APRIL 2010

BLUMENFELD

619

The Frequency of High-Impact Convective Weather Events in the Twin Cities Metropolitan Area, Minnesota KENNETH A. BLUMENFELD University of Minnesota, Minneapolis, Minnesota (Manuscript received 16 March 2009, in final form 28 July 2009) ABSTRACT Thunderstorms frequently produce brief flooding or minor damage, though far fewer lead to major flooding and widespread or significant damage. Outbreaks of such storms exact large tolls on their victims and can compromise, or completely overwhelm, the emergency response infrastructure. This paper derives empirical frequencies and recurrence intervals of ‘‘high end’’ convective weather events in the Minneapolis–St. Paul, Minnesota, metropolitan area from archived tornado, hail, damaging-wind, and high-density daily rainfall data, as well as historical records and accounts. Two classes of high-impact events are analyzed: those with the potential to produce widespread damage or disruption and those virtually certain to do so. Storms in this first class recur within the area, on average, 3 times per year, while the more extreme storms recur every 2–2.5 yr on average. Owing to well-established spatial and temporal inhomogeneities in observed severe weather data, true recurrence intervals are probably somewhat shorter. In the context of ongoing regional population growth, the area is becoming increasingly vulnerable to major damage and potential casualties from these major storm events.

1. Introduction Severe convective weather episodes within major metropolitan regions can threaten the safety and property of hundreds of thousands, or even millions, of people. While the probability of any given high-end severe weather event striking a heavily populated area is limited by the relatively small area occupied by both the event, and the city, virtually every major urban area from Texas and the Deep South through the upper Great Lakes has experienced some form of convective weather disaster in its postsettlement history [e.g., see Grazulis (1993) for lists and discussion of urban tornadoes]. Yet, there has been a growing concern among many regional planners, emergency managers, severe weather forecasters, and scientists, that the worst may be yet to come (e.g., Wurman et al. 2007, 2008; Brooks et al. 2008; Hall and Ashley 2008). Aside from high levels of casualties, a major urban tornado event would likely result in extreme property damage, which increases as a function of property density and wealth (Changnon 2001a).

Corresponding author address: Kenneth Blumenfeld, Dept. of Geography, University of Minnesota, 414 Social Sciences, 267 19th Ave. S., Minneapolis, MN 55455. E-mail: [email protected] DOI: 10.1175/2009JAMC2223.1 Ó 2010 American Meteorological Society

Following the deadly, well-documented Oklahoma City (OKC), Oklahoma, area tornadoes on 3 May 1999, the North Central Texas Council of Governments raised the concern that the Dallas–Fort Worth Metroplex (DFW)— which has a population that is approximately 5 times that of OKC (U.S. Census Bureau 2008)—had not experienced a violent tornado event during its era of maximum population growth and property development, despite being in an area that is historically tornado-prone. Rae and Stefkovich (2000) simulated multiple scenarios involving intense tornadoes in modern DFW and estimated that such an outbreak would affect tens of thousands of residents, and tens of thousands of employees, along with thousands of structures worth billions of dollars. Wurman et al. (2007) simulated violent tornadoes in Chicago, Illinois, and other major U.S. cities and arrived at mortality estimates as high as 63 000. While those estimates have been questioned by Brooks et al. (2008) and Blumenfeld (2008), there is broad agreement that a violent urban tornado outbreak would cause significant loss of life, among other calamities. At the same time, there is ample evidence that tornadoes need not be violent [Enhanced Fujita (EF) 4–5; McDonald et al. 2004] to wreak havoc upon urban areas. Tornadoes hitting the skylines of Nashville, Tennessee, in 1998, Fort Worth, Texas, in 2000, and Atlanta, Georgia,

620

JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY

VOLUME 49

FIG. 1. The TCMA study area, with minor civil division population from the 2000 census.

in 2008 were all in the middle of the Fujita (F) or EF damage scales (e.g., F or EF 2–3), yet each caused some casualties, as well as extensive damage. Additionally, an event need not be tornadic, or strike a city’s central business district, to produce significant destruction. A straight-line wind event in the southern portions of the Minneapolis–St. Paul, Minnesota, area on 30 May 1998 caused numerous injuries, knocked out power to nearly half a million customers, and produced widespread structural damage, on its way to becoming one of the longest and most destructive derechos in recorded U.S. history (Togstad 2008; NOAA/NCDC 1998; Ashley and Mote 2005). A long-track supercell on 10 April 2001 originated near Kansas City, Missouri, and passed through St. Louis, Missouri, where it produced most of its .$1.5 billion hail-related losses (Changnon and Burroughs 2003). Flooding in Dallas, Texas, on 5 May 1995 killed 16 people, after dropping softball-sized hail on Fort Worth (Smith et al. 2001). This paper examines recurrence intervals (‘‘return periods’’) of high-impact convective weather events in the Twin Cities Metropolitan Area (TCMA), defined here as a 9600-km2 rectangular area centered just west of Minneapolis and extending into extreme western Wisconsin (Fig. 1). This study domain contains much of the recent suburban growth and expansion described by Adams (2006), and matches exactly the area used previously for local extreme rainfall research (Skaggs 1998; Blumenfeld et al. 2004). The TCMA has an active group

of emergency managers [the Metropolitan Emergency Managers’ Association (MEMA)] who are in the process of developing a 5-yr preparedness strategy, which requires a complete accounting of all known hazards. Indeed, this article is the result of research aimed at informing those emergency managers about the frequencies of meteorological events that could cause one or more municipalities to draw upon the response resources of neighboring communities. Events in this investigation will fall into one of two strata: significant or extreme (Table 1). Significant events have the potential to cause major damage or disruption, while extreme ones are virtually certain to do so. Though not in the traditional ‘‘Tornado Alley,’’1 the TCMA has a history of major tornado events (Grazulis 1993; Seeley 2006), is within a high-frequency derecho corridor (Johns and Hirt 1987; Coniglio and Stensrud 2004; Ashley and Mote 2005), and is on the northern end of a high-frequency axis of high-end ($33.5 m s21) convective wind events (Doswell et al. 2005). The area is somewhat displaced from the spatial maxima of 5.1-cm hail probabilities reported by Doswell et al. (2005), but it nevertheless has occasional bouts of widespread, damaging hail, the most recent being on 24 August 2006 (NCDC 2006). In recent decades the TCMA population

1 Brooks et al. (2003) did include parts of Minnesota in Tornado Alley (see their Fig. 11 and related discussion).

APRIL 2010

621

BLUMENFELD TABLE 1. Types of high-impact convective events, with thresholds required for inclusion in study.

