AIR Currents

Jan 23, 2014

Editor's Note: AIR is releasing a major update to the AIR Severe Thunderstorm Model for the United States in June 2014. This article provides an overview of some of the most significant changes, many of which will be examined in greater detail in future articles.

The AIR Severe Thunderstorm Model for the United States was first released in 1987 as the industry's first probabilistic model to help companies proactively manage severe thunderstorm losses. Since then, the model has undergone several updates. In the summer of 2014, AIR will release a major update to the model that leverages 10 years of new data and scientific research including additional event reports from the Storm Prediction Center (SPC), high-resolution radar data, over 38 billion dollars in insurance company and Xactware® claims data, and the results of detailed damage surveys.

This latest release encompasses marked enhancements to all three model components—hazard, damage estimation, and insured loss calculation. This article provides a high-level overview of how all of these changes reflect a better view of severe thunderstorm risk across the continental United States. Subsequent AIR Currents articles will delve into these changes in more detail.

Advanced Hazard Modeling

The U.S. severe thunderstorm model captures the spatial extent of macroevents (combinations of all three sub-perils that result from the same atmospheric event or frontal system) as well as the highly localized effects of microevents (the individual tornadoes, hailstorms, or straight-line windstorms that comprise a macroevent). The model's stochastic catalog was built based on historical data from the SPC, a division of the National Oceanic and Atmospheric Administration's (NOAA) National Weather Service, for the period of 1979 (the beginning of the availability of NCAR's Climate Forecast Systems Reanalysis data) through 2011.

Despite the wealth of data from the SPC, one of the biggest challenges that modelers face is underreporting in certain areas due to the high correlation between population density and reported events. While underreporting was most severe prior to 1979, there remains significant underreporting up to the present day. Additional bias was introduced as a result of population growth (the number of reports grew as the population grew) and the surge in event reports in the late 1990s and 2000s due to advances in technology, social media, and the blockbuster 1996 movie "Twister."

Scott StranskyBy: Scott Stransky
Senior Scientist

Kathryn FobertBy: Kathryn Fobert
Science/ Technical Writer

Edited by Nan Ma

To overcome the limitations of the historical data, AIR scientists employed several new smoothing, data augmentation, and statistical detrending methods. Statistical smoothing, for example, allows simulated events to occur in areas where historical storms are possible but have not yet been recorded. The updated model improves upon purely statistical smoothing by leveraging meteorological parameters from reanalysis data to "smart-smooth," thus ensuring physically realistic locations of simulated events.

These high-resolution parameters (expressed as index values) such as wind shear, convective available potential energy (CAPE), and temperature gradients can be used to determine when and where conditions are favorable for severe thunderstorm formation. They also enable the model to capture major outbreaks very similar to those that occurred prior to or after the historical record used in model development, such as the 1974 Super Outbreak, in which over 60 EF-3 or greater tornadoes struck, or the late season EF-4 tornadoes that struck Illinois in November 2013.

Cat Bond Figure 1 Event 1
Figure 1. The model's spatial distribution of all tornado touchdowns indicates increased activity in Dixie Alley (circled region), compared to the previous model. Note that dark shades of red indicate the highest amounts of activity.

Smart-smoothing along with the availability of more simulated events (AIR is making available both a 50,000-year and 100,000-year catalog in addition to the standard 10,000-year catalog) also improves model convergence allowing for a better view of the risk at finer scales.

As critical as the geographic distribution of simulated events is to the accuracy of model results, it is equally important that these microevents have realistic swath sizes (damage footprints). Given the considerable uncertainty in the spatial extent of hail and straight-line wind events in the SPC data, AIR researchers developed a new clustering algorithm that groups SPC reports that are close in space and time to produce more realistic damage footprints for simulated events. Further refinement of hail swaths was achieved by using radar data from the major outbreaks of 2010 and 2011; and since hail extends aloft many thousands of feet, radar data was analyzed from a vertical perspective, using a metric called Vertically Integrated Liquid (VIL).

Measuring Local Intensity

In the AIR model, the measure of intensity for tornadoes and straight-line winds is wind speed; for hailstorms, it is hail impact energy. Based on the latest available data and scientific literature, these parameters were reformulated for the 2014 model update. The updated hail energy measure contains both a horizontal component, which accounts for damage to the sides of buildings, and a vertical component, which accounts for the damage to roofs—the primary source of damage from hailstorms. For tornadoes, wind speed profiles—both in line with and perpendicular to the tornado track—are based on detailed analysis of data from AIR's own damage surveys and detailed claims data from recent events.

