A burst of trouble: thunderstorms can cause greater annual aggregate losses than hurricanes, so insurers must be better prepared.

In May of this year, the insurance industry watched the unfolding of a weather pattern in the United States that resulted in the largest-ever insured loss from an outbreak of severe thunderstorms.

Between May 2 and May 11, 2003, multiple supercell thunderstorms developed and spawned numerous tornadoes, hailstorms and straight-line windstorms. The Storm Prediction Center of the National Weather Service received 422 tornado reports, 1,477 hailstorm reports and 1,055 straight-line windstorm reports during the period. Eighteen states were affected. Of particular significance was the unusually long duration of the weather pattern, which is more typical of a blocked system that can occur during the summer months, when the jet stream is much weaker in strength. Also significant is the location where the pattern built up, as the trajectory of the surface flow allowed warm and humid air from the Gulf of Mexico to return to the region below the jet stream on a daily basis. This recharged the atmosphere and allowed for repeated tornadic events.

From a meteorological perspective, it was one of the largest severe thunderstorm outbreaks in U.S. history. From an insurance perspective, the storms caused the largest-ever insured loss blamed on a thunderstorm system. The most recent loss estimate from the Insurance Services Office's Property Claim Services puts the losses at more than $3.2 billion, and the number could still go higher.

Was this event truly an aberration? With what frequency can insurers expect such losses to occur? When estimating potential losses from infrequent, but high-impact events, extrapolation from historical losses is insufficient. Property values grow, population increases, and tornadoes that once leveled wheat fields may today rip through suburban neighborhoods. Still, a review of recent losses from severe thunderstorms may be revealing to those who might underestimate the loss potential from this often forgotten peril.

"Largest Occurrence Losses From Severe Thunderstorms" (see page 76) shows the largest occurrence insurance losses for each of the past five years. An occurrence is defined as an outbreak of (typically) multiple tornadoes, hailstorms and straight-line windstorms spawned over the course of one or more days as the result of the same atmospheric event or frontal system.

Even more startling are annual aggregate losses. Though the numbers in the table rival insured losses from many hurricanes, individual event losses from thunderstorm systems are generally not as high as from tropical cyclones. Estimates of potential average annual-aggregate losses nationwide, however, are actually higher for severe thunderstorms than for hurricanes, according to AIR Worldwide, a catastrophe modeling firm based in Boston. Tellingly, states at risk from severe thunderstorms account for four of the top 10 states in terms of natural catastrophe loss potential, according to AIR. At the top of the list, severe thunderstorm risk in Texas follows only California earthquake and Florida hurricane risk for natural catastrophe loss potential.

How often can insurers expect to experience losses of a magnitude similar to the May 2003 outbreak? Catastrophe models, such as those widely used for assessing hurricane and earthquake risk, suggest once in roughly 12 years. The previous record of $2.2 billion, set in April 2001, has an estimated return period of about seven to eight years. Yet, despite the magnitude of the potential losses, many insurers treat losses from severe thunderstorms as simply a cost of doing business, convinced that severe thunderstorms are so erratic and the potential damage so localized that it is impossible to manage the risk in the same manner that one might manage risk from hurricanes or earthquakes. On the contrary, recently introduced high-resolution simulation models capture the risk from severe thunderstorms at an extremely localized level, enabling effective and proactive risk-management practices.

Severe Thunderstorm Risk

Thunderstorms are hardly rare. In fact, at any given moment, there are approximately 2,000 thunderstorms in progress around the world, according to the U.S. National Weather Service. Fortunately, most of these are not severe. Only about 10% of the thunderstorms that occur in the United States each year are classified as severe, which the service defines as one "that produces a tornado, winds of at least 50 knots (58 mph), and/or hail at least 0.75 inch in diameter." Still, some 1,200-1,500 tornadoes are reported each year in the United States on average, along with 7,000-10,000 hail reports, and 8,000-12,000 severe straight-line wind reports.

