >15.2 cm of snowfall over 258 000 km2 or more area in the central and/or eastern United States) were defined, as a basis for assessing the surface cyclone track findings. The National Climatic Data Center (NCDC) climate divisions (Fig. 3) were used to determine the spatial distributions of 241 large snowstorms that occurred in 50 winters (1950/51–1999/2000). Only large snowstorms impacting eastern Colorado were used to determine counts for the Southwest climate division. Results of this analysis are biased because of the different sizes of each division; however, for climatological purposes this spatial distribution was viewed as useful. Also, because regions outside the U.S. boundary (e.g., in Canada) or over the Great Lakes and Atlantic Ocean, which might have experienced >15.2 cm of snowfall but were not counted toward the total area for large snowstorms, the total storm counts in divisions along these boundaries would be larger than shown. The regions with the highest number of large snowstorms included the east–north-central, west–north-central, and the central. The Changnon and Changnon (2007) study showed that New York experienced more snowstorms than any other state, suggesting that many of the snowstorms impacting New York were smaller storms including those influenced by the Great Lakes or the Appalachian Mountains, and were not in the large class.
The number of large snowstorms per winter was determined for the area east of the Rockies (Fig. 4). The average was 4.8; however, there was great year-to-year variability (i.e., standard deviation of 2.2) with the numbers ranging from 1 large snowstorm per winter (1973/74, 1980/81, and 1991/92) to 10 storms (1993/94). The winter values did not exhibit any long-term up or down trend during the 50-yr period. A 5-yr running mean also had no trend. The winter counts for each decade (Fig. 5) showed that the frequency peaked in the 1950s with a minimum occurring in the 1970s and 1980s. None of these decadal mean values were significantly different than the average for the rest of the period when using the Student’s t test.
The decadal departure from average values for each climate division (Fig. 6) were investigated to determine how they related to the time distribution of cyclones. Three temporal patterns were apparent: 1) regions with their maximum occurrence in the 1950s followed by their minimum in the 1960s and then a slow increase thereafter (Southwest, west–north-central, and east–north-central); 2) regions with their maximum in the 1960s decreasing to a minimum in the 1980s or 1990s (central, Northeast, and Southeast); and 3) the South, which experienced its minimum in the 1960s (similar to 1 above) but had its maximum in the 1990s. There were few significant decadal differences worth noting; the west–north-central’s value in the 1950s was significantly different from the average of the rest of the period at t = 95% and its minimum value in the 1960s was significantly different at the t = 90% level; the peak value for the east–north-central in the 1950s was significantly different at the t = 99% level; the peak value for the Northeast in the 1960s was significantly different at the t = 90% level; and the peak value for the Southeast in the 1960s was significantly different at the t = 95% level.
b. Surface cyclone characteristics associated with large snowstorms
The decadal average number of cyclone counts was determined for each 5° by 5° grid box (Fig. 7). The spatial distribution of cyclone counts further highlights three primary cyclone tracks associated with large snowstorms: one located from the leeward side of the south-central Rockies east-northeast toward the Great Lakes; a second from the lower Mississippi River basin northeastward toward the Great Lakes; and a third along the coastal mid-Atlantic region northeast to Maine. These cyclone tracks compare well to findings from earlier snowstorm studies (Changnon 1969; Beckman 1987; Branick 1997; Kocin and Uccellini 2005) that identified the “Colorado low” and related “Oklahoma hooker” track, the Gulf of Mexico low and either its northward track impacting the Great Lakes region or an eastward track to off the North Carolina coast, and the development of the northeasters that influenced New England.
The decadal patterns of departure from average cyclone counts were developed for each of the five decades (Figs. 8a–e). The ranked distribution of decadal departures from the average were broken down into three approximately equal-sized groups: above average [decadal departures ≥1 (darkly shaded boxes)], average [decadal departures between −1 and 1 (no shading)], and below average [decadal departures ≤−1 (lightly shaded regions)]. Those boxes with one asterisk had a decadal value significantly different than the rest of the period at the 90% level (using the Student’s t test), those with two asterisks were significant at the 95% level, and those with three asterisks were significant at the 99% level.
The cyclone tracks that originated to the lee of the south-central Rockies (Colorado lows) and moved east-northeast had their highest frequency in the 1950s (e.g., 27 cyclones moving through parts of grid boxes 3 and 8, and 23 cyclones moving through boxes 12, 13, and 17). This maximum frequency of cyclones in this general track explains the peak occurrences of large snowstorms in the Southwest, west–north-central, and east–north-central climate divisions during the 1950s (Fig. 6), with the areas of heavy snow (>15.2 cm) falling north and west of the surface cyclone track. A secondary peak in both Colorado lows and snowstorm occurrence in these three divisions occurred in the 1990s. The minimum frequency of snowstorms in the west–north-central and east–north central divisions in the 1960s and 1970s, and that for the South in the 1960s, can be explained by the pattern of reduced cyclones tracking through grid boxes 3, 8, 9, 12, and 13 during those decades.
