&NOAA/National Severe Storms Laboratory, Norman, OK

Revised manuscript submitted to

07 November 2003

Corresponding author address: Michael C. Coniglio, National Severe Storms Laboratory,1313 Halley Circle, Norman, OK 73069, E-mail: mailto://Michael.Coniglio@noaa.gov

Long-lived convectively produced windstorms, known as derechos, continue to pose a significant hazard to life and property and remain a difficult forecasting and warning problem. Past studies have examined the climatology of derechos and suggest very different distributions of derechos within the United States. This uncertainty in the climatology of derechos is a concern for forecasters, since knowledge of the relevant climatological information is a key piece in the forecast process. A 16-year data set from 1986-2001 is examined to improve the interpretation of derecho climatology by splitting the years into two periods (1986-95 and 1996-2001) and by stratifying the data into three groups based on the intensity of the event.

The results show aspects seen in earlier climatologies, including a southern axis in the southern Plains that is favored in the mid-1980s and early 1990s and a northern axis centered from the upper Mississippi River valley into Ohio that is favored in more recent years. High-end derechos, that require three wind gust reports (or comparable damage) exceeding 38 m s^-1, appear to be favored in the northern corridor during the warm season, particularly in the later period, and are favored along the lower Mississippi River valley during the colder months in both periods.

Altering the criteria to not require three 33 m s^-1 gust reports or F1-type damage (low-end events) significantly increases the number of events that occur in the lower Appalachians and in portions of the southern axis, particularly in the earlier period. To a lesser extent, the inclusion of low-end events also inflates the frequency values in the northern axis in the later period. The overall effect of including the low-end events is to create a distribution that still suggests both a southern and northern axis, and a shift of the primary axis from the southern Plains in the early period to the upper Mississippi valley in the later period, although not to the degree suggested with low-end events included. Therefore, both the length of the data set and the criteria used to define derechos can significantly influence the resulting climatology.

Long-lived convectively produced windstorms, known as derechos, continue to pose a significant hazard to life and property and remain a difficult forecasting and warning problem (Wakimoto, 2001). One of the important early steps in the operational forecast process is the knowledge of relevant climatological information (Johns and Doswell 1992). However, as described below, criteria for identifying derechos and the geographical frequency distribution of derechos are still being debated in the literature. This study provides evidence that helps to improve the interpretation of the derecho geographical distribution.

Johns and Hirt (1987; JH87 hereafter) were the first to develop specific criteria to define derechos and to estimate their preferred geographical regions. They define the term derecho to be associated with an extratropical mesoscale convective system (MCS; Zipser 1982) that produces, what Fujita and Wakimoto (1981) call, a "family of downburst clusters". They identify derechos based on criteria (Table 1) that could be determined from the National Climatic Data Center's publication Storm Data and logs of severe weather events from operations at the National Severe Storms Forecast Center (the predecessor to the Storm Prediction Center; SPC). The geographical distribution of the 70 warm season events identified by JH87 suggests that warm season derechos occur most frequently in a region from the upper Midwest to the Ohio valley and are relatively infrequent in other locations (Fig. 1a).

Bentley and Mote (1998; BM98 hereafter) examined the SPC database of convective wind gust reports between the years of 1986-1995 and identified 113 events from all months of the year in an attempt to better visualize the climatological distribution of derechos. However, their method for identifying derechos differs somewhat from that used by JH87 (Table 1 - see below). BM98 remove the requirement that three wind gust reports of F1 damage (or wind gust estimates or measurements greater than 33 ms^-1) must be separated by at least 64 km. Additionally, they determined whether or not successive reports emanate from the same MCS by temporally mapping the reports instead of examining radar data. Reducing the maximum elapsed time between successive reports to 2 h and requiring that the maximum distance between successive wind reports is no more than 2° of latitude or longitude helps them make this judgment. In contrast to the results of JH87, BM98 suggested that warm season derechos are a much more common occurrence in the southern Great Plains than in the upper Midwest (Fig. 1b) and identified a smaller maximum (in spatial coverage) near the OH/PA border.

Johns and Evans (2000) proposed some explanations for these differences. They believe that the removal of the 33 m s^-1 wind gust criterion and the tighter report density criteria allows for the deficiencies of the convective wind report data base to have a larger effect on the results. They also suggested that clusters of individual thunderstorms or isolated supercells entered the dataset as a result of not examining the radar data and questioned the adequacy of a 10-year period to depict the true climatology.

