1. Introduction

The Bow Echo and Mesoscale Convective Vortex Experiment (BAMEX) is a research study of the life cycles of mesoscale convective systems (Davis et al. 2004). The field phase of BAMEX was conducted from 20 May 2003 through 6 July 2003 over a large portion of the central United States; the base of operations was the MidAmerica Airport in Mascoutah, Illinois (just east of St. Louis, Missouri).

One of the BAMEX goals was to gather an enhanced “ground truth” or verification dataset for proposed and established mechanisms of severe winds in quasi-linear mesoscale convective systems (QLCSs) (e.g., Trapp and Weisman 2003; Weisman 2001; Fujita 1981). Toward this end, aerial and ground surveys of wind damage were conducted (a brief description of each BAMEX event surveyed by our teams, as well as photo documentation of these events, can be found online at http://apollo.lsc.vsc.edu/bamex/index.html) immediately following the occurrence of presumed significant QLCSs.

The location and scope of the surveys were guided initially by National Weather Service (NWS) Local Storm Reports (LSRs) and by weather radar [Weather Surveillance Radar-1988 Doppler (WSR-88D) and airborne Doppler] data. Our working assumption throughout the project was that the significance of a given event would be a function of the storm's appearance on radar, as revealed in fields of radar reflectivity and radial velocity, and also of the number of (and wind speed/damage described within) associated severe wind reports; significance was similarly assumed to be function of finalized reports in the National Oceanic and Atmospheric Administration's (NOAA) Storm Data publication. As we determined at the conclusion of the project and demonstrate below, these were often poor assumptions during BAMEX, even when considered with otherwise intense-appearing bow echoes, such as ones analyzed by Jorgensen et al. (2005).

A report-based assessment of damaging nontornadic wind events can be derived from the derecho criteria of Johns and Hirt (1987). It is important to note, however, that these criteria (e.g., concentrated area of nonrandomly distributed reports of wind damage and/or gusts >26 m s⁻¹) apply strictly to a widespread event, characterized as a “family of downburst clusters,” with a corresponding damage length scale (major axis) ≥400 km (Fujita and Wakimoto 1981). Other damaging wind patterns associated with QLCSs observed during BAMEX include a “downburst” and “downburst cluster,” with 4–40- and 40–400-km major axes of damage, respectively (Fujita and Wakimoto 1981). Illustrations of BAMEX events at these relatively more localized scales are given in sections 2 and 3.

The objective of this paper is to show how the use of severe wind reports in Storm Data (and initially in the LSRs) to assess the scope and severity of events at such scales can be problematic. Explanations and further discussion of this problem are provided in section 4, as are some of the implications, which may include a skewed understanding of how and when systems of thunderstorms cause damage. Some recommendations pertaining to severe wind reporting are offered in section 5.

2. Two examples

In the following two examples from BAMEX, we briefly compare our detailed survey analyses with the corresponding Storm Data severe wind reports. It is not our intent to criticize the warning and postevent operations of any forecast office or county emergency management office, but rather to illustrate the shortcomings of the reports. Additional details of these events can be found in Wheatley et al. (2006, hereafter WTA); discussions of damage survey procedures and limitations, and also of data analyses techniques, can be found in Atkins et al. (2005) and WTA.

On 10 June 2003, an intense cell bow echo (Lee et al. 1992; Klimowski et al. 2004) in the vicinity of Shelby, Nebraska (east-central Nebraska), caused a 30-km-long swath of concentrated¹ wind damage to trees, pivot irrigation systems, and some farm buildings, based on our ground and aerial surveys that were conducted 10–12 June 2003; we have a high degree of confidence that all areas affected by this bow echo were surveyed. As discussed in WTA, the Shelby bow echo was one of three that occurred in this region of the United States on 10 June 2003. We rated most of its damage F0 (Fujita 1981), though an embedded area of F1 damage was also found (Fig. 1).

