Abstract

During a 5-yr period of study from 2000 to 2004, slightly more than 10% of all National Weather Service (NWS) tornado warnings were issued either simultaneously as the tornado formed (i.e., with zero lead time) or minutes after initial tornado formation but prior to tornado dissipation (i.e., with “negative” lead time). This study examines why these tornadoes were not warned in advance, and what climate, storm morphology, and sociological factors may have played a role in delaying the issuance of the warning. This dataset of zero and negative lead time warnings are sorted by their F-scale ratings, geographically by region and weather forecast office (WFO), hour of the day, month of the year, tornado-to-radar distance, county population density, and number of tornadoes by day, hour, and order of occurrence. Two key results from this study are (i) providing advance warning on the first tornado of the day remains a difficult challenge and (ii) the more isolated the tornado event, the less likelihood that an advance warning is provided. WFOs that experience many large-scale outbreaks have a lower proportion of warnings with negative lead time than WFOs that experience many more isolated, one-tornado or two-tornado warning days. Monthly and geographic trends in lead time are directly impacted by the number of multiple tornado events. Except for a few isolated cases, the impacts of tornado-to-radar distance, county population density, and storm morphology did not have a significant impact on negative lead-time warnings.

1. Introduction

Statistics indicate that the National Weather Service (NWS) was able to provide warnings to the general public on about 75% of all tornadoes identified during 2006 (NWS 2007). With the modernization of the NWS (Friday 1994) including the introduction of Weather Surveillance Radars-1988 Doppler (WSR-88Ds; Crum and Alberty 1993) and the Advanced Weather Interactive Processing System (AWIPS) environment (Seguin 2002), automated feature detection algorithms (Mitchell et al. 1998; Stumpf et al. 1998), increasingly sophisticated training (Magsig et al. 2006), and greater conceptual understanding through field programs such as the Verification of the Origins of Rotation in Tornadoes Experiment (VORTEX-95; Rasmussen et al. 1994), the average tornado warning lead time is now 13 min (Erickson and Brooks 2006), thereby saving lives and reducing injuries (Simmons and Sutter 2005). Nevertheless, in lieu of these advances, a significant number of tornado warnings still are issued either simultaneously as the tornado forms (i.e., with zero lead time) or minutes after initial tornado formation but prior to tornado dissipation (i.e., with “negative” lead time). As long as these tornadoes have zero or negative lead times, the potential will exist for the public to remain unwarned in advance of these potentially destructive misoscale wind events. This study asks the questions: What, if any, specific causes prevented these tornado warnings from being issued sooner? What climatological, storm morphological, or sociological factors play a role? Does a delayed NWS warning impact tornado fatalities? Finally, what are areas of focus for improving the current warning system to reduce the number of zero and negative lead time tornado warnings?

2. Data

A compilation of 5 yr of tornado warning reports from 2000 to 2004 were obtained from the National Oceanic and Atmospheric Administration (NOAA)/NWS. These data included information on all confirmed tornadoes and their associated NWS warning; a list of those tornadoes without advance NWS warning was also provided. These reports contained the date and location of each tornado, the time of the event, the time the NWS warning was issued, the weather forecast office (WFO) that issued the warning, the county or parish location, the estimated F-scale rating, the number of fatalities and estimates of the damage. Additional information on each zero and negative lead-time event was obtained from the National Climatic Data Center, including tornado pathlength, width, and duration. County population estimates were obtained from the 1 July 2000 population estimates produced by the Population Division of the U.S. Census Bureau. Storm morphology was determined by manual review of radar reflectivity data, using the classification scheme as suggested by Gallus et al. (2008). Each storm event was classified as either line, cell, tropical (hurricane, tropical storm), or undefined. No radar velocity data were included in this classification scheme.

All zero and negative lead time warnings are associated with tornadoes that were still in progress when the warning was issued. No NWS warnings issued after tornado dissipation were included in this study.

3. Climatology

In an effort to better understand the root causes for the delays in the zero and negative lead-time warnings, especially when compared to those warnings issued well in advance of touchdown, a climatology of the zero and negative lead-time warnings (henceforth, simply referred to together as negative lead-time warnings) was completed. The negative lead-time warnings were sorted as a function of F-scale ranking, geographically by region and WFO, hour of the day, month of the year, distance from radar, county population, and number of tornadoes by day, hour, and order of occurrence.

The database of warned tornadoes from 2000 to 2004 includes 4604 (89.1%) positive lead-time events, 211 (4.1%) zero lead-time events, and 355 (6.9%) negative lead-time events. The negative lead times ranged from −1 to −55 min, with a mean negative lead time of −5.2 min and a median negative lead time of −3 min (excluding zero lead-time events). Fifty tornadoes had a negative lead time of −10 min or more.

a. Negative lead-time warnings sorted by F scale

All negative lead-time warnings were compared against the estimated F-scale damage incurred by the event (Fig. 1). During the 5 yr of study, the distributions of F0, F1, F2, F3, and F4 tornadoes were 59.5%, 28.4%, 9.9%, 1.6%, and 0.5%, respectively. This distribution is nearly identical to the F-scale estimates of those tornadoes with positive lead times, with distributions of 63.9%, 24.2%, 8.1%, 3.0%, and 0.7%, respectively. Note that there were no F5 tornadoes reported during the 5 yr of study.