Event type

Significant threshold

Extreme threshold

Tornado Wind Hail Rainfall Combined severe Combined severe and rainfall

F-2 damage or $1 fatality 33.5 m s21 (65 kt) 5.1 cm (2 in.) 102 mm (4 in.) .2 baseline types .1 baseline severe type and baseline rainfall

F-4 damage rating 41 m s21 (80 kt) 7.6 cm (3 in.) 203 mm (8 in.) .2 event types of which .1 must be extreme .2 event types of which 1 must be rainfall and of which .1 must be extreme

has nearly tripled, while its area has increased by a factor of 4 (Fig. 2; Adams 2006), and there is some concern that this expanded population is not prepared for major convective weather events. Though rare, these are part of the region’s climatology; for instance, the area last experienced a multiple, killer-tornado event on 6 May 1965 (Grazulis 1993; Seeley 2006).

2. Approach a. Definitions This investigation uses severe weather event and highdensity rain gauge data to estimate recurrence intervals for significant and extreme severe weather and rainfall events. Here, significant severe weather events are defined as meeting or exceeding at least one of the following

criteria: F-2 or killer tornado, nontornadic convective wind gust $33.5 m s21 (65 kt), or hail $5.1 cm (2 in.). These thresholds have been applied in previous research to distinguish significant severe events from the lessdamaging but widely reported class of ‘‘marginally severe’’ ones (Grazulis 1993; Doswell et al. 2005), and are higher than those used operationally to define severe weather (Moller 2001). Extreme events are defined by the following criteria: F-4–5 tornado, nontornadic convective wind gust $41 m s21 (80 kt), or hail $7.6 cm (3 in.). F-4 and F-5 tornadoes are often classed together as ‘‘violent,’’ and account for the vast majority of tornadorelated casualties (Grazulis 1993; Ashley 2007). Though the wind and hail thresholds are somewhat arbitrary, they represent ‘‘clean’’ breakpoints in the severe weather databases, and also were selected for their similar statistical rarity to the violent tornadoes: in a pilot investigation

FIG. 2. Population of the TCMA, plotted with the total residing in Minneapolis and St. Paul, 1950–2000, and shown with the TCMA total area. Counties used to define ‘‘official’’ TCMA were obtained from Fig. 1 in Adams (2006); area and population data were obtained from the National Historical Geographic Information System (http://www.nhgis.org).

622

JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY

TABLE 2. Estimated return periods for a 102-mm rainfall occurring over a given duration within the study area, adapted from Huff and Angel (1992). The return period values are approximate and were obtained by examining the 4-in. isohyet from the maps given in Huff and Angel (1992, 54–89).

Storm duration (h)

Approx recurrence interval (yr)

1 2 3 6 12 24

350* 200* 100 50 25 10

VOLUME 49

Since much of the urban infrastructure is designed for 25–100-yr storms (Oberts 1984; Huff and Angel 1992), a 102-mm event would quite likely overwhelm drainage capacity, leading to flash flooding and potential damages. The 203-mm threshold chosen for extreme rain events exceeds the regional 100-yr return estimate for a storm of 24-h duration by approximately 33% (the 100-yr storm is 152 mm; Hershfield 1961; Huff and Angel 1992), but moreover, appears to be the lowest value at which significant damage always occurs.

b. Data and editing

* Denotes value estimated by applying the Huff–Angel log-log fitting procedure to plot of storm duration vs return period for 102-mm rainfall.

for the region, just 1.5% of all tornadoes, 1.4% of all hail reports, and 1.3% of all convective wind reports qualified as extreme. The rainfall thresholds for significant and extreme events in this study are 102 mm (4 in.) day21 and 203 mm (8 in.) day21, respectively. The 102-mm threshold approximates the 10-yr return value for 24-h rainfall at a given point in the area (Huff and Angel 1992). Given the convective nature of most intense rainfall events in the region (Winkler 1988; Changnon 2001b), the actual durations, however, are often shorter than 24 h, as was the case on 23–24 July 1987, when much of the southern and central TCMA received $250 mm (approximately 10 in.) in fewer than 6 h (Kuehnast et al. 1988; Schwartz et al. 1990). The majority of other regional flash-flood events described by Kuehnast et al. (1988), and updated through 2006 by the Minnesota State Climatology Office (MNSCO), have similarly short durations (MNSCO 2008). The 102-mm threshold, therefore, may be viewed as having a return period of anywhere from 10 to over 100 yr, depending on the duration of the storm (Table 2).

Many data sources were utilized, some having different numbers of available years of data (Table 3). Locations of qualifying tornadoes (i.e., F 21) within the study area, 1950–2006, were identified using SVRGIS (Smith 2006), which was developed and modified slightly from the archived Storm Prediction Center (SPC) data available in svrplot2 (Hart 1993). The events from 1950 to 1995 were cross-verified with the listings given in Significant Tornadoes, 1680-1991 (Grazulis 1993) and Significant Tornadoes Update, 1992-1995 (Grazulis 1997). In the case of any discrepancies in F ratings, the latter sources were considered the final authority, if only because of the reasoning provided by the author. There were, however, only two changes made to the dataset: the removal of one nonkiller tornado that Grazulis demoted to F 1; and the reinsertion of a brief F 4 that was counted in the original SPC data, was available in svrplot2, was discussed by Grazulis, but was omitted from SVRGIS. Historical significant tornadoes from 1880 to 1949 from Grazulis (1993) were also included if their descriptions indicated they affected the study area. Qualifying wind and hail events (i.e., $33 m s21, $5.1 cm) from 1955 to 2006 were obtained from the updated files in svrplot2 (http://www.spc.noaa.gov/ software/svrplot2). Wind events on 3 July 1983 and

