Joseph Golden
Senior Meteorologist
Silver Spring, MD

Tornado Climatology

The tornado season tends to begin in the Gulf Coast states during the late winter and early spring, and migrates northward to the Great Lakes and Canadian border by late June and July. Many of the tornadoes that affect the central and eastern Gulf Coast during the late fall and early spring originate over the northern Gulf of Mexico as intense waterspouts. Bluestein and Golden (1993) noted that many of these do not have hook echoes or other classic, distinguishing characteristics on radar. Many of the increasing number of tornadoes being reported along the U.S. West Coast (Monteverdi, 1994), particularly the greater Los Angeles basin (Hales), since the deployment of NEXRAD Doppler radars originate offshore as waterspouts in the presence of an intense upper cold low.

The most definitive study of U.S. tornado climatology from 1680-1991 was published in a large volume by Grazulis (1991). Tornadoes are most likely over the southern and central Plains, southern Mississippi Valley and upper Midwest. It is apparent that the risk of tornadoes is considerably less west of the Rockies and into southern Canada. However, the U.S. averages about 100 tornadoes per year and is the world's capital (Fujita, 1974). Xu (1993) reported on an 11-year study of tornadoes in China, and found several over the plains of the middle and lower Yangtze River, with a peak frequency in July. Tornadoes are also found inland from both southwest and east coasts of Australia and over the plains of Argentina.

The U.S. is the world's tornado capital because of the unique juxtaposition of the north-south Rocky Mountain range and the moisture source from the Gulf of Mexico. Models show that the Rockies allow the inflow of warm, moist air from the Gulf of Mexico to be overlain by colder air flowing over it from the west. These jet stream disturbances are often associated with explosive cyclogenesis and attendant severe weather. Similar geography does not exist elsewhere; for example in China, the elevated Tibetian Plateau is oriented east-west, and there is no comparable moisture source to the south like the Gulf of Mexico.

The primary data source for all climatological records of severe storms, including hurricanes and tornadoes, is the monthly STORM DATA, published by NOAA/NESDIS. It tabulates daily occurrences of all reported severe storm phenomena, by state and date. Most of these reports are gathered at local NWS offices, from both official sources and storm spotters. However, Burgess and Doswell (1988) noted that with decreasing NWS staff and storm surveys by untrained individuals, the number and characteristics of tornadoes (intensity, path length and path width) are becoming increasingly suspect and unreliable. Grazulis (1992, 1993) examined not only tornadoes in STORM DATA, but also visited local NWS offices and libraries nationwide to clean up the U.S. tornado record; he found that about 2000 tornadoes reported out of about 10,000 were improperly designated tornadoes since the late 1880's.

The thirty-year average yearly deaths from flash floods/floods, lightning, tornadoes and hurricanes is shown in Fig. 1. We note that tornadoes are not the leading cause of severe storm fatalities; indeed, the greatest average yearly threat is posed by floods and flash floods, followed by lightning and tornadoes. It might be argued that this situation has partly resulted from the significant improvement in the accuracy and lead-times for tornado and hurricane warnings concomitant with the NWS Modernization (Sheets, 1990; Polger et al, 1995). The trends in U.S. annual tornado frequency since 1953, broken down by intensity, are shown in Fig. 2, after Ostby (1993). The three intensity categories are weak (F0, F1), strong (F2, F3) and violent (F4, F5), as well as for all intensities combined. Most striking is the dramatic rise in reported tornado totals since the mid-1980's. We attribute this rapid rise primarily to the increased reports of weaker tornadoes that in turn have been the result of population increases, increased NWS emphasis on warning verification, and the NWS Modernization ("beating the bushes" by post-storm phone calls to spotters and local officials - see Hales, 1985, 1993), storm chaser reports and eyewitness videos. This figure also points to a major shift in the tornado intensity distribution: During the 1960's and 1970's, studies indicated that about two-thirds of all U.S. yearly tornadoes were weak, one-third strong, and only about 2% violent, these newer data suggest that about 85% of reported tornadoes are weak, 15% strong and less than 1% violent (shown in Fig. 2). Most important, recent studies by Monteverdi et al. on California tornadoes suggest that there may still be significant underreporting of tornadoes west of the Continental Divide.

The recent year-to-year trends in tornado losses are shown in Table 1, which shows data compiled by the NWS/OM from STORM DATA from 1985-95.

Table 1 - U.S. Tornado Impact Statistics

#Deaths In mobile
#Injuries Estimated
1985 94 30%
1986 15 47%
1987 59 41%
1989 50 24% 2490 $1088M
1990 53 13% 1150 $668M
1991 39 51% 854 $798M
1992 39 51% 1300 $764M
1993 33 39% 3990 $368M
1994 69 38% 1067 $518M
1995 30 647 $408M
Source: NWS/OM

The columns show deaths, percentage of fatalities in mobile homes, injuries and dollars of estimated damage in millions. While tornado damage in 1989 was over one billion dollars, it has declined somewhat since that year. However, the Ft. Smith, AK tornado in the last year alone caused $300 million in damage, and the recent major outbreak again in Arkansas may well exceed that figure. Table 2 gives total insured property losses from 1986-1995, based on estimates from Property Claims Services for all wind, hail, snow and tornado losses combined.. These claims have risen from under one billion dollars in the late 1980's to slightly over five billion dollars in 1992, and nearly that amount in 1995 as well.

