Prognostication elevated: models based on numerical weather prediction are being used to generate potential losses from winter storms.

After having experienced one of the costliest hurricane seasons in U.S. history, many insurers anticipated to the relatively quiet winter months with a sense of relief. For some, however, it may be a false sense of security.

According to ISO's Property Claim Services unit, U.S. insured losses from winter storms have totaled more than $15 billion since 1990. More than 10% of those losses, or $1.75 billion, can be attributed to the Blizzard of '93, a storm that has been labeled by many as the "Storm of the Century." Though winter storms often are not thought of as catastrophes on par with hurricanes and earthquakes, the Blizzard of '93 still ranks among the top 20 costliest natural catastrophes in the United States.

The Blizzard of '93 caused the largest insured losses of any winter storm in U.S. history. What should give pause to insurers, however, is that even small and equally likely changes in the storm's initial atmospheric conditions could have significantly changed the storm's evolution and its ultimate financial impact. Things could have been worse.

Loss Drivers

Losses from winter storms are driven by three components: wind, precipitation and temperature. Natural variations in the combination of these components lead to differences in the manifestation of winter storms in different regions of the country. Examples of the types of storms that predominate in different regions include:

* Nor'easters along the East Coast

* Ice storms in the Southeast, South-Central and Midwest states

* Freeze conditions in the South-Central states

* Windstorms along the West Coast

* Lake effect snows along the Great Lakes

* Ocean effect snows along the Atlantic coastline

In the case of the Blizzard of '93, each of the winter storm components contributed to the overall losses. For example, precipitation drove a significant amount of the $300 million insured loss in Georgia, where heavy snow accumulated on roofs in the northern part of the state, resulting in the collapse of numerous industrial building roofs. The combination of heavy snow and strong winds caused widespread power outages throughout inland areas of North Carolina, which contributed to the $130 million of insured losses in that state. Straight-line winds heavily impacted the coastal areas of Long Island and the New York metropolitan area, contributing to the $105 million loss in that state.

Freezing temperatures--including record lows in many areas--led to water damage from burst pipes and, where coupled with precipitation, roof damming. Near-freezing temperatures combined with heavy precipitation, resulting in icing conditions that led to numerous automobile accidents. Auto claims contributed almost $100 million to the total insured loss.

The Butterfly Effect

For the insurance industry, the Blizzard of '93 was the most significant winter storm in U.S. history. However, had the storm's initial conditions been slightly different--a half a degree increase or decrease in temperature, for example, or a similar difference in dew point--a different storm could have resulted. That is, the winter storm of March 1993 might have been only a minor inconvenience, never entering the record books--or it might have been worse than it actually was.

This concept is popularly known in meteorology as the "butterfly effect: It is the notion that, for example, the effects of a butterfly fluttering its wings today in Australia can so disturb the atmosphere as to cause rain to fall in Great Britain tomorrow. Relatively small changes at one location can propagate through the atmosphere to cause measurable, and even dramatic results elsewhere.

The atmospheric conditions that characterize winter storms are highly complex, with centers of high and low pressure, fronts, and a strong jet stream all interacting. This complex interaction can result in a wide range of potential outcomes, all of which are strongly influenced by the initial atmospheric conditions.

Numerical Weather Prediction

Traditional statistical catastrophe models cannot capture complex nonlinear interactions such as the butterfly effect. To produce a wide range of potential storm scenarios requires the use of physical models, such as those based on numerical weather prediction.

Used today by all major meteorological agencies around the world, models based on numerical weather prediction work by putting in motion a three-dimensional grid of atmospheric initial conditions such as temperature, moisture, and pressure. The mathematical equations that represent the meteorological behavior of the atmosphere are then used to simulate how the atmosphere evolves over time.

By employing techniques based on numerical weather prediction, weather forecasters in 1993 were able to project several days in advance that the conditions brewing in the Gulf of Mexico would likely develop into a major winter storm. Based on the meteorological parameters of an incipient storm, models based on numerical weather prediction can simulate a spectrum of possible scenarios that actually could develop. These scenarios--called an ensemble-represent a range of realistic potential storms based on the uncertainty in the evolution of the initial meteorological conditions.