In the LSRs and Storm Data, this event was represented by a mere three reports² located near or at the eastern edge of the damage swath (Fig. 1). The LSRs only mentioned “trees down north of Shelby,” although these are preliminary and often incomplete reports. Nevertheless, one could easily conclude, incorrectly, that this was a fairly isolated, insignificant event. In fact, only after noting the property damage estimate ($1.0 M) and narrative of one of the Storm Data entries (see NCDC 2003a) could one begin to appreciate the scope of the damage:

Elsewhere, vicious high winds over 70 mph did extensive damage north of Shelby in eastern Polk County. Several farmsteads sustained house, outbuilding, grain bin, and tree damage. In one case, a garage was nearly ripped from the house it was attached to. Twenty four center pivot irrigation systems were severely damaged or destroyed.

A similar case of widespread damage, particularly to crops, with only one or two corresponding Storm Data wind (and hail) reports has been shown by Bentley et al. (2002).

In contrast, there were instances during BAMEX in which large numbers of wind reports were listed in Storm Data (and initially in the LSRs) for events that we deemed relatively less significant. One such example is shown in Fig. 2, representing an extensive, squall-line bow echo (Klimowski et al. 2004) that moved through Indiana, Ohio, and Kentucky during the afternoon and evening hours of 4 July 2003 (see WTA). A postevent aerial survey–which facilitates the most comprehensive postevent assessment of the damaging surface winds–was not conducted because of the lack of aircraft resources, and so it is conceivable that unreported damage areas were missed. However, all reported damage was investigated and formed the basis for the ground surveys. Conducted over roughly the week following the event, this effort revealed only a few scattered areas of concentrated F0 damage (see also Fujita 1981), within which some F1 damage was identified.

Storm Data wind reports associated with this bow echo were numerous (55) and widespread on this day, yet relevant property damage amounted to $0.3 M (NCDC 2003b). The narratives of many of these reports (NCDC 2003b) were less informative, for example, “Trees were downed.”

Based on our experience, this could imply a few bent-over samplings, a large grove of snapped hardwood trees with ∼0.5 m diameters, or something in between. Reports on this day of severe wind gusts in absence of damage, such as “A trained spotter reported an estimated wind gust to 60 mph,” were problematic in terms of assessment (see section 4) and, obviously, could not be verified by the damage surveys.

3. Quantification

To summarize, the severe wind reports for these cases from BAMEX served as a poor characterization of the actual scope and magnitude of the surveyed damage. We have attempted to partially quantify this by comparing the number of Storm Data wind reports N_SD with the total area of our surveyed F0 damage A_F0. For consistency with our surveys, which per the mission of BAMEX focused on damage associated with bow echoes, N_SD included only those reports associated with convective storms that were well organized as a QLCS. The value of A_F0 was estimated graphically after digitizing the F0 contours, which enclosed areas of concentrated damage. The ratio N_SD/A_F0 is 3/118 km² = 0.025 km⁻² for the first case; hence, the report density appears to be low. As could be anticipated using Figs. 1 and 2, the ratio is much larger for the second case: 55/394 km² = 0.140 km⁻², again, suggesting by virtue of this higher report density a relatively more significant event. An evaluation of other select BAMEX cases is presented in Table 1.

Although it is unclear what the optimal number of reports should be for a given event, we find that in these BAMEX events, the report density tends to misrepresent the scope of the actual damage. This is not too surprising, in light of the NWS requirement for severe thunderstorm warning verification: one severe wind report per every 16.1 km (10 mi) and 15 min encompassed by a severe thunderstorm warning in a given county³ (e.g., Polger et al. 1994; see also NWS 2005). Hence, for the purpose of verification only, there is little impetus for individual forecast offices to seek additional reports, since only one of possible multiple severe wind reports in the same county and within the 10-mi or 15-min separations can be used to verify a warning. Of course, such multiple reports are still permitted in Storm Data and, in fact, reports are required for the NWS verification database, and hence will appear in Storm Data, in the following circumstances: any event that causes death or injury; any event that causes crop or property damage in excess of $500 000; any report of winds ≥65 kt (33 m s⁻¹; which have caveats of their own; see below); and any hail size report ≥2 in. (5.1 cm; NWS 2005). Nonetheless, for the BAMEX events listed in Table 1, the amount of reporting appears to be related more to time of day of the event (i.e., fewer reports for overnight events) than to property damage.