Fig. 1.

All zero and negative lead-time events plotted as a function of the F scale.

Fig. 1.

All zero and negative lead-time events plotted as a function of the F scale.

An examination of Fig. 1 highlights four outliers in the dataset with negative lead times of −40 min or greater. A summary of these four events is as follows:

  1. A tornado in Angelina County, Texas, formed within a severe squall line causing $5 million in damage but no fatalities. The 30 March 2002 tornado was approximately 170 km from the nearest radar and occurred near dusk at 0105 UTC on a Saturday night.

  2. A tornado occurred on 7 September 2004 in Sumter County, South Carolina, spawned by Hurricane Frances. It caused $1.7 million in damage, but no fatalities. The tornado was approximately 68 km from the nearest radar and in a relatively populated area, but occurred during the early morning hours at 1128 UTC and was 1 of 19 tornadoes warned on by the WFO that day.

  3. A tornado in Baca County, Colorado, formed 16 July 2000 from an undefined (not readily classifiable) convective complex. It caused no damage or fatalities, and no other tornadoes were reported that day. The tornado was approximately 205 km from the nearest radar in a county with a county population density of 0.7 persons per square kilometer (national average county population density is ∼31 persons per square kilometer).

  4. A tornado in Henry County, Tennessee, formed after dark (0425 UTC) on 6 May 2003 from an undefined convective complex. It caused $15,000 in damage and no fatalities.

These four cases highlight four critical challenges to the operational forecaster. First, none of these four tornadoes developed from a parent classical supercell thunderstorm. Tornadoes developing from lines, hurricanes, and less-organized convection are in general more difficult to anticipate and detect. A second challenge for the operational forecaster is the lack of spotter reports, particularly at night and in sparsely populated regions. Spotter reports provide ongoing analysis of storm evolution, and instant communication for when tornado formation begins, lessening the chance for a large negative lead-time warning. Three of the four events occurred during the evening through early morning hours (events 1, 2, and 4). A third challenge is identifying tornadoes when the storms are far from radar. The center of a radar beam at 0.5° elevation at a range of 120 km is ∼1.98 km AGL, meaning that the radar beam is overshooting the lowest 2 km beyond a range of 120 km. Two of the four tornado warning outliers with large negative lead times had tornado-to-radar distances of 120 km or greater. A fourth challenge for the operational forecaster is providing advance warning for situations in which many warnings are needed for a variety of severe weather conditions, such as during Hurricane Frances. Providing advance warnings for isolated events during nonclassical severe weather situations, such as in Baca County, also presents some unique challenges.

A list of all tornadoes classified as F3 or F4 with zero or negative lead times is presented in Table 1. Between 2000 and 2004, 12 tornadoes occurred that were classified as F3 or higher with zero or negative lead time. The distance of the tornado from the radar may have impacted the warning lead time. The four events with the greatest negative lead times are also the same four cases with the farthest tornado-to-radar distance. Nevertheless, the relatively small sample size of F3 and F4 tornadoes in this dataset makes any inferences difficult.

Table 1.

List of all F3 and F4 tornadoes with zero or negative lead times.

List of all F3 and F4 tornadoes with zero or negative lead times.
List of all F3 and F4 tornadoes with zero or negative lead times.

b. Geographical distribution

Are zero and negative lead times more prevalent in certain geographic regions of the country? One may surmise that variations in storm climatology combined with differences in NWS office warning management and experience could lead to varying patterns of negative lead times across the country.

To assess the impact of regional climatology on the warning process, all (positive, zero, and negative lead time) tornado warnings were sorted among four broad geographic regions: southeast (SE), midwest/east (MW), plains, and west (Fig. 2). In general, the lowest proportion of zero and negative lead time warnings was issued in the southeast region, with the plains region having the greatest positive lead time and lowest negative lead time (Table 2). When combined, all tornado warning data fit a gamma distribution, so a Wilcoxon–Mann–Whitney rank sum test was used to determine the uniqueness of these four regions. For positive lead time warnings, the plains region was significantly different from the SE and MW regions at the 99% confidence level, and the West region was significantly different from the SE and MW regions at the 90% confidence level. For negative lead times, the plains region was found to be significantly different from each of the other three regions at the 90% confidence level; none of the other regions were significantly different from any other. Including the zero lead-time events, the plains region was found to be significantly different from all other regions at the 99% confidence level.

Fig. 2.

All data were divided among four geographic regions: midwest/east, southeast, plains, and west.

Fig. 2.

All data were divided among four geographic regions: midwest/east, southeast, plains, and west.

Table 2.

Positive, zero, and negative lead-time statistics for the four geographic regions.

Positive, zero, and negative lead-time statistics for the four geographic regions.
Positive, zero, and negative lead-time statistics for the four geographic regions.

Next, all zero and negative lead-time events were sorted by the NWS WFO that issued each warning. A percentage of the total number of tornado warnings with a zero or negative lead time was calculated for each WFO. These percentages of zero and negative lead times are plotted against the total number of verified tornado warnings issued by each WFO (Fig. 3). WFOs with less than 10 tornado warnings issued during the 5-yr period were not plotted.

Fig. 3.

