Blizzards are extreme winter storms that are defined by strong winds and falling or blowing snow that significantly reduces visibility for an extended period of time. For the conterminous United States, blizzard occurrence by county was compiled from Storm Data for 55 winter seasons from 1959/60 to 2013/14. Spatiotemporal patterns were examined for blizzard seasons (September–August) at annual, decadal, and monthly frequencies. Linear regression and spectral analysis were used to detect any blizzard cycles or trends. Societal impacts such as fatalities, injuries, property damage, and federal disaster declarations were also tallied. Data revealed 713 blizzards over the 55 years, with a mean of 13 events per season. Seasonal blizzard frequency ranged from one blizzard in 1980/81 to 32 blizzards in 2007/08. The average area per blizzard was 83 474 km2, or approximately the size of South Carolina. Blizzard probabilities ranged from 1.8% to 76.4%, with a distinct blizzard zone in North Dakota, western Minnesota, and northern South Dakota. Every month except July, August, and September has reported blizzards with a peak occurrence in December and January. Federal disaster declarations resulting from blizzards totaled 57, with more than one-half of them occurring in the twenty-first century. Storm Data attributed 711 fatalities during the 55-yr study period, with an average of one individual per event; 2044 injuries were reported, with a mean of nearly three per blizzard. Property damage totaled approximately $9.11 billion in unadjusted dollars, with an approximate mean of $12.6 million per storm.
Winter-storm events that include heavy snow, ice storms, frigid temperatures, and blizzards have received widespread media attention in recent years (e.g., Rice and Stanglin 2016; Samenow 2016; McClam 2015). Since 2012, the Weather Channel cable television station (The Weather Company, LLC) has named significant winter storms each season analogous to the naming conventions of tropical-cyclone basins (Weather Channel 2012; Niziol 2015). Some North American winter-weather episodes, such as the Arctic airmass outbreaks in early 2014 and the mid-Atlantic snowstorms of 2010, came to be known by dubious monikers such as the “polar vortex” (LeComte 2015) and “Snowmageddon” (LeComte 2011), respectively. The heightened media interest coupled with an above-average number of winter-storm events has spurred investigation into whether this apparent increase in severe winter weather is part of an overall upward trend and/or a changing climate (e.g., Burnett et al. 2003; Kunkel et al. 2009; Edwards et al. 2014; Lawrimore et al. 2014).
Atmospheric hazards cause significant disruption and damage to various socioeconomic sectors, with annual U.S. dollar losses in the billions (https://www.ncdc.noaa.gov/billions/events, accessed 24 February 2016); winter-storm losses in the United States average $1.2 billion per year (Munich RE 2015). Winter storms may have an adverse impact on agriculture (including livestock), transportation and utilities infrastructure, building integrity (e.g., roof collapse), commerce, school and work schedules, and human health (e.g., Rooney 1967; Babin 1975; Changnon and Changnon 1978; Graff and Strub 1975; Helburn 1982; McKay 1981; Call 2010; LeComte 2012; Potter 2014). These effects are most likely in the conterminous United States in the northern latitudes that have snow and subfreezing weather conditions that may last for several months in the winter (Stommel 1966; Babin 1975; Graf and Strub 1975; Changnon et al. 2008) and in more southerly locations where snow and extreme cold are more rare and the local population is less prepared (Conner et al. 1973; Suckling 1991; Mote et al. 1997). The “superstorm” of 1993 and the New England blizzard of 1978 were two of the top weather events in the United States in the twentieth century (Lipman 2012). In the current century, the Groundhog Day blizzard of 2011 and the eastern winter storm of January 2014 are two winter events that resulted in multibillion dollar losses and collectively resulted in more than 50 fatalities (data taken from Storm Data; see section 2).
An extreme form of winter storms is the blizzard, which combines strong winds with falling or blowing snow to cause the potential of low visibility, deep snowdrifts, and extreme wind chill. As defined by the U.S. National Weather Service (NWS; http://w1.weather.gov/glossary/index.php?word=blizzard, accessed 1 November 2015), blizzards are identified by winds over 16 m s−1 (35 mi h−1) and falling or blowing snow that causes visibility to be less than 400 m (0.25 mi) for at least 3 h. Schwartz and Schmidlin (2002, hereinafter SS2002) and Wild (1997) provide a detailed discussion on the historical evolution of the term blizzard in comparison with current meteorological operational usage.