To help begin to determine how damage intensity is reflected in the Storm Data reports, we computed the ratio N_SD1/N_SD, where N_SD1 is the number of reports of F1 or greater wind speeds (gusts ≥ 63 kt), and compared it with the ratio A_F1/A_F0, where A_F1 is the surveyed F1 damage area. Table 1 suggests a mixed correspondence between the reported gust magnitudes and the intensity of the surveyed damage. This finding is also not too surprising, in light of the NWS-mandated requirement that (severe) wind damage be accompanied by a wind gust in the record. We argue that assignment of a single, peak wind speed for damage to trees, nonengineered buildings, etc. is essentially arbitrary and fraught with potential errors. In the absence of damage, reported wind gusts are estimated values, unless otherwise indicated. This is equally problematic since estimation of wind speed by a human observer is inherently difficult, and tends to be overdone (Doswell et al. 2005).

4. Discussion

We are certainly not the first to raise concerns about reports of severe convective winds. Bentley et al. (2002) questioned the usefulness of Storm Data “damage estimates obtained after a significant weather event.” Johns and Evans (2000) had similar concerns. Weiss and Vescio (1998), and more recently Weiss et al. (2002), have noted that the annual number of severe wind reports has increased substantially over the past 20 yr. Anomalous spikes in the numbers of reported severe wind gusts of 50 kt (26 m s⁻¹), the wind threshold for a severe thunderstorm, have also been noted (Weiss et al. 2002). Besides improved public education/awareness and an increase in population density, explanations for these changes can be attributed to the deployment of the WSR-88D network, enhancement of storm-spotter groups, implementation of a national warning verification program, and of course the NWS policy requiring a wind gust in the severe report (Weiss et al. 2002).

Arguably, the intended purpose of severe wind (and severe hail) reports is to provide a means to verify NWS severe thunderstorm warnings. We will defer the debate of the current verification system [and of suggested future improvements, such as use of polygon warnings instead of county warnings; see Waters et al. (2005)] to another forum. Our debate here is on the use of wind reports in Storm Data as research quality data.

We do not wish to admonish users of these data. For example, in climatological studies of severe convective wind events, and otherwise for models of convective wind hazards, there are few alternatives (see section 5) to the Storm Data reports if a reasonable sample size is desired. It should be clear, however, that such studies will be prone to the misrepresentation identified above.

What is the implication of greatest concern to us? Wind report uncertainties and misrepresentation can potentially act to skew the basic understanding of the capacity of QLCS types to produce damage. In the preceding examples, the large bow echo (as viewed by a weather radar) would likely be perceived (incorrectly) as more intense than the small, cell bow echo. A hypothetical collection of equivalent large bow echo events could provide a composite synoptic and mesoscale environment that then would be treated (again, incorrectly) as a higher threat than would the environment of a collection of equivalent small bow echo events; an analogous statement can be made about the respective characteristics of these events as observed by Doppler radar, and the treatment of these characteristics in the context of warning operations. As an important aside, such small bow echoes, and in fact all but perhaps one of the events in Table 1, would be excluded from derecho databases, owing to damage track lengths that fall below the typical 400-km threshold (e.g., Coniglio and Stensrud 2004). It is reiterated here that derechos and severe QLCSs are not mutually inclusive.

5. Recommendations

In an ideal world with unlimited resources, we would recommend that a postevent damage survey be conducted for all convective wind events associated with a preliminary report of severe wind damage. This would ensure an accurate assessment of the significance of each event, and in turn ensure accurate historical records, climatologies, hazard models, etc. A case for a similar action for tornadoes, hail, and other damaging weather could be made as well. We of course acknowledge that this as an unrealistic recommendation, knowing all too well the time and personnel required for a detailed survey of an extensive bow echo event.