Bentley and Mote (2000a) argued that the 33 m s^-1 wind criterion is unnecessary because Fujita and Wakimoto (1981) make no reference to wind gust magnitudes in the definition of downburst clusters, and is difficult to judge because of the uncertainties in the accuracy of wind gust/damage estimates in the SPC database. In addition, they suggested that an anomalously strong ridge in the central U.S. during 1980 provided a pattern that was unusually favorable for derechos in the upper Midwest (24% of the events in the JH87 data base occurred during June-July 1980) and inflated the JH87 results toward that region. Finally, they believe that the parent convective structure should not restrict the definition of derechos.

Bentley and Sparks (2003) add an additional 118 derechos from the period 1996-2000 and show a reemergence of the primary frequency axis across the upper Mississippi River valley similar to the results of JH87, although a significant secondary axis is still present in the southern Plains. They suggested that the shifts in synoptic patterns favorable to producing derechos, particularly during the warm season, help to explain this northward shift in derecho activity over multi-year time periods.

The above section highlights that several factors, relating to the definition of derechos and the time-period of the investigation, could influence the different estimates of the derecho climatology. This study focuses on the effects of eliminating the higher wind gust criteria used in the definition of derechos by JH87. A 16-year data set is used to examine this effect. The development of the derecho data set and the analysis method is described in section 2. As part of the design of this analysis, criteria are proposed that help to identify the preferred locations of particularly intense derecho events that have been identified in past literature (see Miller and Johns 2000). Evidence of derecho multi-year variability is presented in section 3 that corroborates a recent estimate of the derecho climatology in the literature, but attention is focused on the effects of eliminating the higher wind gust criteria. A summary and a final discussion are presented in section 4.

2. DERECHO DATA SET

In this study, Storm Data and the SPC convective wind database are examined for the years of 1986-2001 to identify derechos. The first two gust criteria that are used in JH87 and BM98 (listed in Table 1) also are used in this study. For the criteria related to the density of reports, a compromise is proposed between JH87 and BM98 such that wind gust reports can be separated by no more than 2.5 h and 200 km. Note that successive wind reports (temporally) can be separated by more than 200 km, but each report must occur within 200 km of the other reports within the wind-gust swath. This situation of successive wind reports that are separated several hundred km frequently occurs with serial derechos (JH87).

Although there is substantial uncertainty in the accuracy and reliability of the wind gust estimates (Weiss et al. 2002), there appears to be enough fidelity to identify events that are noticeably more intense than those that only satisfy the criteria of BM98 and those that only marginally satisfy the criteria of JH87. In making this judgment, the use of Storm Data to supplement the SPC database is crucial because of the very detailed verbal descriptions of the damage that often accompany the most severe events. Therefore, the effect of the higher wind gust criteria is examined by dividing the dataset into three groups that are separated by an estimate of the event's intensity (see criterion #4 of Table 1). The three categories are defined ("low-end", "moderate", and "high-end" events) so that the low-end classification is the same as that used by BM98, the moderate classification discriminates between low-end events and those events that satisfy the JH87 criteria, and the high-end classification emulates characteristics of particularly intense derecho events described in Miller and Johns (2000). Each event is categorized as low-end, moderate, or high-end and is not counted in more than one category. Therefore, the low-end events do not require additional wind gust criteria beyond the 26 m s^-1 threshold and the moderate events satisfy the 33 m s^-1 wind (or F1 damage) criterion of JH87 but not the high-end criteria. The high-end events require at least three reports of wind gusts greater than 38 m s^-1 (75 kt or 85 mph) separated by at least 64 km, with at least two reports occurring during the MCS stage of the event. The MCS stage is reached when a group of convective echoes form a contiguous precipitation shield greater than 100 km in horizontal extent and for longer than 3 h (Zipser 1982; Parker and Johnson 2000).

If no wind gust measurement or estimate is given for the report, a high-end classification can also be met by examining the description of the damage associated with the report, as similarly done by JH87. To determine the types of damage that should qualify for high-end classification, the damage descriptions in Storm Data that are associated with wind gusts of > 75 kt for a wide variety of reports are examined. The complete destruction of mobile homes (usually tossed far from their blocks), significant damage to well-constructed homes and businesses (often including several roofs blown off), and large swaths of trees being flattened within forested areas are examples of this type of damage. Miller and Johns (2000) find this type of damage in several intense derecho events. The F1-type wind damage that helps to define the moderate derecho events also is examined in this manner.