TABLE 3. Data sources used to identify qualifying high-impact convective weather events. Here, AMP is annual maximum daily precipitation. Event type Tornado Tornado Tornado

Wind–hail Rainfall Rainfall Rainfall

Data source

Available record

SVRGIS (Smith 2006) Svrplot (Hart 1993) Significant Tornadoes, 1680-1991, Significant Tornadoes Update, 1992-1995 (Grazulis 1993, 1997) Svrplot TCMA high-density AMP grids (Blumenfeld et al. 2004) MNSCO flash-flood publication (Kuehnast et al. 1988; updated electronically) MN high-density network daily precipitation values (MNSCO online retrieval: http://climate.umn.edu)

1950–2006 1950–2006 1880–1995

Tracks of qualifying tornadoes To validate SVRGIS For historical tornadoes, and to validate tornadoes 1950–95

1955–2006 1970–2002

Approx point locations of wind–hail reports 1958–69 retained but not analyzed

1970–2006

Describes location, coverage of 152-mm (6 in.) rainfall events Search triggered by mention of events in Climate Journal (http://www.climate.umn.edu/ doc/whatsnew.htm)

1970–2006

Remarks

APRIL 2010

BLUMENFELD

21 September 2005 did not originally make it into this study, but were added as extreme events after consulting Storm Data (NOAA/NCDC 1983; NCDC 2005). Daily rainfall data from 1970 to 2002 were edited previously for more than 300 observation points over the TCMA portion of the Minnesota high-density rainfall observer network. The network includes National Weather Service (NWS) cooperative stations, NWS ‘‘backyard’’ observers, and Mosquito Control District gauges, among numerous others (Blumenfeld et al. 2004; information about the network available online at http:// www.climate.umn.edu/doc/hiden_history/appliedmain. htm). Data editing was aided by software that flagged inconsistencies in daily rainfall patterns, allowing the user to identify and remove observations that were missing, that represented multiday accumulations, or that suffered from data entry or transcriptional errors. The editing resulted in a database of all ‘‘acceptable’’ daily rainfall observations over the study area. The data were updated through 2006 by querying the online data retrieval tool at the Minnesota State Climatology Office (http://www. climate.umn.edu/doc/historical.htm). Additional events were identified through the MNSCO’s study of flash floods (Kuehnast et al. 1988), and its updated online flash-flood event inventory.

c. Time series construction An ‘‘event-day’’ methodology, similar to that described by Brooks et al. (2003) and Doswell et al. (2005), was used to identify events. This approach assesses the binary condition of whether a particular event type occurred on a given day within the study area, and ignores the total count and areal extent of reports. For example, four F-3 tornadoes on a given day would be handled identically to a day with just a single F-2 tornado—both would be treated as significant tornado days. Similarly, an event with an isolated extreme wind report will be treated the same as a derecho event associated with numerous extreme-level reports. For each type of event under consideration (e.g., significant wind, extreme hail, etc.), a time series of the number of event days per year was created, with the length of each series determined by the length of the corresponding record (Table 3).

d. Recurrence intervals For temporally and/or spatially continuous data, recurrence intervals for extreme events are often estimated with variants of the generalized extreme value (GEV) distribution (Farago´ and Katz 1990), frequently used for point-based derivation of precipitation design values (Koutsoyiannis 2004). As discussed in detail by Doswell et al. (2005, 2006) and Doswell (2007), however, severe weather data are anything but continuously

623

distributed through space and time. Numerous inhomogeneities related to spatial and temporal variations in severe weather reporting procedures, and the spatial clustering of spotters around urban areas, would lead to significant differences over space in the calculations of recurrence intervals using GEV methods. Recurrence intervals for each event type over the TCMA were estimated empirically from each time series of qualifying event days. Since some years may have multiple qualifying events while other years have none, each time series may be thought of as a modified ‘‘partial duration’’ series (Hershfield 1961). Partial durations are used in extreme rainfall frequency estimation and are obtained by ranking all events and setting the number of events under consideration N equal to the available amount of time in years Ty. Thus, a 50-yr record would examine the top 50 events. The imposition of thresholds for qualifying events in this study means that N and Ty need not be equal; N is set equal to the number of qualifying events, and the recurrence interval, I, is simply Ty/N. Each time series was resampled 10 000 times using a bootstrapping procedure (Wilks 2006), yielding new distributions of recurrence intervals against which the observed recurrence intervals were compared, and from which 95% confidence intervals were obtained. Recurrence intervals were also estimated, based on the number of years with at least one event of interest, because some years had multiple events whereas others had none. In other words, there were ‘‘on’’ years and ‘‘off’’ years, so the estimation of event recurrence based solely on the number of events would not be appropriate, especially for any event types that exhibit episodic or periodic frequency patterns. Using event years, rather than event days, leads to longer recurrence intervals. Since it is unclear which frequency estimation technique is ‘‘best,’’ recurrence intervals are presented as crude ranges, bound by the estimates from the event day and event year methods.

e. Other considerations No attempt was made here to define ‘‘outbreaks,’’ as was done at the national level by Doswell et al. (2006); the term is used in this article only to describe events known by the author, through review of the data and personal experience, to contain multiple, geographically dispersed severe weather reports. Attempting to define and analyze outbreaks at the spatial scale used here would likely result in patterns that merely reflect disparities in severe storm reporting procedures over time, increased presence of storm spotters through time, and the tendency for storm report densities to maximize near populated areas, among other irregularities (Doswell et al. 2005; Doswell 2007). As an example, consider two

624

JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY

VOLUME 49

FIG. 3. The number of days per year in which any qualifying severe weather (tornado, hail, wind) was reported. The raw event day counts are shown as filled circles. The trace shows the 5-yr rolling sum, fit with a 5-yr moving average for smoothing purposes. The vertical axis begins at 21, so that years with counts of zero can be shown.