Percentage of warned tornado events with (a) zero and (b) negative lead times plotted against the total number of tornado warnings issued by that WFO during the 5 yr of study.

Fig. 3.

Percentage of warned tornado events with (a) zero and (b) negative lead times plotted against the total number of tornado warnings issued by that WFO during the 5 yr of study.

As evident in Fig. 3, in general the greater number of tornado events warned by a given WFO, the lower percentage of those warnings are issued with zero or negative lead time. All WFOs with over 100 tornado warnings during the 5-yr period of study averaged less than 10% of warnings with negative lead time.

A few WFOs were found to have exceptionally high zero or negative lead-time rates when compared with other forecast offices. Over 30% of all tornado warnings from two WFOs (labeled “A” and “B”) were issued with zero lead time. Over 40% of tornado warnings from one WFO (“C”) were issued after tornado formation. Indeed, the tornado warning statistics at these three WFOs differ markedly from those at other forecast offices. Parker and Waldron (2002) and Wolf (2002) demonstrate that WFO warning operations can significantly impact warning results.

c. Diurnal climatology

What is the impact of time of day on the issuance of tornado warnings? The numbers of tornado warnings with positive, zero, and negative lead times were plotted by hour (Fig. 4a). For easier interpretation, hourly percentages were subtracted from the daily zero and negative mean percentages (4.2% and 6.1%, respectively) as shown in Fig. 4b.

Fig. 4.

(a) Number of tornado warnings issued each hour with positive, zero, and negative lead times. (b) Percent deviation from the daily average percentage of warnings with zero or negative lead-times plotted as a function of hour of day.

Fig. 4.

(a) Number of tornado warnings issued each hour with positive, zero, and negative lead times. (b) Percent deviation from the daily average percentage of warnings with zero or negative lead-times plotted as a function of hour of day.

Overall, one clear trend emerges. Above-average numbers of zero and negative lead-time warnings are issued between 1300 and 1800 local time (LT) and below-average numbers of negative lead time warnings are issued between 1800 and 0200 LT. This trend likely captures the difficulty in warning on the first tornado of the day. This is explored in more detail in section 3h.

d. Seasonal climatology

As with the diurnal cycle, the seasonal cycle in storm climatology may affect the predictability and detectability of tornadoes and, thereby, impact tornado warning lead times. The percentages of all tornado warnings with zero or negative lead times were plotted as a function of month (Fig. 5a). Nationwide, warned tornadoes with zero or negative lead times peaked during July and August with a relative minimum observed during May. To better discern the causes driving the monthly differences, the data were sorted by region and again plotted as a function of month (Fig. 5b). Note that the percentages are computed from the total number within each region, such that the percentage within each geographic region totals 100%. Tornadoes with zero and negative lead-time warnings peaked in the southeast and west regions in July and August, the peak in the midwest was observed in July followed by June, and in the plains region the peak was observed in September, followed by July and April. Most regions indicated relative minima in tornadoes with zero and negative lead-time warnings during May.

Fig. 5.

(a) Percentage of all warned tornadoes with zero or negative lead times plotted as a function of the month of the year. (b) Same as in (a) but sorted by geographic region. Note that regional percentages are computed from the total number within each region, such that the percentage within each geographic region totals 100%.

Fig. 5.

(a) Percentage of all warned tornadoes with zero or negative lead times plotted as a function of the month of the year. (b) Same as in (a) but sorted by geographic region. Note that regional percentages are computed from the total number within each region, such that the percentage within each geographic region totals 100%.

The distribution of all zero and negative lead times was plotted as a function of month (Fig. 6). No data from any month with fewer than 10 negative lead-time events were shown. In general, the early spring and fall months had a greater number of large negative lead-time events. Ironically, the month with by far the greatest number of tornadoes (May) had among the lowest mean negative lead time. Using a Wilcoxon–Mann–Whitney rank sum test, May lead times were significantly lower than those from March, April, June, August, and October at the 90% confidence level; no other month was significantly different from any other month.

Fig. 6.

Box plots of all zero and negative lead times plotted as a function of the month of the year.

Fig. 6.

Box plots of all zero and negative lead times plotted as a function of the month of the year.

e. Impact of storm morphology

A subsample of 110 positive and 110 negative lead-time events was classified by storm type to better understand the impact of storm morphology on tornado warning lead times. A semirandom sample was chosen; every 5th event was classified in the zero and negative lead-time dataset, while every 40th event was classified from the positive lead-time warning dataset. Each event was manually reviewed using mosaic radar data (see the Web site http://www.mmm.ucar.edu/imagearchive/ for an example), or when not available, level 2 and level 3 radar data archived at the National Climatic Data Center were used. As described in section 2, each storm event was classified as line, cell, tropical (hurricane, tropical storm), or undefined, as described by Gallus et al. (2008). Broken lines of individual storm cells were classified under the cell heading; line events were generally solid convective squall lines. Undefined events were primarily stratiform regions with embedded convective elements. Storm classification was valid only at the time of tornado formation; many storms were in the process of evolving from clusters or lines of cells to more solid convective lines.

Little noticeable difference was found in storm type between those tornado warnings with positive lead time and tornadoes with zero or negative lead time. Positive (negative) lead-time warnings classified as cells, lines, tropical, and undefined were 48%, 34%, 8%, and 10% (57%, 31%, 9%, and 3%), respectively, of the total subset. In summary, there was no statistical relationship found between storm type and tornado warning lead time.