Another major difference in the way that derechos have been identified in literature involves the use of radar data. Unlike BM98, it was decided that radar data should be considered in the definition of derechos for two reasons; to ensure that the wind reports are associated with an extratropical MCS and to ensure that multiple swaths of damage emanate from the same MCS. Regarding the first reason, Bentley and Mote (2000) argue that the parent MCS should not define a derecho, much like a tornado is not defined by its associated thunderstorm, and thus, isolated cells or groups of ordinary thunderstorms that produce sufficiently long swaths of severe wind gusts should be considered as derecho-producing MCSs. However, this study interprets the definitions of MCSs by Zipser (1982) and Parker and Johnson (2000) to be associated with groups of individual thunderstorms that merge to form a common precipitation shield and surface gust front on length scales of at least 100 km. Although we believe that the distinction between an isolated, long-lived supercell and an MCS is not always clear, this MCS definition clearly excludes isolated thunderstorm cells or groups of thunderstorm cells that remain isolated during the majority of their lifetime. Therefore, this study interprets the derecho definition of JH87 as one that intends to separate convective windstorms produced by supercells or groups of isolated cells from those produced by MCSs. Note that this definition allows MCSs that may contain embedded supercells and allows squall lines with limited cross-line extent, but extensive along-line extent. The available WSR-88D level II reflectivity data obtained from the National Climatic Data Center (NCDC), the WSR-88D mosaic reflectivity images at 2-km (available at http://www.spc.noaa.gov/exper/archive/events/index.html) or 4-km resolution (available at http://www.ncdc.noaa.gov/oa/radar/radardata.html) or the archived hourly radar summary charts produced by the National Centers for Environmental Prediction (NCEP) (obtained from NCDC) are used to examine this criterion.

Although the effects of not including the radar data are not quantified in this study, it is worthwhile to briefly discuss these effects in a qualitative manner. BM98 and Bentley and Sparks (2003) attempt to determine whether or not multiple wind gust swaths emanate from the same convective system by temporally mapping the wind gust reports. In terms of the method of identifying derechos in this study, it is found that the isolated, progressive MCSs that produce a compact and steady sequence of reports are often identified much the same with and without the use of radar data. However, it is found that radar information (particularly WSR-88D data) substantially aids in the assessment that multiple swaths emanate from the same convective system, and in many cases, is essential in making this determination. This point is illustrated by plotting the wind reports from 2200 UTC 11 August 2000 to 0830 UTC 12 August 2000 in the northern High Plains region (Fig. 2). Without the aid of radar data, it is very difficult to determine with certainty whether or not the reports emanate from the same convective phenomenon given the regular temporal progression of the reports from east-central Montana to northwestern Minnesota (Fig. 2). WSR-88D data reveals that the first portion of the swath occurring from 2201 UTC to 0425 UTC emanates from an isolated, long-track supercell on horizontal scales < 100 km, whereas the swath from 0400 UTC to 0830 UTC emanates from an MCS that develops along the intersection of a pre-existing boundary and surging outflow ahead of the decaying supercell. Even if one permits a supercell to be considered a derecho-producing convective system, this cannot be considered a derecho since two separate convective systems are involved. Furthermore, the swath that emanates from the MCS does not satisfy the length criteria, and therefore should not be considered a derecho.

Many other severe wind episodes are found that satisfy the wind criteria but never form an organized MCS structure. Additionally, the lack of quality radar information often hinders the ability determine the reports that comprise the derecho, particularly with the serial events in which line segments or individual cells often flank the main squall line, but remain separated from its evolution. These points underscore the importance of examining quality radar data in order to produce the best possible record of derecho occurrences.

The time and location of the first wind report associated with the convection that becomes the parent MCS defines the origin of each event (i.e., wind reports from isolated cells are allowed, but only if those cells later become part of the MCS structure that produces the main swath of reports). Similarly, the time and location of the last wind report associated with the MCS defines the termination of the event. To examine the geographical distribution of the events, the wind reports from each event are mapped to a Cartesian grid with 200 km by 200 km grid cells, which is not substantially different than the 1.83° X 1.83° resolution used by BM98 and the 2° X 2° resolution used by JH87. The distributions are determined by counting the cells that contain at least one wind report from each event and then contouring the sum of each cell over a given time period.