The intense blend of wind and snow from blizzards causes a general cessation of routine societal activities and severe damage to socioeconomic infrastructures. Transportation issues include closures of highways, railroads, and airports, often stranding thousands of travelers and commuters, such as occurred during the February 2015 blizzard in the Northeast (Mutzabaugh 2015). Retail sales and employee absenteeism can be affected, with reduced business and lost wages that potentially cause financial losses to business and industry (Burrows et al. 1979; Matthews 2015). Extensive drifting snow on roofs can lead to structural collapse from extreme snow loads, as occurred during the collapse of the Metrodome football stadium in Minneapolis, Minnesota, during a December 2010 blizzard (CNN 2010). Ranchers in South Dakota experienced devastating impacts from an October 2013 blizzard when an estimated 70 000 farm animals perished in the storm (Goodman 2013b) and area damages totaled $1.7 billion (Goodman 2013a). Although these examples highlight the adverse impacts associated with blizzards, the literature on winter-storm preparedness, as noted by the Federal Emergency Management Agency (FEMA), often fails to adequately discuss the dangers with these most severe winter storms and areas prone to blizzard occurrence (e.g., FEMA 2015a).
With the exception of SS2002, few modern studies have focused exclusively on blizzards, and discussions about blizzards are usually incorporated into an overall discussion on winter storms in the United States or specific regions and case studies (e.g., Zielinski 2002; Changnon and Changnon 2005; Changnon et al. 2006; Thomas and Martin 2007). The purpose of this research is to update the blizzard “climatology” of the conterminous United States that is given in SS2002 and Schwartz (2001), extending the analysis period from 1959/60 through the 2013/14 winter season (see section 4 for a brief summary highlighting comparisons with SS2002). Data were examined for decadal, annual, and monthly spatiotemporal patterns and trends; the additional twenty-first-century data may reveal features that were not readily apparent during the shorter period of record and in the largely pre-NWS-modernization era. Societal impacts from storm events were analyzed for fatalities, injuries, property losses, and crop damages. Federal disaster declarations associated with individual blizzards were also investigated. In comparison with other atmospheric hazards (e.g., hurricanes), federal relief funds for winter events are relatively recent (largely since the late 1990s). The results provide a resource for studies on climate change and variability, winter-weather forecasting, emergency management, transportation planning, and natural-hazard assessment.
2. Data and methods
Blizzard records for the conterminous United States were compiled from Storm Data (SD), a monthly publication of the National Oceanographic and Atmospheric Administration (NOAA) National Centers for Environmental Information [NCEI; formerly the National Climatic Data Center (NCDC)] that is available in digital form (https://www.ncdc.noaa.gov/IPS/sd/sd.htm). Storm Data chronicles severe-weather events and unusual meteorological phenomena by state, detailing the time period, counties and/or forecast zones that were impacted, storm-event type, general storm paths, and total monetary, human-life, and agricultural loss. Since SD relies on submissions from an NWS Warning Coordination Meteorologist (WCM), storm specifics may vary depending on the report preparation and event investigation of the WCM (or designated focal point) and their local NWS office. Storm Data does serve as an official archive of storm events in the United States and is commonly used in severe-weather climatological studies (e.g., Smith et al. 2013; Schultz et al. 2009; Gaffin and Parker 2006; Curran et al. 2000), including extreme winter weather (e.g., Grout et al. 2012; Rauber et al. 2001; Branick 1997).
For September 1959–August 2015, blizzard events from SD were identified using the procedure outlined in SS2002 and Schwartz (2001) and briefly described here. All monthly publications and digital entries (including any later corrections) were reviewed for storm features that reference blizzards. Only events characterized specifically as blizzards in the event-type column and/or text description were used; thus, “near-blizzard conditions” and other similar phrasing for blizzardlike situations were not included. Blizzards that occurred over several states but were caused by the same synoptic-scale disturbance were considered to be a single blizzard event [after a similar method in Branick (1997)].
Blizzards that are reported in SD meet NWS criteria for reduced visibility from blowing snow, sustained winds, and duration, and the reports undergo quality-control procedures by NCEI/NCDC prior to publication (NOAA 2007). Daily synoptic weather charts (from the NOAA Central Library Data Imaging Project; http://www.lib.noaa.gov/collections/imgdocmaps/daily_weather_maps.html, accessed 5 March 2016) and/or other meteorological data (e.g., satellite and radar imagery) were also used to verify the likelihood of blizzard occurrence in more unusual circumstances (e.g., nonwinter, geographically isolated, or no narrative). The investigation identified only one erroneous “blizzard” event, a supposed July 2002 storm in central California; this event was removed from further analysis.