Engineering Enhancements

The 2014 release of the AIR U.S. Severe Thunderstorm Model also features many significant updates to the damage estimation component of the model, allowing for better risk differentiation by building attributes and geography. These enhancements include updates to damage functions based on detailed damage and loss data from recent events, sub-peril-specific probability distributions around the mean damage ratio, and the inclusion of a wide variety of secondary risk features specific to each sub-peril.

The model's damage functions were enhanced based on the latest published research and engineering principles along with post-disaster surveys conducted by AIR engineers in 2008, 2011, and 2013. Notable among these studies are detailed damage survey data from 2011 tornado outbreaks collected by Texas Tech University (TTU) and analyzed by AIR in combination with TTU, and the results of the Roofing Industry Committee on Weather Issues, Inc. (RICOWI) damage survey following the 2011 Dallas/Fort Worth hailstorm, in which AIR was a participant. During that survey, maximum hail size was recorded at more than 100 sites based on observed damage, and the impacts by roof type were measured using a damage scale from 0 (no apparent damage) to 5 (severe damage). RICOWI's survey results were particularly useful in quantifying the relative vulnerability of impact-rated roof coverings.

Capturing Regional and Temporal Vulnerability

The 2014 update to the AIR Severe Thunderstorm Model for the United States features a comprehensive approach to modeling spatial and temporal variations in vulnerability. This approach is based on an extensive, peer-reviewed study that AIR researchers undertook to understand the large number of building codes and design wind standards that exist across the continental United States. This study revealed that the evolution of building codes varies geographically and that in some states, code changes are implemented on the local level. The study results also showed the significant impact that secondary risk features (such as roof covering type, roof geometry, and roof pitch) have on building vulnerability.

As a result, for each location and year-built, AIR researchers defined model buildings in terms of secondary risk features and code enforcement level. The vulnerability of different model buildings was then ascertained by applying modifiers corresponding to the appropriate secondary risk features to the base damage functions. Weighting factors were used to combine the effects of features whose interaction is complex and not necessarily additive. Figure 4 shows the result of such an analysis for a home built in Atlanta in 1996.

Cat Bond Figure 1 Event 1
Figure 4. Example of how AIR researchers develop a damage function for a residential structure built in Atlanta in 1996
Cat Bond Figure 1 Event 1
Figure 2. Roof damage photos from RICOWI's damage survey following the 2011 Dallas/Fort Worth hailstorm (Images courtesy of RICOWI)

The vulnerability component was also informed by results from the Insurance Institute for Business & Home Safety (IBHS) study on the impact of hail on property damage. On February 20, 2013, IBHS produced the first-ever indoor hailstorm to study the devastating effects of hail on property, including different roof types. IBHS also released a detailed report to its member companies summarizing the findings from their analysis of insurance claims from the 2011 hailstorms in Dallas-Fort Worth.

Figure 3. IBHS's full-scale residential house testing of the hail impact resistance of roof coverings (Video courtesy of IBHS)

The AIR model's sub-peril-specific damage functions are based on extensive claims data, including nearly 300,000 insurance company claims and over 4 million claims from AIR's sister company, Xactware. Detailed analyses of these claims data revealed that the wide range of damage to structures of similar construction for the same wind speed or hail impact energy is also sub-peril specific. As a result, the model contains new, sub-peril-specific probability distributions around the mean damage ratio. Touchstone® will allow companies to analyze results for each sub-peril individually as well as for all of three sub-perils combined, thereby gaining further insight into a highly complex risk.


The updates to the 2014 AIR Severe Thunderstorm Model for the United States are a culmination of the latest data and research in the fields of statistics, meteorology, and engineering. While severe thunderstorm is a relatively high frequency peril and individual events typically do not cause loss levels like that of a catastrophic hurricane, it is the aggregation of losses that can lead to unexpected financial results. These losses can be very volatile from year to year due to the myriad of climate and weather factors that give rise to them and the localized pattern of damage. Such volatility combined with the construction of more high value property in harm's way makes the need to understand this risk more critical than ever. AIR's updated model provides companies with a more realistic, detailed, and comprehensive view of severe thunderstorm risk—the sort of robust view that will facilitate the move from treating severe thunderstorm events simply as a cost of doing business, to proactive risk management.




You’re almost done.
We need to confirm your email address.
To complete the registration process, please click the link in the email we just sent you.

Unable to subscribe at this moment. Please try again after some time. Contact us if the issue persists.

The email address  is already subscribed.