It should be noted that this study makes no explicit attempt to correct for many of the non-meteorological factors, such as changes in population characteristics and especially reporting biases (Johns and Evans 2000) that can complicate the interpretation of results. In a related manner, given the fact that the WSR-88D data is widely available for only the more recent years in the data set, it is more difficult to identify events in the earlier years. Combined with the fact that the number of wind gust reports greatly increased in the early to late 1990s (Johns and Evans 2000), this skews the annual number of events toward the more recent years and likely results in an underestimate of the actual number of events prior to the WSR-88D era. Although these factors continue to limit the ability to depict the derecho climatology, the large number of events over a 16-year time period can be used to effectively examine the effect of the higher wind gust criteria on the differences in the estimated derecho climatology of JH87 and BM98 and to gain further insight on the true derecho climatology.

3. RESULTS

a) Multi-year variability

1) WARM SEASON EVENTS

The criteria outlined in section 2 are used to identify 244 derecho events that qualify for at least low-end classification. Among the 244 events from all times of the year, there are 73 low-end derechos (55 May-Aug. events), 116 moderate derechos (82 May-Aug. events), and 55 high-end derechos (31 May-Aug. evens). The derecho events occur year-round, but are primarily a warm season phenomenon (168, or 69%, of the cases occur in the months of May-Aug.).

The regional distribution of events is first displayed for the 168 warm season (May-Aug.) events (including the low-end events) from the years of 1986-2001 to facilitate a direct comparison to the results of JH87 and BM98. Results from all 16 years show two main activity corridors that appear to combine the JH87 and BM98 results; one stretching from Minnesota to western Ohio and another that covers the southern and eastern portions of the Plains through eastern Arkansas (Fig. 3a). This is similar to the distribution of events defined and identified by Bentley and Sparks (2003) for events in the period 1986-2000 from all times of the year (see their Fig. 14), except that the distribution in this study identifies more events further north and west in the upper Mississippi River valley and the northern High Plains and fewer events in the Ohio valley. The two axes also resemble the favored regions of general northwest flow severe weather outbreaks identified by Johns (1982, 1984).

To examine the multi-year variability, the distribution is broken down into the 70 warm season events identified from 1986-95, which is the same time period examined in BM98, and the 98 warm season events identified in the period of 1996-2001; the period in which the availability of WSR-88D data significantly facilitates the identification of derechos.

Despite the differences in the derecho definition between this study and BM98, warm-season derechos (low-end events included) appear to be much more frequent in the southern Plains than in the upper Midwest during the period of 1986-95 (Fig. 3b), as first suggested by BM98 (Fig. 1b). The distribution also suggests an extension of the southern-Plains maximum into the mid-Mississippi River valley (c.f. Figs. 1b and 3b). However, one distinct difference between this study and BM98 is the lack of a maximum in events near the Ohio/Pennsylvania border (c.f. Fig. 1b and Fig. 3b). This is partly because many of the events that were discarded due to inconclusive radar data and due to the failure of satisfying the MCS criterion occurred in this region.

There is a modest degree of confidence that the improved reporting and verification techniques and the wide availability of WSR-88D data allow the identification of most of the actual events from the mid-90s to the present. In the 6-year period of 1996-2001, warm-season derechos were primarily favored in the upper Mississippi River valley through the western Ohio valley (Fig. 3c), as suggested by JH87 (Fig. 1a). However, Fig. 3c also shows that events still occurred in the central and southern Great Plains, which is not evident in the JH87 distribution. A similar shift in the primary derecho regions also is suggested by Bentley and Sparks (2003), who argue that the southern-Plains maximum in the late 80s-early 90s, combined with the lack of events in the upper Midwest, may have been the result of a shift in favorable synoptic conditions on multi-year time scales. Concurrently, BM98 and Bentley and Mote (2000a) suggested that anomalously strong ridging in the central Plains in the period of 1980-83 may have helped produce the pronounced upper-Midwest maximum in events and the lack of southern-Plains events found by JH87.