The counties that were impacted by each blizzard were determined by the SD event listing and/or storm narrative. Earlier SD reports (approximately 1960s–mid-1980s) varied considerably in the geographic preciseness of blizzard occurrence among WCMs and their NWS offices, often being relegated to general regions such as “northeastern Illinois” or “western Kansas.” In those cases, forecast zones were used to assign counties. Later SD reports consistently identified storm events by specific counties and/or forecast zones in each state, and the digital database (since January 1996) was reported in this manner. A 5-year overlap (from 1996 to 2000) existed between SS2002, which relied on printed volumes, and the digital database; no discrepancies in blizzard events between the previous and updated databases were identified.
For each blizzard event, the dates, affected states, and associated socioeconomic impacts were compiled from SD. Property damages (in U.S. dollars) were estimated from insurance adjusters, power utility companies, emergency-management agencies, and newspaper articles (NOAA 2007). Crop-damage data were obtained from the U.S. Department of Agriculture, crop-insurance firms, and other agricultural agencies; although crop-damage data in SD ended in the late 1990s, coinciding with when the NWS phased out their agricultural products to the private sector [J. Logsdon and L. Fisher, National Weather Service–Northern Indiana (IWX), 2016, personal communication]. According to the SD preparation guidelines (NOAA 2007), fatalities and injuries were reported as either directly or indirectly attributable to the meteorological event; indirect fatalities and injuries (e.g., traffic accidents) from winter events may be severely underestimated (Black and Mote 2015; Gall et al. 2009).
The annual blizzard season was defined as from September through August, with January used to define the blizzard year (e.g., the 1960 season refers to all blizzards occurring between September 1959 and August 1960). The annual number of blizzards and annual area affected by blizzards were analyzed for temporal trends using linear regression and a spectral analysis to detect any blizzard cycles. Annual blizzard probabilities for each county were determined empirically by dividing the number of seasons with a blizzard by the number of years in the database (n = 55).
Annual, decadal, and monthly blizzard frequencies were mapped by county for the conterminous United States for a visual examination of the spatiotemporal changes in blizzard events. To account for large variations in county size, the blizzard-count data were normalized by area using the following procedure for each county: 1) averaging the number of blizzards occurring within a 28.196-km radius of the county center [the radius was based on the average size of a contiguous U.S. county area (2497.59 km2)], 2) dividing the resulting blizzard average by the mean contiguous U.S. county area, and 3) multiplying the result by 1000 km2. In other words, the resultant maps (i.e., Figs. 2 and 5–7, described in more detail below) show the average number of blizzards per 1000 km2 that a specific county would have if every county in the continental Unites States were to have the exact same area.
a. Annual blizzard frequency and trends
From the SD reports, the conterminous United States recorded 713 blizzards of various size, intensity, and duration between 1959/60 and 2013/14. The average number of blizzards per season (September–August) is 13.0, with a standard deviation of 7.0, indicating the relatively high interannual variability in blizzard frequency (Fig. 1). Seasonal blizzard frequency ranged from 1 blizzard in the 1980/81 season to 32 blizzards in the 2007/08 season. On the basis of 3-month intervals, blizzard frequencies, as expected, are highest during the boreal winter peak (December–February) at 8.3. Late-season blizzards (March–May) occurred 2 times as often as early-season blizzards (September–November), averaging 3.1 per season as compared with 1.5. Summer blizzards were only reported once, occurring in June 2002.
Seasonal blizzard frequencies displayed a distinct upward trend, with a more substantial rise over the past two decades. A linear-regression analysis revealed a statistically significant positive trend in blizzards over time (β = 0.29; p ≪ 0.000) with a standard error of the estimate of 5.3. The occurrence year explained almost one-half of the variability in seasonal blizzard frequency (coefficient of variation R2 = 0.438; p ≪ 0.000). The modeled increase in blizzard activity showed a nearly fourfold upsurge between the start and end of the study period at 5.9 and 21.6 blizzards, respectively. On the basis of current model trends, the expected blizzard total for a season is 32 blizzards by 2050; uncertainty exists on whether the linear trend will continue or stabilize in the near future.
Figure 1 also displayed a noticeable cycle in blizzard frequency, with a peak every 11–14 years. A spectral analysis was conducted on the blizzard time series from the detrended dataset using standardized residuals from the linear-regression analysis. Although blizzard frequency peaks coincide with a 13- or 14-year cycle, with a secondary peak at 4 years, the peaks could not be separated from independent errors (or “white noise”) and were found to be statistically insignificant as based on Fisher’s test for periodicity (Bloomfield 2000).