major wind events. The first, on 3 July 1983, produced nearly continuous damage from the western suburbs of the Twin Cities to north-central Wisconsin and was featured as an ‘‘Outstanding Storm of the Month’’ in Storm Data (NOAA/NCDC 1983). The approximately 350-km-long F-1- and F-2-equivalent damage swath from this event was only described in the text of Storm Data; neither the line entries in Storm Data nor the SPC database listed it as qualifying for this study. Thus, without consulting the text of Storm Data, this event would not have made it into this study. By contrast, a wind event on 19 May 1996 had lower peak winds, a smaller areal extent, and less overall impact than the 1983 event, yet it triggered eight reports that qualify it for this study and would appear to be more significant in a raw frequency tally. Trapp et al. (2006) described similar difficulties in using severe weather report data to evaluate the relative significance of convective wind events. Additionally, no attempt was made to correct for the frequency bias seen in the latter portion of the study period (Fig. 3). Doswell et al. (2006) used a regressionbased technique to remove secular trends in apparent severe weather frequencies at the national level, but no such ‘‘detrending’’ procedure was used here because of the small study domain and the risk of imparting unsupportable assumptions about high-impact severe weather occurrences. It may be suspected, for example, that the active period ending in the mid-1980s had as many severe weather events as the more recent active period, yet there is no way to know this with any certainty. Any inherent ‘‘underreporting’’ in the early part of the record is maintained, and recurrences intervals, therefore, are likely underestimated: assuming a stable climate, qualifying events will be, on average, at least as common as indicated in this study. Last, no attempt is made here to link the apparent decadal variability,

exhibited by the record of severe weather events in Fig. 3, to low-frequency atmospheric circulation patterns. Doswell (2007) has made a compelling case that severe weather data presently are ill suited for such a venture, and likely will be into the future.

3. Results The TCMA experiences some type of significant convective weather event approximately every year, with multiple events in many years (Table 4). The observed record indicates that individual recurrence intervals for this class of events are generally less than 3 yr, with rainfall events being most common, accounting for 67 event days over a 37-yr period. Wind events have the highest frequencies of the severe weather event types. At the extreme levels, severe weather events generally become more common than rainfall events, but this could very well reflect the somewhat arbitrary threshold selection. For event types that have relatively high annual frequencies, the bootstrapped samples connote accordingly high confidence in the recurrence intervals; short recurrence interval estimates tend to have the small associated potential error. Thus, for significant events, the 95% confidence limits are generally less than one year from the observed recurrence estimates, with the sole exception being for ‘‘combined’’ events—which require that more than one type of significant event was reported (Fig. 4). Extreme events, being far less common than significant events, have relatively long recurrence intervals and large associated potential error (Fig. 5). Infrequent extreme events, such as rainfall, tornadoes, and combined events, have a large degree of uncertainty associated with their estimated recurrence intervals. This uncertainty should be considered as individual event types are discussed.

APRIL 2010

625

BLUMENFELD

TABLE 4. Frequency of high-impact event types within the TCMA. Recurrence intervals are expressed as ranges: the low end of the range indicates the mean recurrence interval-based event days, whereas the high end of the range indicates recurrence interval based on event years. Note that for ‘‘any high impact,’’ ‘‘any severe,’’ and the two ‘‘combined’’ categories, calculations are based on the shortest of the available records. See Table 1 for threshold definitions. Significant events

Event type

Years

Frequency (days)

Any high-impact Any severe Tornado Wind Hail Combined severe Rainfall Combined severe and rainfall

37 (1970–2006) 52 (1955–2006) 127 (1880–2006) 52 (1955–2006) 52 (1955–2006) 52 (1955–2006) 37 (1970–2006) 37 (1970–2006)

127 84 49 44 35 10 67 13

Extreme events

Frequency (yr)

Mean recurrence interval (yr)

36 39 47 22 25 9 29 11

0.3–1.0 0.6–1.3 2.6–2.7 1.2–2.4 1.5–2.1 5.2–5.8 0.6–1.3 2.8–3.4

a. Tornadoes Tornadoes producing F-2 or greater damage and/or at least one fatality recur within the area on average every 2–3 yr. Extreme or ‘‘violent’’ (F 4–5) tornadoes account for over 20% of all high-impact tornado events and recur slightly less than once per decade, on average. Unlike other event types, qualifying tornadoes generally do not tend to affect the area more than once per year; of the 47 yr with qualifying tornado events, only 2 yr featured more than one such tornado day. Tornadoes also exhibit periodic behavior, with multidecadal, interchanging

Frequency (days)

Frequency (yr)

Mean recurrence interval (yr)

18 19 10 13 8 4 5 5

14 15 10 11 7 4 4 5

2.1–2.6 2.7–3.5 12.7 4.0–4.7 6.5–7.4 13 7.4–12.3 9.3

regimes of high and low activity (Fig. 6). Consideration of potential associations with modes of atmospheric circulation is beyond the scope of this paper. While the separation of extreme events from the other qualifying events does help focus the research, some within-class nuances suggest careful interpretation is required. For example, the most recent F-4 tornado occurred on 3 July 1983, embedded within the significant wind event discussed earlier. The tornado’s pathlength was less than 2 km and its width was less than 75 m— both of which are extremely low for F-4 tornadoes (Brooks 2004). Although it did severely damage a small

FIG. 4. Recurrence intervals calculated from each time series of ‘‘significant’’ event days, plotted with 95% confidence intervals obtained from bootstrapped distributions. See section 2a and Table 1 for event type definitions and classifications.