In addition to the locations of favorable synoptic patterns, the frequency of favorable patterns also is suggested to have a considerable impact. BM98 observe that a favorable synoptic pattern in 1995 persisted through much of the summer months and produced 25 derechos (22 are identified in this study), whereas the persistent drought conditions of 1988 allowed only 1 derecho to occur. This study shows, as in Bentley and Sparks (2003), that the events since 1995 show similar tendencies for derechos to occur in groups within persistently favorable flow regimes. In fact, this study identifies 28 derechos in the 46-day period from 15 May - 30 June 1998; a period characterized by frequent intrusions of a strong westerly polar and/or subtropical jet stream on top of a persistent ridge that meandered across the Plains states.

Combined with the estimation of derecho activity from the earlier years presented in JH87, BM98, Bentley and Sparks (2003) and in this study, warm-season derechos can occur almost anywhere east of the Rocky Mountains, but seem to favor both the upper-Midwest/upper-Mississippi River valley and the central and southern Great Plains, depending upon the nature of the mean flow regime. However, it is stressed in section 3b that due to the differences in reporting characteristics within the 16-year period spanned in this study (Johns and Evans 2000; Weiss et al. 2002), the extent to which non-meteorological factors play a role in effecting this apparent shift in the derecho distribution from earlier years remains unclear.

2) COLD SEASON EVENTS

There is less evidence of a regional shift in the distribution of the Sept.-Apr. events within the two time periods examined above, which suggests that the cold season precipitation regimes that favor derechos may be less variable than their warm season counterparts. The results from the entire 16-year period show a primary region in the lower Mississippi River valley and the Gulf Coast states and a secondary axis from the lower Ohio valley into Pennsylvania (Fig. 4a). Bentley and Mote (2000b) suggest a similar distribution for the Sept.-Feb. events identified in their study of cool-season derechos. Most of these events are produced by elongated squall lines, that often show characteristics of both progressive and serial derechos (JH87), in association with a mobile upstream trough and an associated deepening or mature low-level cyclone.

Unlike the maximum frequency axis for the warm season events, the maximum in colder season events remains in approximately the same location for both the 1986-95 and 1996-2001 time periods (in Mississippi; shown in Figs. 4b and 4c). However, there are some differences among these time periods including the northward extension of events into the mid-Mississippi River valley and southeastern Plains region, as well as the appearance of a significant number of events in Tennessee and Kentucky in the later period, as shown by Bentley and Sparks (2003). As with the distribution of the warm-season events, this suggests the more frequent occurrence of synoptically favorable regimes in this region in later years, but the extent to which the non-meteorological factors discussed in Weiss et al. (2002) affect this change in the distribution is not clear. The next section examines the effects of one such factor, namely the effects of including low-end events into the data set, on both the warm and cool-season derecho distributions.

2. Derecho intensity

1) WARM SEASON EVENTS

As done previously, the distribution among the low-end, moderate, and high-end derecho events is first presented in terms of the warm season events to enable a direct comparison to JH87 and BM98. Note that the maximum frequency axis for the low-end events stretches across the southern-Plains (Fig. 5a). The moderate events are found with nearly equal frequency in both of these regions (Fig. 5b). For the high-end evens, the maximum frequency axis is found in the upper Mississippi River valley (Fig. 5c), particularly for the later period (1996-2001) (not shown). Most of these events show primarily progressive-MCS characteristics and contain long-lived and well-defined bow echoes (Fujita 1978) during maturity, some of which may contain embedded supercells or may begin as groups of isolated supercells (Klimowski 2000; Miller and Johns 2000).

The effects of including the low-end events during the warm season are emphasized in Figure 6. The low-end events appear to account for the largest percentage of the total number of events in the lower Appalachians, particularly in eastern Tennessee and the western Carolinas, where the low-end events account for greater than 50% of the total number of events. Additionally, locations in the lower Ohio valley also show areas with greater than 40%. As stated before, the Ohio valley region contained many severe wind episodes that were not included as derechos because of inconclusive radar data and due to the failure of satisfying the MCS criterion. Figure 6 shows that the exclusion of low-end events further contributes to lowering the frequency values in this region, as compared to BM98 (Fig. 1b).