In addition to an increased frequency over time, blizzard occurrence (shown as the average number of blizzards per 1000 km2) between the 1959/60 and 2013/14 seasons showed a wider geographic distribution (Fig. 2). Only six states in the conterminous United States (Alabama, Florida, Louisiana, Mississippi, South Carolina, and Tennessee) do not have any officially reported blizzards for their counties in the SD publication database; meteorological data and the spatial pattern of snowfall distribution from the March 1993 “Storm of the Century” (Lott 1993) suggest that blizzard conditions may have occurred in eastern Tennessee, western South Carolina, and northeastern Alabama.
Blizzard activity was strongly concentrated in the northern Great Plains, particularly in the Dakotas and western Minnesota (or the “blizzard zone” from SS2002). Over the study period, counties in this active region averaged between approximately 26 and 42 blizzard events per 1000 km2—values that are not found outside the blizzard zone (Fig. 2). Nearly all 119 counties in North and South Dakota and 39 counties in western Minnesota averaged at least one blizzard or more per year (i.e., at least 55 blizzards occurred during the 55-year study period). Along the Minnesota border with the Red River, five counties in eastern North Dakota (Pembina, Walsh, Grand Forks, Traill, and Cass) reported over 100 blizzards during the study period. Traill and Cass Counties had the maximum blizzard frequency for the contiguous United States at 111 blizzards and the highest blizzard average per 1000 km2 at 42.4. In addition to favorable synoptic-scale conditions, explanations for the high blizzard occurrence in eastern North Dakota (as opposed to the entire blizzard zone) could include 1) local WCM decisions on how winter events should be cataloged, thus creating many of the abrupt changes in blizzard occurrence at state and/or county warning-area boundaries, 2) relatively higher potential for population impact, leading to increased blizzard-event classification, and 3) local terrain differences, such as the transition zone from forests in the east to open plains in the west, that may have a large effect on wind speeds (J. Logsdon and L. Fisher, IWX, 2016, personal communication).
Blizzard totals remained relatively high on the blizzard-zone margins (Fig. 2). Eastern Minnesota, northwestern Iowa, northern Nebraska, and southeastern Wyoming had the average blizzards per 1000 km2 range from approximately 17 to 25. Blizzard averages decreased from 9 to 16 blizzards per 1000 km2 for eastern Minnesota, Nebraska, western Kansas, eastern Colorado, and isolated areas of Montana and Idaho. As with the Dakotas, relatively sharp gradients between neighboring states (e.g., New Mexico and Colorado) were observed and may reflect local WCM decisions on blizzard reporting.
Outside the blizzard zone and its periphery, blizzard totals were generally between 1 and 11 blizzards per county (or an average of 9 or less per 1000 km2) with higher frequencies generally in the northern and high-altitude locations, with a few exceptions. A secondary blizzard (geographic) maximum occurred in eastern Maine (Hancock and Washington counties) with 27–48 blizzards (or an average of approximately 9–17 per 1000 km2); this pattern coincides with typical New England northeaster tracks. Higher regional blizzard totals (n = 12–26) also occurred for counties in coastal New England, interior Maryland (Garrett County), and lakeside New York (Erie County). With the exception of northwestern Georgia, blizzards were overall absent in the Southeast and the low-elevation areas in western states (Arizona, California, Nevada, and Washington). The blizzards that have occurred in the southeastern United States were largely confined to high-elevation locations in the Appalachian Mountains.
b. Annual area affected by blizzards
The total area affected by blizzards each year displayed strong interannual variability (Fig. 3, top panel), ranging from a minimum of 149 652 km2 in 1981 to a maximum of 2 167 847 km2 in 1971. As one would logically expect, annual blizzard area was directly correlated to annual blizzard frequency (correlation coefficient r = 0.52; p ≪ 0.000), but some winter storms produce blizzards over vast areas in an otherwise low-frequency year. For instance, the Storm of the Century during 12–14 March 1993 created blizzard conditions over 720 109 km2 (278 142 mi2) from Georgia to Maine, occurring at the end of a relatively quiet blizzard season (n = 10). The Storm of the Century accounted for nearly one-half (44%) of the total affected blizzard area (1 634 707 km2) for the 1992/93 season. By comparison, the blizzards for the 2013/14 season affected an almost identical total area (1 663 152 km2) but occurred during a very active season (n = 29). Although blizzard counts displayed a positive trajectory over time, the total area affected by blizzards each season displayed no significant linear trend (β = 1937 km2; p = 0.67).