We do recommend, however, that damage surveys become an integrated part of future field programs of severe and hazardous weather phenomena. The surveys should include NWS personnel whenever possible; proper training of NWS personnel, emergency managers, and field researchers is essential. The data—and in fact current weather—should be shared immediately with relevant NWS and emergency management offices. (In hindsight, we did not do this well during BAMEX, although with only a few dedicated surveyors for the entire BAMEX domain and project duration, we had neither the personnel nor resources to do this efficiently.) Collaborative efforts between universities and forecast offices could aid the collection of damage information outside of special field programs, given proper training of the university personnel. These efforts would stand the highest probability for success if organized as a cooperative project, such as those fostered by the University Corporation for Atmospheric Research's Cooperative Program for Operational Meteorology, Education, and Training.

Research into other damage mapping techniques is also recommended. Satellite remote sensing offers a currently underutilized way to constrain the area potentially affected by thunderstorm winds, and could even complement traditional surveys. One possible application to refine is the use of multispectral satellite data, as employed by Yuan et al. (2002), to map and characterize tornado damage from the 3 May 1999 tornado outbreak, and by Bentley et al. (2002), to determine the areal extent of crop damage due to wind and hail. Images are available from other earth-observing satellites that have pixel widths as low as 1 m, and offer promising damage assessment applications. Striking images of the 28 April 2002 La Plata, Maryland, tornado track provided by the National Aeronautics and Space Administration's Earth Observing-1 Advanced Land Imager (with 10-m panchromatic pixels) support this concept. Issues related to the cost and/or access to these very high-resolution data would need to be addressed, however.

Beyond damage mapping, other steps can be taken to improve the severe wind reports. We echo Weiss and Vescio's (1998) recommendation that it be stated explicitly in Storm Data whether a reported wind gust was measured or estimated. Strictly speaking, this is already enabled in the Storm Data system⁴: high wind entries must now also include whether the gust (peak 5-s averaged wind speed) was estimated (by damage) or measured (by known calibrated anemometers). Nonetheless, guidance still needs to be offered on how to interpret the rather specific “estimates” that continue to appear in the publication. For example, a brief perusal of the February 2004 (print) issue shows “estimated” gusts ranging between 50, 51, 52, 53, 56, 61, 67, 70, and 97 kt.

Our inclination is to propose that as many reports as is feasible be gathered and recorded, regardless of the initially perceived event significance. As one of the reviewers warned, however, this could actually have a deleterious effect on the database, in the form of unforeseen biases. So, we instead support the collection of the highest quality data possible; another reviewer suggested that state climatologists, who in fact were the authors of Storm Data during the 1950s and 1960s, might be beneficial to this process. Precise locations should be sought always, and then translated into descriptive wording in the narratives. Good examples of this from past Storm Data entries include estimates of tree size (in terms of diameter and height, rather than “large” or “tall”), exact location of damage (e.g., “near the intersection of County Road 100W and County Road 900N,” instead of “2 miles north of town,” which can be rather vague in rural areas), and an approximate area over which damage occurred (number of city blocks if urban, number of square miles if rural); an integration of geographic information system and global positioning system capabilities into Storm Data could help in this regard. Additional descriptors akin to tornado pathlength and path width could be useful, although a path per se is more illusive and difficult to define since some severe wind reports can be associated with distinct cells as well as large convective systems. Moreover, our experience suggests that typical severe convective systems tend not to damage or “blow down” everything they encounter (Fujita 1981).

Finally, we encourage investigators to seek additional sources of information that can be used to assess the severity of convective wind events. These may include utility companies (locations of downed lines and poles), local police departments, state climate centers and agriculture extension offices, the U.S. Department of Agriculture Farm Service Agency, and pivot irrigation system replacement companies, which to varying degrees are already used by some forecast offices and research teams. Another is the insurance industry (e.g., see Changnon and Changnon 1990). A partnership between this group and the NWS would seem mutually beneficial, and result in a higher quality database and subsequent high wind risk assessments.

In summary, it is our hope that this paper has served to further raise the awareness of issues with the Storm Data database, and increased the appreciation for the need of producing more representative assessments of damage associated with severe convective wind events. A panel to discuss ways of improving severe wind reporting seems warranted.