Table 2- 1986-1995 Wind, Hail, Snow and Tornado Losses

Note: Losses are based on estimates from the property claims Services (PCS) in New York. PCS losses account for property insured under fire and allied lines coverages, inland marine fixed property coverages, and comprehensive automobile coverages.
Year Wind, Hail, Snow & Tornadoes
1986 $849,247,500
1987 $875,000,000
1988 $1,314,000,000
1989 $1,777,000,000
1990 $2,545,000,000
1991 $2,378,000,000
1992 $5,055,000,000
1993 $4,355,000,000
1994 $4,306,000,000
1995 $4,850,000,000
Total $28,304,247,500

Most interesting by contrast in Table 1 is the fact that after 1985, the yearly tornado deaths were well below the 30-yr average shown in Fig. 2. There is no discernible trend in either tornado deaths or injuries, although 1989 was a bad year for both injuries and damages. Another interesting point is that there are generally 40-50% of yearly tornado deaths occurring for occupants of mobile homes. This comes at a time when the only affordable housing for a growing percentage of American disadvantaged and the elderly is manufactured housing (e.g., mobile homes). A striking example was the tornado outbreak across Florida and the Southeastern U.S. documented by the NOAA Survey Team Report on the March 13-14, 1993 Superstorm. We should stress that while the risks to occupants of manufactured housing from all types of severe windstorms appears to be rising, there are mitigation measures that can be taken. Most important among these, which arose after the many fatalities in the large mobile home park devastated by the 1991 Andover, KS tornado, is the requirement for a fixed, permanent shelter at every mobile home park in wind-prone areas. The wind engineering community has examined the manufactured home construction and recommended better windstorm protection through improved anchoring systems (Marshall, 1994). After detailed examination of the damage to permanent and mobile home construction in Miami after Hurricane Andrew, Marshall (1993) found that damage to manufactured homes ranged from loss of roofing to total destruction. In general, units that conformed to the HUD safety standards suffered less damage than did pre-HUD units. However, conventional residential construction adjacent to manufactured home parks performed better, in some instances significantly better, than did manufactured homes, including HUD-labeled units. Based on Marshall's (1993) windspeed assessments and damage to roof and wall coverings at fastest-mile speeds of up to 42.5 m/s and significant structural damage at wind speeds ranging 44.7 to 53.6 m/s. At wind speeds ranging from 53.6 m/s to 60.3 m/s, there were numerous instances of HUD-labeled units suffering total destruction. Marshall used these results to recommend some improvements to the HUD wind load requirements in its construction and safety standards for manufactured housing.

The trends in NWS warnings since the first deployment of NEXRAD Doppler radars have been gratifying. Lead-times for tornado warnings up to the early 1980's had been negative, that is, the tornado had already touched down and been producing damage on the ground for several minutes by the time the first NWS warning was issued. Beginning around 1990, and coincident with the first NEXRAD radar deployment at Melbourne, FL, the NWS warning lead-times have been positive and increasing steadily. The latest NWS warning statistics for 1996 are shown in the Table, and while these are preliminary figures, suggest tornado warning lead-times averaging over 18 minutes, a CSI of 0.46, a POD of 0.826, and false alarm ratio (FAR) of 0.489. The first three figures have all improved steadily since 1990. However, the FARs have slightly increased over the past few years, after dropping during the early part of the decade. The reasons for more NWS false alarms are not clear, but may relate to newer forecasters replacing more experienced ones who have retired and being somewhat overzealous with the new technology.

The trends above in tornado climatology and warnings suggest that, overall, things seem to be improving. Indeed, the recent major tornado outbreak over Arkansas on March 1, 1997 was well-warned, lead-times ranging from 18-35 minutes. However, there are still significant tornado forecast and warning challenges, highlighted by the major tornadoes with 44 deaths in the Southeast during the 1994 Palm Sunday outbreak, the Tulsa, OK tornado, and the Ft. Smith, AK tornado last year. Another problem may be lurking with the recent rise in false-alarm rates. There are potential social response issues as well that the increased warning lead-times may cause: The AP reported that after being warned of the tornado that struck Arkadelphia, AK, 20 minutes later, young Katie Kolb did the once-unthinkable act: she traveled into its path to retrieve her sister. There are also increasing numbers of people putting themselves in harm's way by chasing tornadoes with video cameras or by watching and waiting too long after being warned at home. A new controversy has arisen in the literature lately by Schmidlin and King (1996, BAMS) who question current NWS recommendations to mobile home residents to seek permanent shelter or lie in a culvert when warned of an approaching tornado. These authors question such advice and suggest instead that it may be safer to outrun the tornado in an automobile. This needs rational thought and debate, and much more scrunity of the evidence, pro and con. The fact remains that a large fraction of the total deaths in the Wichita Fall, TX tornadoes of 1979 and Huntsville, AL in 1994 were people that were caught in their cars during rush hour trying to get home ahead of the tornado.

For references, please contact the author.

Societal Aspects of Weather

Workshop's Main Page

Table of Contents