The Great Plains region is particularly prone to severe thunderstorms. The reasons are meteorologically straightforward:

* Warm moist air from the Gulf of Mexico, the maritime air mass, flows inland at the surface.

* Cool upper-level air, the continental air mass, is funneled down the east side of the Rocky Mountains.

* The land between the Rocky Mountains and the Appalachian range is mostly flat and so does not impede airflow.

* Jet-stream winds aloft dynamically energize the thunderstorms.

Severe thunderstorms are not confined to Tornado Alley, an area comprising parts of approximately 15 central states. They can, in fact, occur virtually anywhere east of the Rockies, as evidenced by an F-4 (wind speeds of 207-260 mph) tornado that ripped through eastern Maryland in April 2002. While part of a larger system that caused $1.7 billion in damages across 16 states, a La Plata, Md., tornado cost more than $250 million alone.

Modelers distinguish between the so-called "microevent," which is any individual tornado, hailstorm, or straight-line windstorm, and the "macroevent," which represents the larger atmospheric system and can include many--even hundreds--of microevents. Catastrophic losses typically result from the macroevent, which affects whole regions over several days. But large losses also can result from individual tornadoes or hailstorms, particularly if they occur in an urban setting. A single hailstorm packing three-inch hailstones wreaked havoc across north St. Louis in April 2001, resulting in losses of $1.2 billion in April 2001, according to Property Claims Service.

Answering the Questions

The fundamental questions that models must answer are where, how big and how often? The historical data modelers use to construct their catalogs of simulated severe thunderstorms are available at the microevent level from the Storm Prediction Center, a division of the National Weather Service. The center's database includes information on more than 39,000 tornadoes, 108,000 hailstorms and 145,000 straight-line windstorms that have occurred over the past 50 years. Using statistical clustering techniques and meteorological expertise, these data are then combined into macroevents.

Despite the abundance of historical data, however, serious shortcomings result from significant underreporting of historical microevents. Underreporting is most severe for the period between 1955 and the early-1970s, when no formal reporting system existed. In the 1970s, the National Weather Service launched a campaign to recruit volunteers to make and report "spot observations." The level of reporting shows a steady increase since then. In the early 1980s, the NWS began to issue warnings of severe thunderstorms. Increased warnings brought about increased awareness and undoubtedly contributed to the increase in the number of reports. An even more significant increase in the number of reported events has occurred since the 1990s, when Doppler radar systems were installed at local and regional weather stations and the density of coverage increased dramatically as a result.

As can be seen in "Reported Storms Between 1955 and 1998" (see above), the most severe historical underreporting exists for hail and straight-line windstorms. Both show an almost exponential growth in the number of reports over time. While there has been an increase in the number of reported tornadoes as well--and particularly for weaker F-0 and F-1 tornadoes--underreporting of these events has been less acute; tornadoes are simply less likely to go unnoticed.

A critical element of accurately assessing current severe thunderstorm risk, therefore, is data augmentation to compensate for this historical underreporting. Using sophisticated statistical techniques, augmentation consists of temporal filling (microevents are added to remove or alleviate time trends), spatial filling (microevents are added in geographic locations where underreporting is known to exist), and statistical smoothing to produce a spatially smooth surface of event frequency.

These smoothed frequency distributions are developed separately for each of the three perils at high resolution. A resolution of 1km2, for example, results in nearly 7 million modeled locations across the United States. The high resolution is necessary because the perils associated with severe thunderstorms are often quite localized and affect, in many cases, only a small area within a ZIP code. The high degree of spatial resolution enables risk assessment down to the address level when detailed exposure information (construction type, building characteristics, etc.) is available.

Estimating Losses

Estimating potential losses from severe thunderstorm events involves the use of damage functions, which relate the parameters that define the intensity of the microevents to the physical damage done to buildings and their contents.