The classified storms also were sorted by geographic region (Fig. 7). The southeast region was dominated by linear events whereas the plains and west regions were dominated by cell-based events. The midwest region had more lines than cells in the positive lead-time sample, but more cells than line events in the negative lead-time sample. Overall, there were no clear differences between those storm types with positive and negative lead times. A greater number of cell-based tornadic storms in the plains and west regions could partially explain the higher positive lead times in those regions.

Fig. 7.

(a) A subsample of 110 tornado warnings with positive lead time classified by storm type and geographic region. (b) A subsample of 110 tornado warnings with either zero or negative lead time classified by storm type and geographic region.

Fig. 7.

(a) A subsample of 110 tornado warnings with positive lead time classified by storm type and geographic region. (b) A subsample of 110 tornado warnings with either zero or negative lead time classified by storm type and geographic region.

All tornado events classified by storm type were sorted by month (Fig. 8) to discern any impact on seasonal variations with tornado warning lead time as discussed in section 3d. No clear relationship was evident between the fraction of negative lead-time warnings and the dominant storm type. Quasi-linear systems were more prevalent during March and April and could be responsible for the higher negative lead times during those months. Tropical systems dominated the tornado statistics during September but, in general, were not responsible for many of the highest negative lead times observed during that month.

Fig. 8.

A subsample of 220 of all of the warned tornado events classified by storm morphology sorted by month.

Fig. 8.

A subsample of 220 of all of the warned tornado events classified by storm morphology sorted by month.

f. Impact of population

Spotter reports are a valuable tool used by forecasters in evaluating the development and severity of a storm and are a critical component to the decision-making process for tornado warning issuance. In general, it is assumed that the greater the population density of an area, the greater the chance that any tornadoes that develop will be reported to the WFO. If true, we can expect smaller negative lead-time warnings within more densely populated areas.

To test this assumption, tornado warnings with zero and negative lead times were plotted against the county population density in which the tornado began (Fig. 9). No statistical significance is found relating negative lead time and county population density.

Fig. 9.

Box plots showing the mean and range of county population density (persons per square kilometer) plotted as a function of negative lead time (min).

Fig. 9.

Box plots showing the mean and range of county population density (persons per square kilometer) plotted as a function of negative lead time (min).

g. Impact of distance from radar

As with county population density, no statistical relationship could be found between negative lead times and the distance of the tornadoes from radar (Fig. 10). However, the radar-to-tornado distance appears to play some role in at least a few events, as shown in Table 1. Many studies have shown a direct relationship between tornado reports and distance from radar (e.g., Ray et al. 2003). The current study has not focused on unwarned tornadoes, and it is possible that this type of study may show a much greater impact of radar coverage on warning performance. We also did not consider the impact of the elevation of the radar on tornado warning performance, which would be expected to have some influence in high-terrain regions.

Fig. 10.

Box plots showing the mean and range of tornado distance (km) from the nearest WSR-88D plotted as a function of negative lead time (min).

Fig. 10.

Box plots showing the mean and range of tornado distance (km) from the nearest WSR-88D plotted as a function of negative lead time (min).

h. High-impact events

Does the number of tornado warnings issued in a day by a WFO impact the warning lead time? Is the first tornado report of the day more difficult to warn in advance? Are certain geographic regions more prone to experiencing tornado outbreaks (multiple tornadoes within a WFO county warning area per day), and if so, how does this impact the tornado warning lead time? To address these questions, the tornado warning data were parsed by the number of tornado warnings per day, per hour, and the order of occurrence.

All tornado warning data were separated by the number of confirmed tornado warnings per day (Table 3). A total of 5170 confirmed tornado warnings were issued during the 5 yr of study. Of these, 14.9% (772) of the warnings were the only confirmed tornado warnings issued by a WFO during that calendar day. Another 13.3% of warnings were associated with two warnings per day, 10.6% of warnings were associated with three warnings per day, and the remainder (61.2%) of warnings was associated with days during which four or more confirmed tornado warnings were issued by a single WFO during one calendar day.

Table 3.

Tornado warning statistics and lead times listed as a function of the number of confirmed tornado warnings per day per WFO.

Tornado warning statistics and lead times listed as a function of the number of confirmed tornado warnings per day per WFO.
Tornado warning statistics and lead times listed as a function of the number of confirmed tornado warnings per day per WFO.

Interestingly, the ratio of warnings with zero or negative lead time decreases with increasing number of verified tornado warnings per day; in other words, the more warned tornadoes per day, the smaller the ratio of zero or negative lead-time warnings. As shown in Table 3, the percentage of warnings with negative lead time decreases from 21.6% of days with one confirmed warning to only 2.1% for days with 20 or more confirmed warnings. In part because there are proportionally fewer zero and negative lead-time warnings on high tornado warning days, the average tornado warning lead time increases from 11.8 min with a single tornado warning per day to 16+ min for tornadoes on days with four or more confirmed warnings. In an effort to minimize false alarms, operational forecasters are likely reluctant to issue that first tornado warning of the day until they are given sufficient evidence that an ongoing storm is capable of producing a tornado.

