US Severe Convective Storm Model for Tornado & Hail Risk
EQECAT released its initial US Severe Convective Storm Model in 2004, and since then the model has been enhanced to incorporate the latest science. The US tornado risk analysis model addresses the loss potential associated with weather systems with thunderstorms that produce damaging tornado outbreaks, hail storms, and straight-line winds and is part of the RQE™ (Risk Quantification & Engineering) cat modeling platform.
Tornado/Hail Risk Model for the United States
All regions of the continental United States have varying degrees of thunderstorm risk. Areas of high risk include Tornado Alley, a high-frequency tornado risk zone that spans much of Texas, Oklahoma, Colorado, Kansas, Nebraska, and South Dakota. However, deadly and costly tornadoes have occurred in a much broader swath of the central US.
EQECAT integrates updated building codes and construction practices with the latest science and engineering to produce this state-of-the-art tornado risk analysis model for the United States. Both the hazard and vulnerability components of the US Severe Convective Storm Model offer unique modeling innovations, including:
- Specialized hail vulnerability functions for structures and automobiles
- A finite hazard footprint model that produces a robust representation of the focused and severe damages arising from tornadoes
- Embedded weather system modeling to simulate event occurrences of tens to hundreds of tornadoes
- Hail storms
- Straight-line wind events spanning multiple states
The RQE™ platform enables robust risk measurement and differentiation calculations. EQECAT’s unique approach to uncertainty is especially relevant to regions experiencing infrequent tornado touchdowns. Our robust methodology, which retains the full breadth of uncertainty through each step of the model calculation and reports multiple loss outcomes for any one event, allows re/insurers to:
- Gain confidence in decisions related to low probability events that have significant financial consequences
- Avoid surprises
- Set rational expectations about risk
The US Severe Convective Storm Model features include:
Event Frequency Model Based Upon Adjusted Historical Data
The frequency of severe convective storm events is based upon NOAA's Storm Prediction Center (SPC) public record of tornadoes and hail reports from 1950 onwards. The records contain details about each tornado event, including the tornado identifier, the tornado's Fujita classification, date, time of occurrence, maximal width, the tornado's length, number of injuries and fatalities, and geographic starting and ending position (latitude and longitude). In addition, the hail records provide date, time of occurrence, geographic position, and hail size, from a minimum of 0.75 inches up to 5 inches and above.
Improvements in tornado and hail tracking technology and reporting have led to an apparent increased frequency in tornado activity. The use of Doppler radar beginning in the mid-1980's and an improved understanding of tornado and hail formation processes enabled improvements to the reporting process especially with low F-Scale tornadoes. Therefore, historical data must be adjusted before a valid model may be developed. EQECAT's solution to the problem of reporting bias is to focus on data reported since the early 1990s, an approach that EQECAT validated using NOAA research. The resulting de-trended historical data set is used to model EQECAT's probabilistic stochastic event set for tornadoes, hail storms, and straight-line winds. Its robust probabilistic set includes approximately 840,000 events. The model also accounts for temporal and spatial clustering of events. The probabilistic set is evaluated against the historical data set for completeness and validation.
Statistical Model Testing and the Elimination of Model Bias
Tornado Damage in Joplin, Missouri - 2011
Source: Columbia Daily Tribune
The nature of severe convective storm events includes events that can:
- Endure for as long as seven days;
- Produce hundreds of tornadoes, hail storms and straight-line wind events, each with a damaging footprint of perhaps several acres;
- Produce damage in tens of states across the central US. The US Severe Convective Storm Model's stochastic event simulation has undergone stringent statistical convergence testing to ensure that the model produces a robust and credible representation of the insured risk from severe convective storms.
Convergence testing was done individually for separate components of severe convective storms (tornadoes, hail storms, and straight-line winds) and upon the combined model to demonstrate a consistent measure of risk.
Model Validation from a Broad Array of Sources
The annual probability of a tornado touchdown upon a specific parcel of land is very small, and the effectiveness of a loss model for severe convective storms requires identifying all potential sources of validation.
- Property Claim Services (PCS) data covering the time since 1970 provides one perspective of event frequency and severity, although increasing urbanization and insurance values limit the effectiveness of this data.