When tornadoes come into contact with buildings, roofs are often the first element to be stripped by the force of the wind. Loss of the first shingle allows wind to penetrate under and lift the next shingle. Unsecured slates may peel off, metal roofs may roll up and off, overhangs and eaves on the windward side become vulnerable. Once the roof fails, walls may collapse from loss of lateral support ordinarily provided by the roof system.

Of the three microevents associated with severe thunderstorms, straight-line winds are often the most overlooked, but can account for as much as 40% of total losses. These localized wind events can produce wind speeds in excess of 100 mph. One variation is the microburst, a surface wind generated by a downdraft. Winds from microbursts have been known to reach 146 mph. Finally. losses from hail are based on intensity, which is measured in terms of "hail impact energy." This is a function of hailstone size, the number of hailstones (per square meter, for example), and wind speed, which propels the hailstones and exacerbates their gravitational impact.

Because tornadoes, hail, and straight-line windstorms inflict damage differently, structural engineers develop damage functions individually for each peril. The functions provide detailed breakdowns of loss estimates by coverage, construction class and occupancy. Separate damage functions provide loss estimates for buildings, contents and time element. They are extensively validated based on actual claims data and the results of detailed post-disaster field surveys.

Managing the Risk

Insurers can use high-resolution severe thunderstorm models to make proactive risk management decisions and increase the profitability of their portfolios. Specifically, modeling helps insurers minimize accumulations of risk, strategically price their policies and determine various reinsurance and risk transfer strategies.

Catastrophe modeling provides insurers with the estimated loss potential for individual risks, books of business or an entire portfolio. Results can be analyzed for pricing decisions and risk transfer strategies. Model output may also be used to run sensitivity tests, develop underwriting guidelines, analyze policy conditions, make sound decisions regarding the purchase of reinsurance, estimate consistent loss costs for catastrophe-prone areas, and execute overall catastrophe risk management.

Output can be customized to any desired degree of geographical resolution down to the street address. Catastrophe modeling software also provides summary reports of exposures, comparisons of exposures and losses by geographical area, and detailed information on potential large losses caused by extreme "tail" events.

Insurers face a significant risk from severe thunderstorms. Billion-dollar occurrence losses are no longer uncommon, and annual aggregate losses can exceed those from hurricanes. Insurers need to adequately prepare for these large losses. Catastrophe modeling can provide the framework to appropriately manage the risk.

Largest Occurrence Losses From Severe Thunderstorms

An occurrence is typically defined as an outbreak of multiple
tornadoes, hailstorms and straight-line windstorms spawned over the
course of one or more days as the result of the same atmospheric
event or  frontal system.

                                                       Estimated Loss
Year   Period          States                            ($ Millions)

2003   May 2-11        Ala., Ark., Colo., Ga., Iowa,           $3,130
                       Ill., Ind., Kan., Ky. Mo.,
                       Miss., N.C., Neb., Ohio,
                       Okla., S.C., S.D., Tenn.

2002   April 27-May 3  Ark., Ga,, Ill., Ind., Kan.,            $1,675
                       Ky., Md., Mo., Miss., N.C.,
                       N.Y., Ohio, Pa., Tenn., Texas,
                       Va., W.Va.

2001   April 6-12      Ark., Colo., Iowa, Ill., Ind.,          $2,200
                       Kan., Ky., Mich., Minn., Mo.,
                       Neb., Ohio, Okla., Pa., Texas,
                       Wis.

2000   March 28-29     La., Texas                                $520

1999   May 3-7         Ala., Ark., Fla., Ga., Ill.,            $1,485
                       Ind., Kan., Ky., La., Mo.,
                       Miss., N.C., Neb., Ohio,
                       Okla., S.C., Tenn., Texas

Source: AIR Worldwide, Insurance Services Office Property Claim
Services

Beverly Porter is director of communications and Tim Doggett is research scientist at AIR Worldwide Corp., Boston.