The lead time of warnings with only positive lead time also jumps; lead time increases from ∼16 min for days with one to three warned tornadoes to ∼19 min for days with four or more confirmed tornado warnings per day per WFO. One possible reason for this increase in positive lead time with number of tornado warnings per day is the issuance of warnings downstream for ongoing tornadic storms. These “continuance” warnings likely have a much greater warning lead time, although this could not be verified with this particular dataset. Surprisingly, very little discussion of the impact of continuance warnings on tornado warning lead-time statistics is found in the meteorological literature. A second possible reason for the increase in positive lead time is that weather conditions that are able to produce multiple tornadoes per day within a WFO County Warning Area (CWA; or immediately downstream across neighboring NWS CWAs) are likely stronger and more organized and therefore easier to anticipate and detect than more marginally severe systems (e.g., Guillot et al. 2008).

The impact of tornado order on warning lead time was examined. Warning and lead-time statistics were calculated from all verified tornado warnings issued by a WFO on days with four or more tornado warnings (Table 4). The first confirmed tornado warning of the day has a high (19.5%) ratio of having a zero and negative lead time. However, the ratio of zero and negative lead time warnings remains relatively steady after and including the second tornado of the day. The average of positive lead-time warnings increases gradually with tornado order from 16.4 min for the 1st tornado of the day to ∼22 min for the 10th tornado of the day. The average of negative lead-time warnings remains steady regardless of tornado occurrence order. As discussed previously, the first tornado of the day is the most difficult on which to warn, as evident by the high ratio of zero and negative warnings. After the first tornado of the day, however, it is primarily the increased positive lead time that dominates the lead-time average. Tornadoes warned in succession are more likely to be warnings on tornadoes already in progress, whereas the first tornado warning of the day is likely a warning based on the anticipation of what will occur. Bieringer and Ray (1996) also found a much lower warning lead time for the first tornado events of the day.

Table 4.

Tornado warning statistics and lead time as a function of the order of tornado occurrence for days with four or more confirmed tornado warnings issued by a single WFO in one calendar day (0000–2359 LT).

Tornado warning statistics and lead time as a function of the order of tornado occurrence for days with four or more confirmed tornado warnings issued by a single WFO in one calendar day (0000–2359 LT).
Tornado warning statistics and lead time as a function of the order of tornado occurrence for days with four or more confirmed tornado warnings issued by a single WFO in one calendar day (0000–2359 LT).

Could the number of tornadoes per day or tornado order have a diurnal impact on lead time? As discussed in section 3c, a review of all warned tornadoes showed an above-average percentage of negative lead-time warnings between 1300 and 1800 LT and below-average percentage between 1800 and 0200 LT (Fig. 11a). A review of all single tornado days showed a very similar pattern with an above-average percentage of negative lead time warnings issued between 1400 and 1700 LT with a below-average percentage of negative lead time warnings issued between 1700 and 2200 LT (Fig. 11b). However, a review of tornadoes from days with four or more tornadoes showed no discernable diurnal pattern (Figs. 11c and 11d). From these data, the diurnal pattern appears limited to weaker storm systems (i.e., those systems with less than four tornadoes per day). Furthermore, the diurnal pattern does not appear to be a function of day versus night. An above-average percentage of negative lead time events is recorded prior to 1700–1800 LT, and a below-average percentage is recorded after this time. For weakly forced events, perhaps forecasters simply have a much lower expectation of tornadogenesis during the early afternoon than during the evening. Nevertheless, the exact cause of the diurnal pattern is unknown.

Fig. 11.

Hourly deviation from the daily percentage of events with negative lead time. Only those hours with 20 or more observations are plotted. Dataset includes (a) all data, (b) days with one tornado, (c) the first tornado from days with four or more tornadoes, and (d) all tornadoes after and including the fourth tornado of the day from days with four or more tornadoes.

Fig. 11.

Hourly deviation from the daily percentage of events with negative lead time. Only those hours with 20 or more observations are plotted. Dataset includes (a) all data, (b) days with one tornado, (c) the first tornado from days with four or more tornadoes, and (d) all tornadoes after and including the fourth tornado of the day from days with four or more tornadoes.

The number of tornadoes associated with a given tornado day may be expected to change with season. The average number of tornado warnings issued by a WFO per tornado day was calculated monthly (Fig. 12a). Typically, those months with more isolated, single tornado days (e.g., August) were found to have the highest proportion of zero and negative lead-time warnings (Fig. 12b). Thus, much of the seasonal variability in the number of zero and negative lead-time warnings can be directly attributed to the number of multitornado events per month.

Fig. 12.

(a) Monthly average of the number of confirmed tornado warnings issued per tornado day. (b) Monthly average of the number of confirmed tornado warnings issued per tornado day plotted as a function of the percentage of warnings with zero or negative lead times.

Fig. 12.

(a) Monthly average of the number of confirmed tornado warnings issued per tornado day. (b) Monthly average of the number of confirmed tornado warnings issued per tornado day plotted as a function of the percentage of warnings with zero or negative lead times.