In terms of the preferred southern and northern frequency axes found in the overall warm season distribution (Fig. 3a), there are locations within the southern axis in which the low-end events account for > 40% of the total number of events, specifically in eastern Oklahoma and southern Arkansas (Fig 6). In comparison, the low-end events generally account for only 20-30% of the total number of events in the upper Mississippi River valley through the western Ohio valley (Fig. 6). The low-end events in the southern axis are largely accounted for in the period 1986-95 (Fig. 7a), whereas the low-end events in the upper Mississippi River valley and Ohio valley occur almost exclusively in the period 1996-2001 (Fig. 7b). Low-end events from both time periods contribute to the inflation of events in the lower Appalachians (Figs. 7a and 7b).

Figures 6 and 7 show that the removal of the higher wind gust criterion (33 m s^-1 in table 1) inflates the numbers in all locations, but appears to have the greatest impact on the distribution of warm-season events in the southern U.S in the earlier years. To reinforce this point, the moderate and high-end warm season events are combined into one distribution (Fig. 8). The most apparent effect is the shrinking of the distribution in the lower Appalachians (c.f. 3a and 8a). The effect on the primary frequency axes is subtler when viewing the entire 16-year distribution. The southern axis is similar in frequency to the northern axis when low-end events are included (Fig. 3a). With the low-end events excluded (emulating the events considered to be derechos by JH87), the frequency values for the northern axis become slightly larger and appear to cover a larger geographical area than those for the southern axis (c.f. 3a and 8a).

The effect of eliminating the low-end events is more clearly seen when comparing the distribution for the moderate and high-end warm season events for the time period of 1986-95 (Fig. 8b) to the distribution over the same time period with low-end events included (Fig. 3b). From 1986-95, the southern Plains still contains the primary warm season axis, but the frequency values and their geographical coverage are significantly reduced as compared to those found in this study with the low-end events included (c.f. 3b and 8b) and also in BM98 (Fig. 1b). Concurrently, the frequency values in the northern axis are essentially unchanged (c.f. 3b and 8b). As hypothesized by Johns and Evans (2000), the distribution presented by BM98 falsely suggests that derechos were several times more likely in the southern Plains than the northern Mississippi River valley in this time period. Indeed, figure 8b shows frequency values in the southern axis that are much more comparable to those in the northern axis. Therefore, the removal of the 33 m s^-1 wind criterion, which allows the identification of low-end events in this study, appear to play a significant part in the distinct southern Plains maximum in derechos shown in BM98.

The removal of low-end events also reduces the frequency values in the northern axis for the time period of 1996-2001 (c.f. 8c and 3c). However, this reduction isn't as apparent as compared to the reduction in frequency values and geographical coverage of the southern axis in the earlier period, which is to be expected from figure 6. The overall effect is to create a distribution that still suggests a shift of the primary axis from the southern Plains in the late 80s to early 90s to the upper Mississippi River valley from the late 90s-early 2000s although not to the degree suggested by BM98 and Bentley and Sparks (2003).

2) COLD SEASON EVENTS

For completeness, the regional distributions for the 18 low-end derechos, the 34 moderate derechos, and the 24 high-end derechos for the remaining months (Sept.-Apr.) are shown in Fig. 9. This reveals that the largest concentration of low-end events is found in a relatively confined area from southern Louisiana through central Alabama (Fig. 9a), where 30-50% of the total number of events are low-end derechos (not shown). This decreases the large frequency values shown in this region (Fig. 4a), but does not significantly impact the location of this maximum frequency axis along the Gulf coast states. Also notice that most of the colder-season derechos that comprise the majority of the events from the southern Plains through the Ohio valley are either moderate or high-end events (Fig. 9b) and are found in the later period (Fig. 4c). The high-end events alone are favored along the western Gulf coast states, but also show an extension into the Ohio valley, similar to the overall distribution. The overall distribution with low-end events excluded (not shown) still shows a maximum in the number of events in the lower Mississippi River valley, but with frequency values that are more comparable to those in the Ohio valley (as compared to figure 4a).

4. SUMMARY AND CONCLUSIONS

A data set of 244 derecho events over a 16-year period is used to examine the effects of changing the study period and the criteria for identifying derechos and to gain further insight into the interpretation of the underlying geographical distribution of derechos. This study focuses on the choice of BM98 to remove the requirement preferred by JH87 of three 33 m s^-1 wind gust reports. In this study, events that do not meet this criterion are termed "low-end" events. This study also proposes new criteria to help identify particularly intense derecho events (high-end events) and identifies their preferred geographical regions.