626

JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY

VOLUME 49

FIG. 5. As in Fig. 4, but for ‘‘extreme’’ events. Note difference in vertical scale.

section of town, the tornado’s overall impact was limited by its fleeting existence (NOAA/NCDC 1983; Grazulis 1993). By contrast, a long-track tornado family on 23 June 1952 traveled nearly 250 km, passing from largely rural areas of southern Minnesota into the southern and central TCMA, producing several segments of sustained F-2 damage (Grazulis 1993). While the actual length is unknown, we do know the last third or so of the storm’s life was over an area that is now thoroughly developed and heavily populated, though it was not at the time. Even if the tornado was only capable of F-2 damage, which we cannot know but may have reason to doubt (see Brooks 2004), it still would produce considerable damage over the TCMA portion of its path. Whether it did constitute or would constitute a higher-impact event than a very brief F-4 is a matter of speculation. Three substantial tornado events stand out as probable worst-case scenarios, which would likely prove catastrophic over the modern-day TCMA. The first, on

12 June 1899 in New Richmond, Wisconsin, is noted as one of the deadliest tornadoes in U.S. history, and is the only estimated F-5 in this study (Grazulis 1993). The tornado occurred when a traveling circus was performing before hundreds of out-of-town visitors; the official death toll of 117 includes only those who could be identified, and thus, was believed by many to be substantially underestimated (Boehm 1900). The second event, on 5 April 1929, produced two, long-track, estimated F-4 tornadoes in the central, northern, and eastern portions of the study area, among others in Wisconsin (Grazulis 1993). The tornadoes affected portions of the TCMA that were decidedly rural at the time but are now thoroughly developed and heavily populated (Fig. 7), and thus, a similar event over the modern TCMA would likely result in more destruction. The third event, on 6 May 1965, produced six significant and extreme tornadoes (four F-4 tornadoes, plus one F-2 and one F-3 tornado) over the TCMA in a 3-h period, resulting in

FIG. 6. As in Fig. 3, but for tornadoes, and with 10-yr rolling sums (instead of 5 yr).

APRIL 2010

BLUMENFELD

627

FIG. 7. (left) Damage to Columbia Heights, MN, on 5 Apr 1929 (courtesy of Minnesota Historical Society), with (right) satellite view of same area as of 2007 (courtesy of Google Maps).

13 deaths (Fig. 8). In many ways, this outbreak could help ‘‘calibrate’’ a modern event: the tornadoes all had positive warning lead times, which, combined with ‘‘wall-towall’’ media coverage and the TCMA’s first use of civil defense sirens for tornadoes, drastically reduced the element of surprise common with outbreaks of the time (Grazulis 1993; NWS 2008; Seeley 2006). Additionally, two of the tornadoes struck heavily populated areas near Minneapolis. Thus, although many scenarios could worsen the outcome—for instance, tornadoes crossing the area during rush hour—a modern violent tornado event would be similar to this one in many regards.

b. Nontornadic wind events Convective wind events capable of significant damage are the most prevalent of all severe weather types in this

study, recurring on average about every 1–3 yr. Again, however, owing to detection problems early in the record, they may be more frequent than indicated here. Over the 52-yr study period 42 days had qualifying wind events, but these were distributed over just 22 yr. Thus, whereas most qualifying tornado years have just one event day, wind events tend to be clustered in time. For instance, 1980 had five significant wind event days, and 1998 had six. Extreme wind events recur within the area approximately every 4–5 yr, and usually come just once per qualifying year. Unlike tornadoes, the TCMA has experienced major wind events recently, and it is therefore likely that the risk from such storms is well understood by the public. The 30 May 1998 derecho attracted a lot of local media attention, as parts of St. Paul were without

FIG. 8. Tracks of 6 May 1965 tornadoes. First touchdowns were the middle two tornadoes. Tracks were digitized by author from hand-drawn maps in Kuehnast et al. (1975).

628

JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY

power for a week (NOAA/NCDC 1998). A large supercell on 21 September 2005 caused extensive downburst winds that resulted in thousands of downed trees and one fatality in Minneapolis, overshadowing the F-2 tornado that touched down, briefly, to the north of the wind swath (NCDC 2005).

c. Hail The hail frequency characteristics fall between those of winds and tornadoes: wind recurs more frequently, tornadoes less frequently. Moreover, hail does exhibit some temporal clustering, but not as much as wind events, and more than tornado events. For instance, the wind event days to event years ratio was 2:1, hail was 1.4:1, and tornadoes were approximately 1:1. The TCMA experiences significant hail events approximately every 1–2 yr, and extreme events recur approximately every 7 yr. Distinguishing between significant and extreme events may not be as useful for hail as for tornado and wind events. The most damaging hail event in the TCMA to date was 15 May 1998. Here, the hail was barely large enough to qualify for this study, but was driven by 25–35 m s21 (50–65 kt) winds and resulted in nearly $1 billion across the TCMA (NOAA/NCDC 1998). Thus, hail size alone may not be a good gauge of ‘‘significance,’’ since the conditions in which it falls can determine its damage potential.

d. Heavy rainfall Heavy rainfall is most common of the high-impact convective weather types, and is the only one to have more events than available years. Rainfall events tend to be clustered in time, even more so than wind events. Thus, the 64 qualifying event days occurred during 29 of the 37 available years. In this case, detection early in the period is not a problem, as the high-density network observer numbers have remained steady since 1970. Extreme rainfall is somewhat rare, and recurs less frequently than both extreme wind and extreme hail events. Of course, the selection of the 203-mm threshold could influence the perceived rareness of these events. If instead the regional 100-yr, 24-h value of 152 mm had been designated as the threshold, 18 events would have been counted as extreme. The higher threshold was selected because every instance of it being met or exceeded resulted in severe damages, including infrastructural collapse.

e. Combined events Combined-hazard events appear to be somewhat rare, and judging the significance of their combined effects is