The average blizzard size of individual events has decreased substantially, particularly in the twenty-first century (Fig. 3, bottom panel). Unlike the total area, the average area has significantly decreased over time (β = −1892 km2; p ≪ 0.000). The average area impacted by the 713 blizzards over the study period is 83 474 km2 (or 32 229 mi2), approximately the size of South Carolina. The 24–27 January 1978 Midwestern winter storm remains the largest individual storm to impact the most area (1 054 799 km2), eclipsing the event with the second-highest total area (821 409 km2) that occurred 14–16 December 2000 over the northern Great Plains and upper Midwest. The two smallest blizzard events were recent and were confined to single counties: Cherokee County (1179 km2) in extreme western North Carolina on 5 March 2013 when a strong pressure gradient developed on the backside of a Cape Hatteras low, and Door County (1248 km2) in northeastern Wisconsin on 18 January 2005 when a low pressure system was rapidly generated across the western Great Lakes region.
Trends suggested that smaller blizzards events such as these 2013 and 2005 examples were driving some of the higher annual-count totals and were decreasing the average blizzard area. When the blizzard area was examined by quartiles, the lowest quartile (i.e., the smallest 25% of all blizzards) showed that nearly one-half of all events (49.7%) have occurred since 2000. As with other atmospheric hazards, reports from sources other than staffed and automated weather stations, such as emergency-management officials and newspapers, may also increase the number of smaller events. The digital SD database denotes that the primary information source for smaller blizzards (i.e., isolated counties) often stems from trained storm spotters and law enforcement. In the past decade or so, social media (e.g., Facebook, Twitter, or NWSChat) may have also alerted the NWS to potential blizzard conditions, particularly those in remote regions and impacting small areas. As a consequence, as the geographic precision for blizzard-impacted areas improves, the number of events may increase, thereby decreasing the annual-average blizzard area.
c. Annual blizzard probability by county
The annual probability of a blizzard for an individual county was calculated as the total number of years with a blizzard occurring during a season divided by the total number of years in the study period (n = 55). Probabilities ranged from 1.8% to 76.4% and were plotted by county in a five-class, equal-interval choropleth map (Fig. 4). Blizzard probabilities were highest in the blizzard zone and surrounding area, with blizzards expected on average at least every other year (or 50%). The highest blizzard-probability class (61.6%–76.4%) occurred for all counties in North Dakota, 16 counties of western Minnesota largely bordering the Red River, and 5 counties in northern South Dakota. Four counties in east-central North Dakota (Barnes, Benson, Cass, and Traill) tied at 76.4% for having the highest blizzard probability in the conterminous United States. Blizzard probabilities remained relatively high (approximately one year in three or more) throughout the central plains, upper Midwest, and eastern Rockies, an area comprising Minnesota, Nebraska, Kansas, Iowa, eastern Colorado, and portions of Montana.
Beyond the blizzard-zone region of the northern and central Great Plains, blizzard probabilities diminished considerably. Maine has the largest blizzard probabilities outside the central United States at 25% or more (one year in four), including three northeastern counties (Aroostook, Hancock, and Washington) that had blizzard probabilities comparable to the blizzard-zone periphery at 40%. Other New England counties in eastern Massachusetts, southern New Hampshire, and Rhode Island also had regionally high blizzard probabilities of about one blizzard expected every five years; isolated counties in mountainous regions (e.g., Garrett County, Maryland) and lake-effect-snow areas (e.g., Muskegon County, Michigan, and Erie County, New York) had similar probabilities. The lowest nonzero blizzard probabilities of 1.8% (or one documented blizzard) occurred for 375 counties, largely across the southern and western fringes of where blizzards were recorded.
d. Decadal variability in blizzard frequency
The geographic distribution of blizzards by decade displayed distinct periods of concentrated blizzard events contrasted with phases of more widespread activity (Fig. 5). In spatial terms, blizzards in the 1960s (Fig. 5a) and 1980s (Fig. 5c) were more focused in the northern Great Plains blizzard zone with secondary activity around the extreme Northeastern coast. Although the blizzard activity was also more concentrated in the 2000s (Fig. 5e), blizzard reports shifted westward from the Dakotas-centered blizzard zone to the mountainous regions of the western United States (Rockies, Cascades, and Sierra Nevada), with a secondary geographic blizzard peak limited to southeastern Maine. In comparison, blizzard reports from the 1970s (Fig. 5b) were prevalent throughout the north and were shifted south of their mean position to include states such as New Mexico, Oklahoma, and North Carolina where severe-winter-weather events were rare. The 1990s (Fig. 5d) also had widespread blizzard activity that also included much of the western United States as well as the entire Northeast corridor. On the basis of the 2010–14 seasons, blizzard activity for the 2010s (Fig. 5f) demonstrated a trend toward more geographically extensive blizzard reports, with current activity widely spread throughout the central United States and Rockies; the blizzard zone of the northern Great Plains remains the most active blizzard area, however.