The overall distribution over the entire 16-year period (with low-end events included) suggests two primary axes of warm season events; one from the southern Plains/Arkansas region and one stretching from the upper Mississippi River valley through the Ohio valley. The southern axis is preferred in the earlier years (1986-95), while the northern axis becomes prominent in later years (1996-2001). These results provide some supporting evidence to the results of Bentley and Sparks (2003), who show a similar northward shift in primary derecho activity and suggest that this shift is tied to the location and frequency of favorable synoptic patterns over multi-year periods.

The high-end warm season events are found to be favored in the northern axis, particularly in the later years (1996-2001), and are a relatively infrequent occurrence in the central and southern Plains. The high-end derecho distribution for the colder months (Sep.-Apr.) shows a maximum frequency axis along the western Gulf coast states, centered in Mississippi, and an extension of events into the Ohio valley region, similar to the overall distribution of colder season events.

The greatest effect of removing the low-end criterion for the warm-season events is found in the lower Appalachians, where low-end events account for greater than 50% of the events in some locations. In terms of the effects on the two primary frequency axes mentioned above, low-end events account for greater than 40% of the total number of events in portions of the southern Plains/Arkansas region. Low-end events from the time period 1986-95 largely account for this inflation of frequency values. Low-end events also are found to inflate the frequency values in the upper Mississippi River valley region through the western Ohio valley in the time period from 1996-2001, but generally account for only 20-30% of the total number of events. With the low-end events removed, the overall regional distribution of warm season events still suggests two similar high frequency axes over the entire 16-year period; one across the upper Midwest through the Ohio valley, as previously suggested by JH87, and another across the southern Plains, as suggested by BM98. The southern axis is somewhat more prominent in the period of 1986-95, but not to the extent suggested by BM98, while the northern axis is somewhat more prominent in the time period of 1996-2001, but not to the extent suggested by JH87 and Bentley and Sparks (2003).

The inclusion of colder season, low-end derechos inflates the frequency values by 30-50% in a relatively confined region from Louisiana to central Alabama. However, the overall distribution of colder season events still maintains the suggestion of a maximum in the number of events in the lower Mississippi River valley and an extension in the frequency of events northeast into Ohio.

This study also briefly examines the effects of including radar data. The use of radar data is found to eliminate severe wind episodes that are associated with either isolated supercells or groups of individual cells. In addition, the determination of the requirement that multiple swaths of reports be associated with the same convective system is greatly aided with the use of radar data and is illustrated with an example. Although this study shows no quantitative results on the total effects of including radar data in the analysis, this change is found to have an impact on the frequency of events in the Ohio/Pennsylvania area, in which the frequency values suggested by BM98 are significantly reduced (the elimination of the low-end criteria also contributes to a reduction of warm season events in this region). With more years of quality WSR-88D data available, future studies should attempt to quantify this effect in more detail, particularly the percentage of events that are falsely identified as emanating from the same convective system by failing to examine the radar data and the percentage of severe wind episodes that emanate from supercells or groups of cells that fail to organize into a MCS.

Although we have shown some of the effects of non-meteorological factors, it is stressed that there are other non-meteorological factors that complicate the interpretation of the results that aren't considered in this study (Johns and Evans 2000), particularly the explosion of low-end reports since the mid to late 80s (Weiss et al. 2002). Future studies also should attempt to quantify these factors in order to improve the interpretation of the derecho climatology.

ACKNOWLEDGMENTS: We greatly appreciate the constructive comments and suggestions provided by four reviewers, Robert Johns, and Dr. Harold Brooks. Motivation and ideas for this work came from discussions with Steven Weiss and several of the forecasters from the NOAA/Storm Prediction Center. It would not have been possible to complete this work in any reasonable amount of time without the software to view the SPC convective wind data base (svrplot 2.0) developed by John Hart and Paul Janish of the Storm Prediction Center. This research was supported by NSF grant ATM-0138559.

Bentley, M.L., and T.L. Mote, 1998: A climatology of derecho-producing mesoscale convective systems in the central and eastern United States, 1986-1995. Part I: Temporal and spatial distribution. Bull. Amer. Meteor. Soc., 79, 2527-2540.

--------, and --------, 2000a: Reply to "Comments on a climatology of derecho-producing mesoscale convective systems in the central and eastern United States, 1986-1995. Part I: Temporal and spatial distribution". Bull. Amer. Meteor. Soc., 81, 1055-1060.