To determine the relationship between the number of tornadoes per day and geography, the number of confirmed tornado warnings per day (with either positive or negative lead time) was calculated as a function of geographic region (Fig. 13). Over 60% of tornado days in the West region are single, isolated tornado events, followed by just over 50% of tornado days in the midwest region. Given that negative lead time is more often associated with isolated, single-tornado events, we can expect these regions to have slightly lower average warning lead times when compared to other regions. A review of Table 2 shows that indeed the west and midwest regions have the lowest average lead times. Contrarily, the plains and southeast regions, which experience many more multiple-tornado events, have a much higher overall tornado warning lead time.

Fig. 13.

The percentage of confirmed tornado warnings per day estimated and plotted as a function of geographic region.

Fig. 13.

The percentage of confirmed tornado warnings per day estimated and plotted as a function of geographic region.

The average number of confirmed tornado warnings per tornado day also was calculated for each NWS WFO. Then, each WFO average was plotted against the percentage of warned tornadoes with zero and negative lead times (Figs. 14a and 14b). An examination of zero lead-time warnings shows only a slight negative trend with the average number of confirmed warnings per tornado day. However, a strong correlation exists between the percentage of negative lead-time warnings and the average number of confirmed tornado warnings per tornado day. In other words, those WFOs with the greatest number of multitornado events are likely to have a much lower ratio of negative lead-time tornado warnings.

Fig. 14.

Percentage of warned tornado events with (a) zero and (b) negative lead times plotted against the average number of confirmed tornado warnings issued per WFO per tornado day.

Fig. 14.

Percentage of warned tornado events with (a) zero and (b) negative lead times plotted against the average number of confirmed tornado warnings issued per WFO per tornado day.

Next, the tornado warning data were separated by the number of confirmed tornado warnings per hour per WFO (Table 5). Statistics indicate that the ratio of tornado warnings with zero or negative lead time decreases with increasing numbers of tornado warnings per hour. As a result, the average lead time increases from 14.3 min for one tornado warning per hour to ∼17 min for three or more tornado warnings per hour. There appears to be a difference between one or two tornado warnings per hour and three or more warnings per hour; however, there is little difference in the statistics within each group (one to two warnings versus three or more warnings).

Table 5.

Tornado warning statistics and lead times listed as a function of the number of confirmed tornado warnings per hour per WFO.

Tornado warning statistics and lead times listed as a function of the number of confirmed tornado warnings per hour per WFO.
Tornado warning statistics and lead times listed as a function of the number of confirmed tornado warnings per hour per WFO.

During the 5 yr of study, 16 tornado “outbreaks” of three or more tornadoes occurred in which a WFO issued three or more warnings during a single weather event, each with either zero or negative lead time (Table 6). A review of these cases points to several commonalities among these events. First, many of these tornadoes occurred in rapid succession, often times with two or more tornadoes occurring within the same hour and, sometimes, even simultaneously. Seven of the events had three or more tornadoes occur within 1 h and with zero or negative lead time. Second, seven of the events were associated with outbreaks where the WFO had to issue 10 or more tornado warnings within a short period of time. Under such conditions, even the best WFO operations could become overwhelmed. Third, half of the events occurred on the weekend, almost twice the number expected by chance. Subtle differences in office management and staffing at the WFO could have a considerable effect during high-impact outbreak situations (Waldstreicher 2005).

Table 6.

List of all tornado outbreaks with three or more tornado warnings all with zero or negative lead times issued from the same WFO.

List of all tornado outbreaks with three or more tornado warnings all with zero or negative lead times issued from the same WFO.
List of all tornado outbreaks with three or more tornado warnings all with zero or negative lead times issued from the same WFO.

4. Relationship between negative lead time and fatalities

Overall, tornadoes with zero or negative lead-time warnings account for 10.9% of all tornado warnings and are associated with 8.5% of all tornado fatalities. During the 5 yr of study, there were 12 fatal tornadoes with zero or negative lead-time warning with a total of 17 fatalities. A review of these events (Table 7) reflects several difficulties in the warning process. First, three-fourths of the events took place either late at night and/or on the weekend, when the public was least likely to hear and act upon a warning. Second, some storm types remain challenging, especially tornadoes spawned from hurricanes; in this case, such storms were responsible for 5 of the 17 deaths. The “undefined” case in Texas also posed a challenge, with two isolated tornadoes occurring in an otherwise mostly benign, heavy rain event. No initial advance warning was given, despite the short distance from radar in a heavily populated region.

Table 7.

List of all fatal tornadoes with zero or negative lead-time warnings.

List of all fatal tornadoes with zero or negative lead-time warnings.
List of all fatal tornadoes with zero or negative lead-time warnings.

5. Conclusions

A comprehensive review of NWS tornado warning statistics from 2000 to 2004 examined in detail those warnings with zero or negative lead time. Zero and negative lead-time warnings were sorted by F-scale rating, geography, WFO, time of day, month, storm type, county population density, distance from radar, and tornado order during multiple tornado events. Our results are summarized as follows:

  • Providing advance warning for the first tornado of the day remains a difficult challenge. Diurnal trends show that the hours during the early afternoon, primarily during the time when the first tornadic storms are developing, have the highest ratio of negative lead-time warnings (Fig. 4). Statistics indicate the first tornado of the day has 3 times the proportion of zero or negative lead-time warnings than does the second tornado of the day for days with four or more tornado warnings within a WFO county warning area (Table 4). In addition, likely due to the inclusion of “continuance” warnings, the national average of positive lead-time warnings increases with each successive tornado warning per day.