The peak number of large snowstorms in the Southeast and Northeast appears to be related to the greater number of cyclones tracking through grid boxes 23, 24, 28, 29, 32, 33, 36, and 37 (coastal mid-Atlantic states track). The snowstorm minimum in the Northeast division (Fig. 6) related well to the below-average cyclone counts in grid boxes 28 and 33 in the 1980s and 1990s (Figs. 8d,e). The peak (1960s) and minimum (1980s) decadal large snowstorm counts in the central region appear best related to the shift in number of cyclones moving through boxes 23, 24, and 33 (decadal departure from average values ranging from 2.2 to 7.6 in the 1960s and −0.8 to −5.4 in the 1980s).
The average orthogonal distance from the cyclone track to the edge of the heavy snow (>15.2 cm) was calculated for the duration of the 241 large snowstorms (Fig. 9). The average distance was 201 km with the area of heavy snow located to the left of the cyclone track; however, the values ranged from −80 km (four storm tracks located within the heavy snowfall region) to 1127 km. Approximately 86% of all large snowstorms had an average distance equal to or less than 400 km. Based on examination of the 241 large snowstorms, a conceptual model of the relationship of cyclone track to the leading edge of the heavy snowstorm was developed (Fig. 10). The most common direction of the cyclone tracks associated with large snowstorms was southwest to northeast (approximately 230°), however, the distance to the edge of the heavy snow generally decreased during the event, and thus its axis was approximately 250°. This model did not consider strength or size of cyclones, but rather it examined the distance from the cyclone track to the area of heavy snow related to large snowstorms. A similar conceptual model was found by Changnon (1969) in his evaluation of Illinois heavy snowstorms between 1900 and 1960.
The minimum surface pressure was used as a measure of cyclone intensity (Table 1). Using once-daily measured pressure values taken from the daily weather map series (NOAA 2007), the lowest pressure found was 959 hPa and the highest of the lowest pressures was 1013 hPa. The distribution of cyclone minimum pressures (Table 1) showed that 67.7% of all large snowstorms had a surface cyclone pressure between 980 and 999 hPa at one time during its track across the United States. A limitation associated with using daily weather maps is that the minimum pressure values do not always occur at the time of observation.
Average surface cyclone speed for the duration of the each large snowstorm was computed (Table 2). These speeds ranged from less than 483 to more than 1930 km day−1, a wide difference. Nearly 57% of cyclones experienced an average speed of between 805 and 1287 km day−1. It was not uncommon for cyclone speed to vary from day to day, and thus the average speed for the duration of the storm represents a value easily compared among large snowstorms. These results were similar to those found by Beckman (1987).
Large snowstorms, those creating large economic impacts, occurred on average about five times per winter in the eastern two-thirds of the United States. However, large year-to-year variability existed, with three winters only experiencing 1 large snowstorm and one winter experiencing 10 large snowstorms. These U.S. snowstorms, which produced >15.2 cm in a 1–2-day period over an area >258 000 km2, were most common in the east–north-central region (i.e., western Great Lakes), the west–north-central region (i.e., northern high plains), and the central region (i.e., lower Midwest). Regions outside the U.S. boundary (e.g., in Canada) or over the Great Lakes and Atlantic Ocean, which might have experienced >15.2 cm of snowfall, were not counted toward the total storm area and this may have biased the results in climate divisions near these political/geographic boundaries.
Although the national trend in the occurrence of large snowstorms through 50 winters showed a maximum in the 1950s and a minimum in the 1970s and 1980s, the regional analysis of large snowstorms indicated three regional trends. The Southwest, west–north-central, and east–north-central divisions experienced a peak frequency in the 1950s with a minimum in the 1960s. The central, Northeast, and Southeast divisions experienced their peak in the 1960s with a minimum in either the 1980s or 1990s. The South division experienced its minimum in the 1960s, with a maximum in the 1990s. These regional differences in time distributions indicate the need to examine changes in the occurrence of climate extremes at spatial scales smaller than continental or national.
The regional time distributions in large snowstorm frequency were found to be related to temporal changes in cyclone counts (examined using the daily weather maps) associated with three primary tracks: 1) from the south-central Rockies east-northeast toward the upper Great Lakes; 2) from the southern Mississippi River Valley northeast toward the upper Great Lakes; and 3) from the coastal mid-Atlantic region northeast toward Maine. The average distance from the storm track to the edge of the heavy snow was 201 km, with nearly 86% of all storm tracks located less than 400 km from the heavy snow. Average cyclone movement associated with these large snowstorms ranged from less than 483 to more than 1930 km day−1 with more than 57% of cyclones moving between 805 and 1287 km day−1. The minimum surface pressure associated with these large snowstorms ranged from 959 to 1013 hPa with more than 67% of all storms having a minimum surface pressure between 980 and 999 hPa. There are a number of ways to further extend this research, and future efforts will go into mining available datasets to expand the results described herein.
This research was funded by a grant from NOAA and NASA, as part of the Climate Change Enhanced Data Set Project, under NA16GP1585. The views expressed herein are those of the authors and do not necessarily reflect the views of NOAA or NASA or their subagencies. This project was also supported through a Northern Illinois University (NIU) University Research Apprenticeship Program (URAP) award in the spring of 2006. The authors thank Jodi Heitkamp of Northern Illinois University’s cartography laboratory for developing the figures for this manuscript and Stan and Suzy Changnon for their editorial comments.
Corresponding author address: David Changnon, Meteorology Program, Department of Geography, Northern Illinois University, DeKalb, IL 60115. Email: firstname.lastname@example.org