VOLUME 49

difficult. In general, it appears the combination of event types is typically not as important as many single-type events, and often combined events are still dominated by one particular event type. Only two events—1 July 1997 and 21 September 2005—had qualifying reports from each of the three severe weather categories. With the first event, the hail was relatively minor in comparison with the F-2–3 tornadoes, which were less significant (in terms of impact) than the widespread extreme wind event across the northern TCMA. The 2005 event also seemed to be largely wind focused, though 15% of the estimated $200 million in damage was attributable to hail (NCDC 2005). In both cases, the highest-magnitude reports from all of the severe weather types were within the same storm-sized area, generally comprising less than 2000 km2. In addition, only the 3 July 1983 event had more than one extreme report type. One might think of that day’s brief F-4 tornado as merely punctuating an extremely energetic and damaging straight-line wind event. Considering that the regional pilot study indicated that the probabilities of a randomly selected tornado, wind, or hail report qualifying as extreme by this study’s definition were 0.015, 0.013, or 0.014, respectively, it should not be surprising that such events do not occur together with much regularity. Qualifying heavy rainfall events that combined with one or more severe weather types were also somewhat rare, with only 13 such dates (out of 37 yr), and only two qualifying severe weather events associated with extremelevel rainfall. In those cases, the intense flooding had a far greater impact than the severe weather event. Thus, it appears that most high-end severe weather events (85%) do not pose serious flooding threats, and most high-end convective rainfall events (80%) do not pose serious severe weather threats. These statements bear careful interpretation, however, because short-term flooding events associated with severe convection could prove catastrophic, even if the 24-h rainfall totals do not satisfy either of the rainfall criteria used here. The daily data used for rainfall in this study do not permit calculation of short-duration rainfall rates, though it is quite likely that many of the storms analyzed here exceeded short-term design capacity. Smith et al. (2001) demonstrated the short-term flood danger posed by the extreme rainfall rates associated with supercell thunderstorms. One historical event bears special consideration. The hard-to-define storm on 20 August 1904 was reported initially as a tornado, owing almost certainly to the severe and extensive damage it wrought. The storm claimed 16 lives and damaged the downtowns of both Minneapolis and St. Paul, as well as many neighboring areas. The large areal coverage may, in fact, be what led

APRIL 2010

629

BLUMENFELD

TABLE 5. Comparisons of frequencies of high-impact severe weather events for TCMA between NSSL maps, study results for same NSSL base period, and study results for entire available period. Expected annual frequencies were estimated from NSSL maps, and then multiplied by 1.5 to account for the 50% size increase of the TCMA over the NSSL maps’ grid cells. See Brooks et al. (2003) and Doswell et al. (2005) for discussion of mapping techniques.

Qualifying event type

Expected annual frequency (NSSL base period)

Actual frequency using NSSL base period

Total frequency (study period)

Tornado Extreme tornado Wind Hail

0.15–0.23 (1921–95) 0.02–0.03 (1921–95) 1.1–1.5 (1980–94) 1.2 (1980–94)

0.45 0.09 1.0 0.73

0.39 (1890–2006) 0.08 (1890–2006) 0.85 (1955–2006) 0.67 (1955–2006)

Grazulis (1993) to doubt that the entire event was tornadic. Other evidence suggests the storm was primarily nontornadic. For instance, the New York Times (21 August 1904; 22 August 1904) reported that the tornado winds lasted 15 min (at any one place), and also that the storm took an improbable track, northeastward to downtown St. Paul, before ‘‘[sweeping] onward to Minneapolis,’’ which lies well west of downtown St. Paul. If the storm was not tornadic, then it would appear to stand out from the other wind events in the extreme class: it had a 1-min average of 49 m s21 (96 kt), and maximum gust of 81 m s21 (157 kt), before ripping the recording equipment from its mountings atop a newspaper building in downtown St. Paul (Seeley 2006). Knowing the true nature of this storm is difficult, but it was either yet another violent tornado, or a singularity among nontornadic wind events.

4. Discussion and concluding remarks This investigation has examined the recurrence intervals of convective weather events most likely to cause damage, casualties, and disruption to the Twin Cities Metropolitan Area. On average, the area experiences 1–3 of these events per year. About half of the threat comes from rainfall capable of flash flooding, with nontornadic convective wind events being the most common of the severe weather event types. Each mode of highimpact convective weather (tornadoes, winds, hail, heavy rainfall) has at some point in the TCMA’s history caused extensive damage, and even death. Indeed, extreme convective weather systems, virtually guaranteed to cause significant damage and disruption, recur somewhere within the TCMA approximately every 2–3 yr, with wind being the most common of these extreme events. The severe weather frequencies presented here differ from those of Brooks et al. (2003), Doswell et al. (2005), and supplemental maps (see http://www.nssl.noaa.gov/

hazard; hereinafter NSSL maps),2 despite employing the same or similar baseline thresholds for some categories (e.g., F-41 tornadoes, 65-kt winds, and 2-in. hail). Appropriate event frequencies from the NSSL maps can be estimated over the TCMA, converted to annual frequencies, and multiplied by 1.5 to account for the 50% size increase in the TCMA over the NSSL maps’ analysis units. The annual frequencies of qualifying tornado and extreme tornado events are generally much higher in this investigation (by factors of 2–4), while the NSSL maps’ hail and wind frequencies are up to 80% higher; these results are obtained even when identical reference periods are used (Table 5). Some of the difference between the frequencies here and in the NSSL maps may be attributable to different methodologies, and thus, direct comparison may not be appropriate. Frequencies here are presented as is, and are based entirely upon the number of event days within the TCMA over the study period. Near misses are not considered. Brooks et al. (2003) and Doswell et al. (2005), however, employed Gaussian kernel density estimation, which allows gridcell event frequencies to be influenced by neighboring cells, so that resulting maps reflect the smoothness of the assumed underlying distributions. In that way, near misses—when an event does not strike the target grid cell but does strike a neighboring one— would be important and would influence the frequencies and probabilities within the TCMA. In the severe wind and hail cases, the presence of higher-frequency values to the south and southwest might enhance local probabilities. For significant tornadoes, this study included all killer tornadoes, as well as F-2 and greater tornadoes, consistent with definitions used by Grazulis (1993), 2 The NSSL wind and hail maps are visually identical to those in Doswell et al. (2005). The legends, however, are different, with the NSSL stating that the maps show day yr21, and Doswell et al. (2005) stating they show percent annual probability. The former is assumed here to be correct (they imply yearly recurrence) and the latter in error (they imply centennial recurrence).