  • In general, the more isolated the tornado event, the less is the likelihood that an advance warning is provided. Isolated, single tornado per day events have 10 times the ratio of zero and negative lead-time warnings than do days with 20 or more tornadoes (Table 3). Average lead time increases from 11.8 min for single-tornado warning days to 19.3 min for days with 20 or more confirmed tornado warnings within a given WFO county warning area.

  • Monthly and geographic trends in lead time are directly impacted by the number of multiple tornado events. Because lead times are skewed by continuance warnings, those months during which tornado outbreaks are most common (e.g., May) have a much lower proportion of zero and negative lead-time warnings (Figs. 5a and 12). Geographic regions, such as the plains and southeast, where multiple-tornado events are more common, are observed to have greater average warning lead times (Table 2, Fig. 13). WFOs that experience many large-scale outbreaks have a lower proportion of warnings with negative lead time than WFOs that experience many more isolated, one- or two-tornado warning days (Fig. 14).

  • During a tornado “outbreak” several tornadoes can be on going simultaneously, leading to a stressful situation for NWS operations (Andra et al. 2002). However, on average these situations do not negatively impact tornado warning lead times nor do they lead to an increase in the ratio of zero or negative lead-time events (Table 5), with some notable exceptions (Table 6). Staffing shortages may arise during some unanticipated tornado outbreak situations.

  • A clear impact of storm morphology on tornado warning lead times was difficult to discern from this 5-yr dataset. No direct relationship was found between storm type and the ratio of zero or negative lead-time warnings (Fig. 7), and no statistical relationship could be found between storm type and tornado warning lead time. It is possible that the higher proportion of linear and tropical systems in the midwest and southeast regions could be negatively impacting positive tornado warning lead times (Fig. 7, Table 2). Some isolated cases (Fig. 1, Tables 6 and 7) were shown to pose some unique challenges for the forecaster, particularly from tropical systems and less-organized convection. Gallus et al. (2008) show a relationship between storm morphology and the type of severe weather produced. Other studies have shown some impact of storm type on tornado warning lead time. Trapp et al. (1999) found tornadoes resulting from nondescending vortices, often associated with quasi-linear convective systems, were associated with shorter warning lead times. Guillot et al. (2008) found higher tornado warning lead times for isolated supercells and strong convective lines but less lead time for weaker, less-organized systems. However, stronger systems are more likely to produce multiple-tornado events; in this case, tornado warning lead times for strong events may be artificially enhanced by including more continuance warnings.

The National Weather Service vision for 2025 is to provide tornado warning lead times of 45 min or greater (J. Hayes 2007, personal communication). The conclusions from this paper identify those situations and times that are most vulnerable to failure in the current tornado warning process. If we are to continue to improve tornado warning lead times, special attention should be given to two specific areas: (i) warning on the first tornado of the day (or rather, first tornado from a given storm complex) and (ii) warning on nontypical (weak and/or isolated) severe storms. Simply raising awareness of these “gaps” in the warning process is a critical first step in improving tornado warning operations. Similar work is needed to examine the causes of no-warn events and false alarms.

One short-term solution for improving our understanding of the tornado warning process is to create a detailed metadata archive to accompany all NWS tornado warnings. Such information should include items such as why each tornado warning was issued, on what criteria the warning was based, and whether or not the warning was a continuance warning of a tornado already in progress. Such details could yield long-term dividends in understanding and improving tornado warning statistics.

The long-term solution to improving tornado warning lead times will ultimately depend upon our ability to provide “warn on forecast.” As shown in this manuscript, our inability to predict the first tornado of the day greatly limits our overall tornado lead time. The use of additional observing tools along with improved conceptual models, statistical nowcasting techniques, data-mining algorithms, or some blending with numerical weather prediction models will likely be needed.

Finally, it is noteworthy that despite the delay in issuance of the tornado warnings discussed in this manuscript, over 80% of these zero and negative lead-time warnings still provided valuable positive lead times for those downstream in the path of the ongoing storm. Of these warnings, the tornado remained on the ground on average for another 7 min, providing critical life-saving information to the general public.

Acknowledgments

We thank Brent Macaloney at NWS Headquarters for supplying us with the tornado record data used in this study, and Harold Brooks, Rodger Brown, Michael Richman, Liz Quoetone, and three anonymous reviewers for excellent suggestions to improve this manuscript. Thanks to James Hocker for his GIS expertise. This work is supported by the Engineering Research Centers Program of the National Science Foundation under NSF Award 0313747. Any opinions, findings, conclusions, or recommendations expressed in this material are those of the authors and do not necessarily reflect those of the National Science Foundation.

