Tornado Risk Analysis in United States

Abstract eng:
Perils of tornado and hail cause large amount of loss every year. The growth in population and wealth only increases the loss in the future. Since 1949, based on the data provided by Property Claims Services, tornado, hail and straight-wind losses account for more than 40 percent of total natural losses in United States (.Converium [2]) Given the high frequency of tornado and damaging hail in continental United States (about 9000 single tracks per year) and the area exposed to them, quantifying these risks will be an important advancement in pricing them for insurance/reinsurance purposes. In the absence of realistic physical model, which looks at these perils in a cluster/outbreak basis, it is not possible to underwrite these risks effectively. This paper focuses on the tornado risk. A tornado windfield model is developed based on the model used by Wen and Ang[6]. The windfield is varied, as the storm/intensity evolves and dissipates, along the path. This provides a better resolution for calculation of wind speed at various locations when estimating the losses for a given tornado. The model is calibrated to the specifications given in the Fujita intensity scale to generate the appropriate range of wind speeds for each intensity level. In order to estimate the tornado hazard, a stochastic event-set is generated. The event-set uses the information provided by SPC (Storm Prediction Center [5]) along with the dataset given by Grazulis[3] (1921-1995). The SPC information is corrected for repetition of the same storm in different location (state or counties) and the gaps were filled based on statistical inference from neighboring areas as well as using Grazulis data. The storm parameters such as number of single tracks within an outbreak and number of outbreaks per year are de-trended. This event-set along with different timeframe criteria is used to come up with a definition of outbreak. This was done with a (re)insurance application in mind. These historicalbased outbreaks are used in a Markov chain process to generate a large number of outbreaks representing 35,000 years of simulated data. This event-set is later used to come up with temporal frequency contours (return period) and loss analyses. The results will focus on the frequency of occurrence of tornadoes of different intensities (Fujita scale) in contiguous United States. The spatial distribution of annual loss will be discussed and conclusions are made with regard to convergence and overall uncertainty in the results.

• Export as: BibTeX | MARC | MARCXML | DC | EndNote | NLM | RefWorks
• Add to your basket