- Historical records of crop-hail insurance loss payments provide a complementary and useful source of validation for the hail model.
- There are many scientific studies of hail undertaken for suppression studies, designs of structures, crop and property insurance risk assessments, and aircraft operations.
- A fourth form of data is limited to hail-produced losses in events that cause more than $5 million (in year of occurrence) in property damage, labeled by the insurance industry as catastrophes. In some years the crop-hail data was collected for individual storms, allowing examination of storms on a weekly and monthly basis. This data has been compiled on various geographical scales and offers considerable spatial information on patterns of crop-damaging hail.
- Modeled versus observed burning cost evaluations of stable insurance portfolios were undertaken in many regions of the country to develop greater confidence in the model.
- The US SCS hazard model has also been reviewed and validated by Dr. Harold Brooks. Dr. Brooks is a Research Meteorologist and heads the Mesoscale Applications Group at NOAA/National Severe Storms Laboratory.
Stochastic Event Set
EQECAT uses stringent acceptance criteria in the development of the stochastic event set. The resolution grid of tornadoes, hail storms, and straight-line wind events is constant across the landscape, which provides consistent results. Since all geographies are modeled with equal confidence, even the most specialized portfolio will be modeled to the same high standards as a market portfolio. The model includes spatial and temporal clustering, which is an essential aspect of severe convective storm risk.
The tornado risk analysis model incorporates vulnerability curves that are developed from reviews of historical loss data and wind risk studies conducted by EQECAT and ABS Consulting over the last 20 years.
The tornado/hail model specifically models wind (tornado and straight-line winds) and hail (damage due to the kinetic energy associated with impact).
The model calculates damage to structures (building damage), contents, and damage related to time (loss of operations or loss of use). Separate, independent vulnerability functions are used for calculating losses related to each coverage type. Time-element vulnerability functions are a function of structural and contents damage.
US Severe Convective Storm Model Specifications
The US Severe Convective Storm Model helps reinsurance and insurance clients perform a variety of crucial analyses and provide risk perspectives, including:
- Understanding US severe convective storm risk correlation at the site, policy and portfolio levels;
- Pricing and managing US severe convective storm risk at the site, policy, and portfolio levels;
- Communicating US severe convective storm risk to important stakeholders, including rating agencies, regulators, shareholders, and counterparties.
The US Severe Convective Storm Model specifications include:
Import and risk evaluation is geocoded at latitude and longitude coordinates, ZIP Codes, and county levels. When exposure data is provided at aggregate levels (ZIP Code and county), the model adds refinement to loss results by disaggregating data to finer resolution points based on weighted distribution of values for the purpose of analysis and risk estimation.
Hazard Analysis Resolution
The hazard is gridded into cells of approximately 10 km2, smaller than the typical ZIP Code. The tornado touchdown grids are elongated rectangles aligned in a north-easterly direction approximating the most likely tornado path. Hail storm grids are square.
The model's geographic coverage includes the 48 contiguous United States (including the District of Columbia).
Lines of Business
Lines of business modeled in the US Severe Convective Storm Model include:
Structure Types and Occupancies
With many different structure types and dozens of occupancy categories for each line of business, the US tornado/hail model differentiates risk across hundreds of combinations.
Risk metrics include loss exceedance curves, including OEP/AEP, AAL, TVAR, event-by-event losses with associated uncertainty, as well as simulations of historical events. Reporting of results supports multiple levels of refinement:
- Total aggregate portfolio
- ZIP Code or county
- Detailed output by policy and site
In addition, RQE’s Year Loss Table (YLT) uniquely features three dimensional output:
- Simulation year
- Sample outcomes
Instead of reporting mean losses with standard deviations, each loss in the YLT represents one possible outcome for the associated event. This allows users to retain the full distribution of uncertainty when using model output in dynamic financial analysis and capital modeling. Conventional event loss results and other risk metrics can be derived from the YLT with arithmetic or simple database queries. YLT and event loss results are supported at the portfolio level.
Other risk metrics are supported at multiple levels of refinement, from total aggregate portfolio results to detailed output by policy and site.
All major insurance policy structures and reinsurance treaty types are modeled.
Request US Severe Convective Storm Model information from EQECAT.