By repeatedly changing the initial conditions of actual storms by small, but equally likely amounts, a large ensemble of potential storms can be generated that reflects various combinations of intensity, spatial footprint and temporal evolution. In this fashion, a very large catalog of realistic storms is produced that reflects the true likelihood of various combinations of atmospheric conditions and resulting storm characteristics. As is observed in the historical record, most simulated storms are of minimal intensity and cause little damage. However, a reasonable number are of moderate intensity, a few result in storms like the Blizzard of '93, and a very few result in even more extreme storms and higher insured losses.

Catastrophe models based on numerical weather prediction can be used to better understand the conditions that contributed to losses from historical events such as the Blizzard of '93. They also reveal that while aspects of the Blizzard of '93 were fairly infrequent occurrences for particular affected regions, slightly different atmospheric conditions could cause even more extreme winter weather.

Of course, the real purpose of catastrophe models is not to estimate losses from historical storms, but to anticipate the likelihood and severity of future events so companies can appropriately prepare for their financial impact. Because winter storms result from the complex interaction of atmospheric conditions, a model is required that accurately evolves these conditions and accounts for the uncertainty inherent in storm development. To this end, a fully probabilistic, numerical weather prediction-based winter storm model is the best tool for insurers and reinsurers to objectively and accurately quantify the risk to their own portfolios from this peril.

Key Points

* The Blizzard of 1993 caused $1.75 billion in insured losses.

* Losses from winter storms are driven by wind, precipitation and temperature.

* Physical models such as those based on numerical weather prediction are needed to produce a wide range of potential storm scenarios.

The Blizzard of 1993

On March 12; 1993, a newly formed center of low pressure moved into a zone of existing instability and abundant moisture over the Gulf of Mexico. An extratropical cyclone rapidly developed and swept over northwest Florida during the early hours of March 13. An intense squall line preceded the rapidly moving cold front as it raced across Florida producing torrential downpours, hurricane force winds, a 12-foot storm surge, and 11 confirmed tornadoes. At several hundred miles across, the storm covered an area about 10 times that of a typical hurricane.

An intense pressure gradient developed as a result of the storm's rapidly falling central pressure. Fierce winds coupled with the heavy snow paralyzed airports and highways across the eastern third of the country, resulting in the cancellation of more than 25% of the nation!s flights over a two-day period. Some interstate highways were rendered impassable and, at one point, more than three million people were without power.

Over its five-day lifespan, the Blizzard of '93 affected 26 states and impacted the lives of nearly 100 million people, nearly half the U.S. population. The storm resulted in 270 fatalities and insured property damage of $1.75 billion in 1993 dollars.

Blizzard of 1993 Statistics

* Hurricane force winds along the East Coast--ranging from 109 mph in Florida to 144 mph in New Hampshire

* Record-breaking temperatures were recorded in Burlington, Vt., minus 12, and Birmingham, Ala., 2 degrees

* Record snowfalls were recorded across the South and Mid-Atlantic states--50 inches in Mount Mitchell, N.C., and 47 inches in Grantsville, Md.

U.S. Catastrophe Losses by
Cause of Loss, 1983-2002 (1)

Water Damage                 0.2%
Civil Disorders              0.6%
Fire (6)                     2.3%
Wind/Hail/Flood (5)          4.2%
Earthquakes (4)              9.8%
Winter Storms               10.5%
Terrorism                   12.4%
All Tropical Cyclones (3)   28.0%
Tornadoes (2)               32.1%

(1) Catastrophes are all events causing direct
insured losses to property of $25 million or
more in 2002 dollars. Adjusted for inflation by
ISO.

(2) Excludes hurricanes and tropical storms.

(3) Includes other geologic events such as
volcanic eruptions and other Earth movement.

(5) Does not include flood damage covered by the
federally administered National Flood Insurance
Program.

(6) Includes wildland fires.

Note: Table made from pie chart.

Source: Insurance Services Office Inc. (ISO)