e. Monthly blizzard frequency
Blizzards have been reported in all months except July, August, and September. Monthly blizzard occurrence highlighted a more active blizzard season (December, January, February, and March; Fig. 6) and a less active blizzard period during the transitional seasons (October, November, April, and May; Fig. 7). Blizzard values were highest in January, ranging from an average of 9.3 to 13.6 per 1000 km2 throughout the blizzard zone and extending into northern Iowa. Actual blizzard totals were highest in December (n = 168), with the highest December average blizzard values being approximately 7 blizzards per 1000 km2 and confined to extreme eastern North Dakota. February and March have similar average blizzard values (ranging from 0.4 to 6.5 per 1000 km2), although March has higher blizzard activity throughout the blizzard zone and central Great Plains. In the transitional seasons, November (n = 65) and April (n = 56) were 3–4 times as likely to have a blizzard as October (n = 17) or May (n = 7). Blizzard averages in November and April were comparable (0.4–4 per 1000 km2) yet contrast geographically (Fig. 7). November blizzard occurrences were concentrated along an axis from Colorado to Minnesota, with the highest values in eastern South Dakota, whereas April blizzards were highly focused in the western Dakotas, Montana, and northeastern Colorado; the straight diagonal axis in the central plains is a result of the trajectory of Colorado lows and the averaging technique employed. June (not shown) had only one reported blizzard, occurring in 2002 over a small mountainous region of Montana; this is the latest-season blizzard reported in SD.
f. Blizzard-related disaster declarations and societal impacts
For locations to receive a federal disaster declaration, a three-phase process through a bureaucratic hierarchy is followed. First, the local jurisdiction determines the financial and humanitarian needs that cannot be met and then requests emergency assistance from the state. The state governor may declare a state of emergency, invoke state disaster-plan protocols, and utilize state resources. Second, the governor may request federal aid if local and state government resources combined are deemed insufficient and the situation meets the guidelines of the Stafford Act for disaster relief and emergency assistance (FEMA 2015b). Third, FEMA conducts a preliminary damage assessment and gives a recommendation to the president. Between 1953 and 2014, the number of major disaster declarations totaled 2202, nearly all of which were atmospheric-related hazards.
In comparison with mesoscale events (i.e., tornado outbreaks) and tropical cyclones, blizzards composed a small proportion of federal disaster declarations and were a subset of the more generalized “winter storm” category. Between the 1959/60 and 2013/14 seasons, federal disaster declarations due to blizzards totaled 57, displaying a visible increase from the mid-1990s that coincided with the increase in reported blizzard activity (Fig. 8). Over one-half of the blizzard declarations (n = 33) occurred in the twenty-first century alone, with the active 2009/10 blizzard season over the mid-Atlantic and Great Plains being the highest individual season (n = 9). In comparison, the disaster declarations in earlier decades (n = 24) were highly concentrated in the late 1970s and late 1990s and were largely confined to the Dakotas and northeastern Michigan (Iosco County). The first federally declared blizzard declaration occurred in the 1974/75 season with the 10–12 January 1975 blizzard that impacted the Dakotas, Minnesota, Iowa, Nebraska, and Missouri.
Storm Data reports attribute 711 fatalities to blizzards during the 55-yr study period, with an average loss of life of approximately one individual per event. Fatalities ranged from a minimum (and most common value) of no fatalities per storm to a maximum of 73 deaths associated with the late January Midwest blizzard in 1978. Although blizzard frequencies increased substantially in the twenty-first century, blizzard fatalities actually decreased when compared with earlier decades. From the 1959/60 season through the 1999/2000 season, blizzards accounted for 679 fatalities, with an average of 1.55 deaths per event, as compared with only 32 fatalities, with an average of 0.12, for the 2000/01–2013/14 seasons. Reported injuries yielded similar results. All blizzard events for the study period totaled 2044 injuries, with a mean of 2.87 per blizzard. As with fatalities, most events had zero reported injuries. The maximum number of injuries reported (n = 426) occurred with the March 1993 superstorm over the densely populated East Coast. The period from the 1959/60 season through the 1999/2000 season yielded 2011 injuries, with a mean of 4.59 injuries per event; the 2000/01–2013/14 seasons had only 33 injuries, averaging 0.12. Although more recent (since the late 1990s) SD reports make a distinction between direct and indirect causation, indirect fatalities and injuries from winter events (e.g., snow-related vehicle crashes) are significantly underestimated in the overall SD database (Black and Mote 2015).