--------, and --------, 2000b: A synoptic climatology of cool-season derecho events. Physical Geography, 21, 21-37.

--------, and J.M. Sparks, 2003: A 15 yr climatology of derecho-producing mesoscale convective systems over the central and eastern United States. Clim. Research., 24, 129-139.

Fujita, T.T., 1978: Manual of downburst identification for Project NIMROD. SMRP Res. Paper 156, University of Chicago, 104 pp.

--------, and R.M.Wakimoto, 1981: Five scales of airflow associated with a series of downbursts on 16 July 1980. Mon. Wea. Rev., 109, 1438-1456.

Johns, R.H., 1982: A synoptic climatology of northwest flow severe weather outbreaks. Part I: Nature and significance. Mon. Wea. Rev., 110, 1653-1663.

--------, 1984: A synoptic climatology of northwest flow severe weather outbreaks. Part II: Meteorological parameters and synoptic patterns. Mon. Wea. Rev., 112, 449-464.

--------, and W.D. Hirt, 1987: Derechos: widespread convectively induced windstorms. Wea. Forecasting, 1, 32-49.

-------, and C.A. Doswell III, 1992: Severe local storms forecasting. Wea. Forecasting, 7, 588-612.

-------, and J.S. Evans, 2000: Comments on "A climatology of derecho-producing mesoscale convective systems in the central and eastern United States, 1986-1995. Part I: Temporal and spatial distribution". Bull. Amer. Meteor. Soc., 81, 1049-1054.

Klimowski, B.A., R. Przybylinski, G. Schmocker, M.R. Hjelmfelt, 2000: Observations of the formation and early evolution of bow echoes. Preprints, 20^th Conf. on Severe Local Storms, Orlando, FL, Amer. Meteor. Soc., 44-47.

Miller, D.J., R.H. Johns, 2000: A detailed look at extreme wind damage in derecho events. Preprints, 20^th Conf. on Severe Local Storms, Orlando, FL, Amer. Meteor. Soc., 52-55.

Parker, M.D., and R.H. Johnson, 2000: Organizational modes of midlatitude mesoscale convective systems. Mon. Wea. Rev., 128, 3413-3436.

Wakimoto, R.M., 2001: Convectively driven high wind events. Severe Convective Storms, AMS Meteor. Monogr., C. Doswell III, Ed., 28, Amer. Meteor. Soc., 255-298.

Weiss, S.J., J.A. Hart, and P.R. Janish, 2002: An examination of severe thunderstorm wind report climatology: 1970-1999. Preprints, 21^st Conf. Severe Local Storms, San Antonio, TX, Amer. Meteor. Soc., 446-449.

Zipser, K.A, 1982: Use of a conceptual model of the life cycle of mesoscale convective systems to improve very-short-range forecasts. Nowcasting, K. Browning, Ed., Academic Press, 191-204.

Table 1. Criteria used to identify derecho events in JH87, BM98, and in our study.
#	JH87 criteria	BM98 criteria	Our criteria
1)	There must be a concentrated area of convectively induced wind gusts greater than 26 m s-1 that has a major axis length of 400 km or more.	Same as JH87	Same as JH87
2)	The wind reports must have chronological progression.	Same as JH87	Same as JH87
3)	No more than 3 h can elapse between successive wind reports.	No more than 2 h can elapse between successive wind reports.	No more than 2.5 h can elapse between successive wind reports.
4)	There must be at least three reports of either F1 damage or wind gusts greater than 33 m s-1 separated by at least 64 km during the MCS stage of the event.	Not used	Low-end: Not used Moderate: Same as JH87 High-end: There must be at least three reports of either wind gusts greater than 38 m s-1 or comparable damage (see text), at least two of which must occur during the MCS stage of the event.
5)	The associated MCS must have spatial and temporal continuity.	The associated MCS must have spatial and temporal continuity with no more than 2° of latitude and longitude separating successive wind reports.	The associated MCS must have spatial and temporal continuity and each report must be within 200 km of the other reports within a wind-gust swath.
6)	Multiple swaths of damage must be part of the same MCS as indicated by the available radar data.	Multiple swaths of damage must be part of the same MCS as seen by temporally mapping the wind reports of each event.	Same as JH87