REFERENCES

REFERENCES
Andra
,
D. L.
,
E. M.
Quoetone
, and
W. F.
Bunting
,
2002
:
Warning decision making: The relative roles of conceptual models, technology, strategy, and forecaster expertise in 3 May 1999.
Wea. Forecasting
,
17
,
559
566
.
Bieringer
,
P.
, and
P. S.
Ray
,
1996
:
A comparison of tornado warning lead times with and without NEXRAD Doppler radar.
Wea. Forecasting
,
11
,
47
52
.
Crum
,
T. D.
, and
R. L.
Alberty
,
1993
:
The WSR-88D and the WSR-88D Operational Support Facility.
Bull. Amer. Meteor. Soc.
,
74
,
1669
1687
.
Erickson
,
S.
, and
H.
Brooks
,
2006
:
Lead time and time under tornado warnings: 1986–2004.
Preprints, 23rd Conf. on Severe Local Storms, Atlanta, GA, Amer. Meteor. Soc., 11.5. [Available online at http://ams.confex.com/ams/23SLS/techprogram/paper_115194.htm.]
.
Friday
Jr.,
E. W.
,
1994
:
The modernization and associated restructuring of the National Weather Service: An overview.
Bull. Amer. Meteor. Soc.
,
75
,
43
52
.
Gallus
Jr.,
W. A.
,
N.
Snook
, and
E. V.
Johnson
,
2008
:
Spring and summer severe weather reports over the Midwest as a function of convective mode: A preliminary study.
Wea. Forecasting
,
23
,
101
113
.
Guillot
,
E.
,
V.
Lakshmanan
,
T.
Smith
,
G.
Stumpf
,
D.
Burgess
, and
K.
Elmore
,
2008
:
Tornado and severe thunderstorm warning forecast skill and its relationship to storm type.
Preprints, 24th Int. Conf. on Interactive Information Processing Systems for Meteorology, Oceanography, and Hydrology, New Orleans, LA, Amer. Meteor. Soc., 4A.3. [Available online at http://ams.confex.com/ams/pdfpapers/132244.pdf.]
.
Magsig
,
M. A.
,
T.
Decker
, and
N. M.
Said
,
2006
:
Builds five and six of NOAA’s NWS weather event simulator.
Preprints, 22nd Int. Conf. on Interactive Information Processing Systems for Meteorology, Oceanography, and Hydrology, Atlanta, GA, Amer. Meteor. Soc., P7.7. [Available online at http://ams.confex.com/ams/pdfpapers/104624.pdf.]
.
Mitchell
,
E. D.
,
S. V.
Vasiloff
,
G. J.
Stumpf
,
A.
Witt
,
M. P.
Eilts
,
J. T.
Johnson
, and
K. W.
Thomas
,
1998
:
The National Severe Storms Laboratory tornado detection algorithm.
Wea. Forecasting
,
13
,
352
366
.
NWS
,
cited
.
2007
:
NOAA’s NWS national performance measures FY 2006–FY National Weather Service.
.
Parker
,
S. S.
, and
H.
Waldron
,
2002
:
Dramatically improving warning services—One office’s experience.
Preprints, 21st Conf. on Severe Local Storms, San Antonio, TX, Amer. Meteor. Soc., P9.5. [Available online at http://ams.confex.com/ams/pdfpapers/46785.pdf.]
.
Rasmussen
,
E. N.
,
J.
Straka
,
R.
Davies-Jones
,
C.
Doswell
III
,
F.
Carr
,
M.
Eilts
, and
D.
MacGorman
,
1994
:
The Verification of the Origins of Rotation in Tornadoes Experiment: VORTEX.
Bull. Amer. Meteor. Soc.
,
75
,
997
1006
.
Ray
,
P.
,
P.
Bieringer
,
X.
Niu
, and
B.
Whissel
,
2003
:
An improved estimate of tornado occurrence.
Mon. Wea. Rev.
,
131
,
1026
1031
.
Seguin
,
W.
,
2002
:
AWIPS—An end-to-end look.
Preprints, Interactive Symp. on the Advanced Weather Interactive Processing System (AWIPS), Orlando, FL, Amer. Meteor. Soc., 147–150
.
Simmons
,
K. M.
, and
D.
Sutter
,
2005
:
WSR-88D radar, tornado warnings, and tornado casualties.
Wea. Forecasting
,
20
,
301
310
.
Stumpf
,
G. J.
,
A.
Witt
,
E. D.
Mitchell
,
P. L.
Spenser
,
J. T.
Johnson
,
M. D.
Eilts
,
K. W.
Thomas
, and
D. W.
Burgess
,
1998
:
The National Severe Storms Laboratory mesocyclone detection algorithm for the WSR-88D.
Wea. Forecasting
,
13
,
304
326
.
Trapp
,
R. J.
,
E. D.
Mitchell
,
G. A.
Tipton
,
D. W.
Effertz
,
A. I.
Watson
,
D. L.
Andra
, and
M. A.
Magsig
,
1999
:
Descending and nondescending tornadic vortex signatures detected by WSR-88Ds.
Wea. Forecasting
,
14
,
625
639
.
Waldstreicher
,
J. S.
,
2005
:
Assessing the impact of collaborative research projects on NWS warning performance.
Bull. Amer. Meteor. Soc.
,
86
,
193
203
.
Wolf
,
P. L.
,
2002
:
Toward better warning decision-making and verification statistics: Improvements at the Wichita National Weather Service office.
Preprints, 21st Conf. on Severe Local Storms, San Antonio, TX, Amer. Meteor. Soc., 10B.4. [Available online at http://ams.confex.com/ams/SLS_WAF_NWP/techprogram/paper_46960.htm.]
.

Footnotes

Corresponding author address: Jerald Brotzge, University of Oklahoma, 120 David L. Boren Blvd., Suite 2500, Norman, OK 73072-7309. Email: jbrotzge@ou.edu