Property damage totaled approximately $9.1 billion ($9,011,291,100) in unadjusted dollars, with an approximate mean of $12.6 million per blizzard ($12,620,856). The minimum was zero, and the maximum was $4.5 billion ($4,545,000,000) from the Midwest blizzard in 1978. Crop damage totaled $439.5 million ($439,542,000), for a mean of $1 million ($1,003,520) per blizzard. The minimal was zero, and the maximum crop damage was $227,250,000 from the Midwest blizzard of 1978. As previously noted in section 2, crop damage was not reported in the period from 2000/01 through 2013/14 and its statistics may be influenced by changing reporting methods in SD.
4. The updated blizzard climatology: Comparisons with SS2002
The previous blizzard climatological description, or “climatology,” conducted by SS2002 from 1959/60 to 1999/2000 was updated to the 2013/14 winter season, increasing the dataset by 14 seasons and 275 blizzard events. Although blizzard counts by county were still examined, data were mapped here as average blizzards per 1000 km2 on the basis of the mean size of a conterminous U.S. county to account for variable county size (see section 2). In addition to updating the analyses conducted in SS2002, the major new blizzard investigations included event-frequency cycles, decadal patterns, and trends in federal disaster declarations.
Frequency: Average blizzards per year increased from 10.7 to 13. The blizzard range expanded from 1 to 27 (the 1980/81 and 1996/97 seasons) to 32 in 2007/08. December, January, February, and March continued to be the most active blizzard months, with December (not January) having the largest number of events; average blizzard values were the largest in January in the blizzard zone, however. Blizzards were reported in all months except July, August, and September.
Trends: Annual blizzard counts continued to show a significant upward trend, producing a linear regression model with a steeper positive slope and higher model confidence than were originally reported in SS2002 for the 1959–2000 period. The model yielded a quadrupled (as opposed to doubled) increase from 5.9 blizzards at the beginning of the study period to 21.6 blizzards at the end.
Cycles: Spectral analysis showed a blizzard cycle with a peak at 13–14 years; the results were not statistically significant, however, given the relatively short data record.
Area: Average blizzard size decreased by nearly one-half, from 150 492 km2 (or about the size of Iowa) to 83 474 km2 (or about the size of South Carolina). Smaller blizzard events (often encompassing only two counties) over the past two decades were partially driving the significant downward trend in average blizzard area; this decline was not apparent previously.
Geography: Blizzards were reported in Arizona and Arkansas since 2000; hence, only six southern states now do not have reported blizzards in SD. The earlier defined blizzard zone remained intact, but a secondary blizzard zone has developed in northeastern Maine. Decadal results overall indicated an increased geographic distribution of blizzards.
Deaths and injuries: Blizzard fatalities and injuries decreased substantially despite the increased number of events; indirect casualties may be underreported, however.
Federal disaster declarations and monetary impacts: Federal disaster declarations are becoming more common, with 33 of the 57 total blizzard-related disaster declarations occurring since 2000. Blizzard-related losses totaled nearly $8.6 billion in unadjusted dollars, with an approximate mean of $19.6 million per blizzard in the original study period and of $414.3 million in the 2000/01–2013/14 seasons.
A blizzard climatology for the conterminous United States was developed by updating and expanding upon the initial method outlined in SS2002. Using the SD and the NCEI Storm Event database for the 1959/60–2013/14 winter seasons (n = 55), blizzard events totaled 713 and annually averaged 13. Observations indicated a strong positive trend in annual blizzard frequency, especially in the 2000s as highlighted by the 2007/08 season, which recorded 32 blizzards. A potential explanation for the generally larger annual blizzard counts since the 1990s is that it may be an artifact of greater consistency in common blizzard terminology and reporting parameters, NWS modernization and blizzard detection capabilities, and increased spatial precision from smaller forecast zones and/or geographically specific narratives. Several studies have noted the potential impacts of technological improvements and population changes on entries in SD (e.g., Smith et al. 2013; Agee and Hendricks 2011; Trapp et al. 2006; Branick 1997). These discussions on SD limitations have centered on phenomena that are mesoscale or smaller and not on relatively large events (e.g., blizzards and winter storms) that are much less likely to be missed by human and/or automated observers across the study area and time period (S. Hinson, NOAA/NCEI, 2016, personal communication). As such, these factors may have a minimal impact on blizzard events in SD as compared with other weather phenomena and alone do not account for the overall high spatiotemporal variability in blizzard counts throughout the period, especially given that many of the more recent reports are occurring in areas with consistently high population and station-network densities. Investigations into the physical mechanisms of these patterns, including any climate change implications, are needed.