630

JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY

whereas Brooks et al. (2003) used F 2 or greater only. Thus, tornado frequencies should be somewhat higher here. Discrepancies in F-4–5 tornado frequencies between this study and the NSSL maps likely result from events being included here that were not included in the earlier studies. This study, for example, analyzed tornado paths—any path that affected the study area was included—whereas Brooks et al. (2003) analyzed tornado ‘‘touchdown points.’’ Thus, methodological differences likely account for many of the differences in reported frequencies. As explained in sections 1 and 2, one goal of this research was to give local emergency managers a frequency analysis of high-impact convective weather based on actual events within the area now known as the TCMA. The severe weather frequencies (and recurrence intervals) presented—especially for hail and wind—are likely artificially too low (and too long), owing to detection problems early in the study period. Though detection and observation of high-impact events should continue to improve, meaningful differences in reporting and handling procedures are likely to continue into the foreseeable future (Doswell 2007). Thus, it is unclear whether the addition of time will clarify local severe weather frequencies. Tornado frequencies present a challenge in the TCMA, because it has been 25 yr (as of this writing) since the last violent single tornado in the area, and over 40 yr since the last multiple-tornado event. Hazard mitigation research has shown that awareness, concern, and action related to natural hazards all decay as the time since the last event increases (Prater and Lindell 2000). Considering the large population growth since the 1965 outbreak, and that most residents have not experienced a serious tornado outbreak in the region, communicating the risk to the public might require an extensive outreach campaign. For emergency preparedness officials, and certainly for members of the media and the public, understanding the qualitative points may be more important than trying to quantify them: thunderstorms capable of serious damage and disruption strike the TCMA regularly. Major damage from a convective weather episode is ‘‘normal,’’ and tornadoes—including long-lasting and violent ones—are part of the area’s history, and should be expected to be part of its future. As many hazards scholars have argued, identifying dangerous natural phenomena and understanding their statistical properties do not make us safe from them (Tobin and Montz 1997). Furthermore, risk is not just geophysical in origin, but is also social, cultural, and political (e.g., Pelling 1999). Thus, this work is really just one piece in a larger puzzle. If the goal is to protect the

VOLUME 49

people of the TCMA from convective weather hazards, then understanding the convective weather is one issue; understanding the people we wish to protect is another. The next major tornado outbreak in the TCMA is going to affect a much larger and more diverse population than that of 1965. Do all groups have equal access to warning information? Can the disabled and elderly be notified and moved to safety quickly? Are people responding to warnings based on their beliefs about the unlikelihood of tornadoes hitting urban areas? Do people know what to do when stuck in traffic and severe weather is approaching? Ultimately, it is the answers to these questions—and not what we know about our climatology—that will determine how the TCMA fares during its next high-impact convective weather event. Acknowledgments. I thank Dr. Richard Skaggs for his advice and guidance throughout the preparation of this manuscript, and for his keen insight into statistical techniques. I also thank Drs. Charles Doswell and Harold Brooks, both for their comments regarding interpretation of discrepancies between the frequency estimates in this paper and their earlier work, and also for their helpful comments on an earlier version of this article. I extend sincere appreciation to Dr. Walker Ashley and two anonymous reviewers, whose thoughtful analyses and helpful feedback dramatically improved this manuscript. REFERENCES Adams, J. S., 2006: A new Twin Cities: The shape of things to come. CURA Reporter, Winter 2006, 10–19. Ashley, W. S., 2007: Spatial and temporal analysis of tornado fatalities in the United States: 1880–2005. Wea. Forecasting, 22, 1214–1228. ——, and T. L. Mote, 2005: Derecho hazards in the United States. Bull. Amer. Meteor. Soc., 86, 1577–1592. Blumenfeld, K. A., 2008: Comments on ‘‘Low-level winds in tornadoes and potential catastrophic tornado impacts in urban areas.’’ Bull. Amer. Meteor. Soc., 89, 1578–1579. ——, R. H. Skaggs, and J. A. Zandlo, 2004: Using a dense precipitation gage network to estimate annual maximum daily precipitation. Preprints, 14th Conf. on Applied Climatology, Seattle, WA, Amer. Meteor. Soc., P4.4. [Available online at http://ams.confex.com/ams/pdfpapers/67883.pdf.] Boehm, A. G., 1900: History of the New Richmond Cyclone. Dispatch Job Printing Co., 217 pp. Brooks, H. E., 2004: On the relationship of tornado path length and width to intensity. Wea. Forecasting, 19, 310–319. ——, C. A. Doswell III, and M. P. Kay, 2003: Climatological estimates of local daily tornado probability for the United States. Wea. Forecasting, 18, 626–640. ——, ——, and D. Sutter, 2008: Comments on ‘‘Low-level winds in tornadoes and potential catastrophic tornado impacts in urban areas.’’ Bull. Amer. Meteor. Soc., 89, 87–90. Changnon, S. A., 2001a: Damaging thunderstorm activity in the United States. Bull. Amer. Meteor. Soc., 82, 597–608.