The revised dataset highlighted a discernable cycle in annual blizzard frequency, peaking approximately every 11–14 years, with a secondary peak every 4 years. A spectral analysis revealed that the peaks are not statistically significant; the result is not unexpected, given the relatively short data record. The noticeable blizzard cycles may also just be an artifact of the data and method employed. Additional analysis (e.g., storm track variability or 11-yr sunspot cycles) is needed to understand the apparent blizzard periodicity that occurs independent of the positive trend in blizzard counts.
While annual blizzard counts increased substantially in the past two decades the average area covered by blizzard events displayed a marked decrease; a regression analysis showed the trend to be significant. The average size of a blizzard event across the entire study period was 83 474 km2, or approximately the size of South Carolina, a nearly 50% reduction in the average blizzard size of 150 492 km2 reported in SS2002 for the 1959/60–2000/01 period. As with blizzard counts, the trend in blizzard event sizes may be related to reporting protocols, increased detection capabilities in remote regions since the NWS modernization era, and greater blizzard awareness through social media. Between the 1990/91 and 2013/14 winter seasons, blizzard events averaged only 61 088 km2, whereas blizzard events in the premodern era (1959/60 to 1989/90) averaged 79 811 km2. Yet a two-tailed t test between the premodern era (n = 30) and postmodern era (n = 25) was not statistically significant (using the α = 0.05 criterion).
Blizzard probabilities on the county level ranged from 1.8% to 76.4%. As expected, blizzard probabilities were highest in the blizzard zone and surrounding areas. The highest blizzard probabilities were in the entirety of North Dakota, 16 counties in western Minnesota (primarily bordering the Red River), and 5 counties in northern South Dakota. Maine has the largest blizzard probabilities outside the central United States at 25%. On a county level, three counties were at 40%, which is similar to the blizzard-zone periphery. The lowest nonzero blizzard probabilities of 1.8% occurred for 375 counties, primarily across the southern and western margins of recorded blizzards.
There were 711 fatalities reported during the 55-yr study period, or an average of one fatality per blizzard. Most storms had no fatalities, and the maximum of 73 deaths was associated with the 24–27 January 1978 Midwest blizzard. Reported injuries had a total of 2044, with a mean of 2.87 injuries per blizzard. Most events had zero injuries, and the maximum number of 426 reported occurred with the March 1993 superstorm, primarily over the densely populated East Coast. Blizzard-related fatalities and injuries displayed a substantial decline in the twenty-first century when compared with previous decades. This declining trend could stem from improved forecasting, warning, and communication mechanisms or from shifts in coding and reporting practices in SD.
Blizzards composed a small proportion of federal disaster declarations and were a subset of the more generalized winter-storm category. Between the 1959/60 and 2013/14 seasons, federal disaster declarations due to blizzards totaled 57, with the first blizzard-specific declaration occurring in January 1975. Over one-half (n = 33) of the declarations occurred in the twenty-first century. Changes in the Stafford Act made federal winter-weather disaster declarations easier, especially after the active 2009/10 season that impacted the mid-Atlantic and Washington, D.C., areas (Lindsay and McCarthy 2015). Atkinson (2010) noted that the increases in blizzard federal disaster declarations since 1987 in the central United States were a function of an increase in storm activity (in particular, in late-season storms that have higher snowfall potential), the increased infrastructure and population in previously remote regions, and changes in federal disaster guidelines for reporting winter-weather events into subcategories (e.g., snow and ice).
In addition to understanding spatiotemporal variations in blizzard occurrence, the blizzard database and the developed climatology are useful for more specialized analyses. Blizzards are an important component for winter-weather risk assessments at various spatial scales (NWS warning areas, emergency management districts, transportation corridors, etc.). Meteorologists, social scientists, and emergency-management agencies will find utility from a preparedness standpoint, recognizing blizzards as a severe and costly atmospheric hazard that is separate from the generalized winter-storm category.
Future research will focus on examining the physical mechanisms for the trends and large interannual variability in blizzard events. The synoptic circulation characteristics associated with the geographic distribution and frequency of blizzards will be investigated, particularly the prevailing storm-track position in relation to changing blizzard activity. Teleconnection patterns (e.g., the Pacific–North American pattern, North Atlantic Oscillation, and Arctic Oscillation) and other sources of climate variability will also be examined for potential linkages.
The authors thank Candace Boren and Nicholas Eckstein at Ball State University for assistance with the GIS data compilation. We also thank the anonymous reviewers for their helpful suggestions in improving the paper.