APRIL 2010

BLUMENFELD

——, 2001b: Thunderstorm rainfall in the conterminous United States. Bull. Amer. Meteor. Soc., 82, 1925–1940. ——, and J. Burroughs, 2003: The tristate hailstorm: The most costly on record. Mon. Wea. Rev., 131, 1734–1739. Coniglio, M. C., and D. J. Stensrud, 2004: Interpreting the climatology of derechos. Wea. Forecasting, 19, 595–605. Doswell, C. A., III, 2007: Small sample size and data quality issues illustrated using tornado occurrence data. Electron. J. Severe Storms Meteor., 2 (5), 1–16. ——, H. E. Brooks, and M. P. Kay, 2005: Climatological estimates of daily nontornadic severe thunderstorm probability for the United States. Wea. Forecasting, 20, 577–595. ——, R. Edwards, R. L. Thompson, J. A. Hart, and K. C. Crosbie, 2006: A simple and flexible method for ranking severe weather events. Wea. Forecasting, 21, 939–951. Farago´, T., and R. Katz, 1990: Extremes and design values in climatology. WMO Tech. Doc. 386. Grazulis, T. P., 1993: Significant Tornadoes, 1680-1991. Environmental Films, 1326 pp. ——, 1997: Significant Tornadoes Update, 1992-1995. Environmental Films, 117 pp. Hall, S. G., and W. S. Ashley, 2008: The effects of urban sprawl on the vulnerability to a significant tornado impact in northeastern Illinois. Nat. Hazards Rev., 9, 209–219. Hart, J. A., 1993: SVRPLOT: A new method of accessing and manipulating the NSSFC severe weather data base. Preprints, 17th Conf. on Severe Local Storms, St. Louis, MO, Amer. Meteor. Soc., 40–41. Hershfield, D. M., 1961: Rainfall frequency atlas of the United States. U.S. Weather Bureau Tech. Paper 40, 61 pp. Huff, F. A., and J. Angel, 1992: Rainfall frequency atlas of the Midwest. Illinois State Water Survey Bull. 71, 141 pp. Johns, R. H., and W. D. Hirt, 1987: Derechos: Widespread convectively induced windstorms. Wea. Forecasting, 2, 32–49. Koutsoyiannis, D., 2004: Statistics of extremes and estimation of extreme rainfall: I. Theoretical investigation. Hydrol. Sci. J., 49, 575–590. Kuehnast, E. L., D. G. Baker, and J. W. Enz, 1975: Climate of Minnesota, Part VIII—Precipitation Patterns in the MinneapolisSt. Paul Metropolitan Area and Surrounding Counties. University of Minnesota, 37 pp. ——, ——, and J. A. Zandlo, 1988: Sixteen Year Study of Minnesota Flash Floods. Minnesota Dept. of Natural Resources, 72 pp. McDonald, J. R., and Coauthors, 2004: A Recommendation for an Enhanced Fujita Scale (EF-Scale). Texas Tech University, 95 pp. MNSCO, cited 2008: Minnesota flash floods: 1970-2008. [Available online at http://climate.umn.edu/doc/flashflood.htm.] Moller, A. R., 2001: Severe local storms forecasting. Severe Convective Storms, Meteor. Monogr., No. 50, Amer. Meteor. Soc., 433–480. NCDC, 2005: Storm Data. Vol. 47, No. 9, 208 pp. ——, 2006: Storm Data. Vol. 48, No. 8, 384 pp. NOAA/NCDC, 1983: Storm Data. Vol. 25, No. 7, 64 pp. ——, 1998: Storm Data. Vol. 40, No. 5, 432 pp.

631

NWS, cited 2008: The May 6 1965 tornadoes. [Available online at http://www.crh.noaa.gov/mpx/HistoricalEvents/1965May06/ index.php.] Oberts, G. L., 1984: Surface water management: Precipitation frequency analysis. Metropolitan Council Publication 10-84-007. Pelling, M., 1999: The political ecology of flood hazard in urban Guyana. Geoforum, 30, 249–261. Prater, C. S., and M. K. Lindell, 2000: Politics of hazard mitigation. Nat. Hazards Rev., 1, 73–82. Rae, S., and J. Stefkovich, 2000: The tornado damage risk assessment: Predicting the impact of a big outbreak in Dallas-Fort Worth, Texas. Preprints, 20th Conf. on Severe Local Storms, Orlando FL, Amer. Meteor. Soc., 295–296. Schwartz, B. E., C. F. Chappell, W. E. Togstad, and X.-P. Zhong, 1990: The Minneapolis flash flood: Meteorological analysis and operational response. Wea. Forecasting, 5, 3–21. Seeley, M. W., 2006: Minnesota Weather Almanac. Minnesota Historical Society Press, 208 pp. Skaggs, R. H., 1998: Intensity of extreme rainfall over Minnesota. Minnesota Department of Transportation Research Rep. 199809U, 33 pp. Smith, B. T., 2006: SVRGIS: Geographic Information System (GIS) graphical database of tornado, large hail, and damaging wind reports in the United States (1950-2005). Preprints, 23rd Conf. on Severe Local Storms, St. Louis, MO, Amer. Meteor. Soc., P2.6. [Available online at http://ams.confex.com/ams/ pdfpapers/115381.pdf.] Smith, J. A., M. L. Baeck, Y. Zhang, and C. A. Doswell III, 2001: Extreme rainfall and flooding from supercell thunderstorms. J. Hydrometeor., 2, 469–489. Tobin, G. A., and B. E. Montz, 1997: Natural Hazards: Explanation and Integration. Guildford Press, 388 pp. Togstad, W., cited 2008: Origins and synoptic setting for the 30–31 May 1998 Minnesota to New York derecho. [Available online at http://www.crh.noaa.gov/mpx/?n5derechoevent.] Trapp, R. J., D. M. Wheatley, N. T. Atkins, R. W. Przybylinski, and R. Wolf, 2006: Buyer beware: Some words of caution on the use of severe wind reports in postevent assessment and research. Wea. Forecasting, 21, 408–415. U.S. Census Bureau, cited 2008: Table 2: Population estimates for the 100 most populous metropolitan statistical areas based on July 1, 2006 population estimates: April 1, 2000 to July 1, 2006. [Available online at http://www.census.gov/Press-Release/ www/releases/archives/cb07-51tbl2.pdf.] Wilks, D., 2006: Statistical Methods in the Atmospheric Sciences. 2nd ed. Academic Press, 627 pp. Winkler, J. A., 1988: Climatological characteristics of summertime extreme rainstorms in Minnesota. Ann. Assoc. Amer. Geogr., 78, 57–73. Wurman, J., C. Alexander, P. Robinson, and Y. Richardson, 2007: Low-level winds in tornadoes and potential catastrophic tornado impacts in urban areas. Bull. Amer. Meteor. Soc., 88, 31–46. ——, ——, ——, and ——, 2008: Reply. Bull. Amer. Meteor. Soc., 89, 90–94.