We use performance cookies to collect information about how you use our website (for instance which pages you visit most often). Cookies help us to improve your online experience with Aspen. Find out more here

close

Aspen Opinion

US Tornadoes and Hail: Volatility, Predictability and Urban Risk

October 28, 2015

Dr. Giovanni Leoncini, Aspen Re’s storm risk scientist, discusses the current challenges of increasing volatility, seasonal predictability relating to El Niño and tornado risk in urban areas.

The 2013 El Reno and 2011 Joplin tornadoes are reminders of just how destructive severe thunderstorms can be. Losses associated with convective storms, including hail and "straight-line" winds, can also be significant and even a relatively benign year, such as 2014, resulted in US$13 billion of insured losses.1 While hail and straight-line wind damage can also be costly, most of the recent scientific publications have focussed on tornadoes.

Increasing Volatility

In a previous Aspen Opinion we reviewed evidence to suggest that the volatility of tornadoes is increasing and there have now been a number of research papers substantiating this point. In the case of a tornado, the actual ground area affected relative to the area where it may potentially occur is very small. This means that the probability of an area experiencing a tornado is actually very low – even in high risk areas such as Oklahoma.

Nonetheless, when a tornado does occur, it often results in a 100 percent loss and yet buildings that are one block away from the destruction can be completely unscathed. The combination of low probability and high damage generates a large spatial volatility effectively “built in” to the hazard. This spatial volatility together with large inter-annual changes in tornado counts builds up a large volatility of losses. Tornado observations of the last few years indicate that the volatility of tornadoes is actually increasing, although the average number of tornadoes per year is roughly constant.

The initial paper suggesting such an increase in the volatility of the annual counts of tornadoes was presented by Brooks in 2013 at the European Conference on Severe Storms. Others have subsequently studied the problem from different angles with consistent results. For example, Tippett (2014) used different metrics and agreed that in recent years the total number of tornadoes has changed dramatically from year to year. Both these studies are based on the Local Storm Report database, collected by the National Weather Service, which provides a wealth of information, but presents weaknesses in its compilation. For example, tornadoes prior to 1973 were rated long after the event through newspaper reports and other anecdotal evidence. However, this and other limitations are now relatively well known and have been taken into account.

The changes in variability are statistically significant and have also been observed in the 2000s when reporting practices were constant. Tippett also links the changes in volatility to changes in the environment where storms form and evolve. More precisely, changes in a storm's relative helicity (a measure of how conducive the wind flowing into a storm is to rotation) explain part of the increased volatility of tornadoes. Once a storm starts rotating, it has a greater chance of generating a tornado. As a result, changes in volatility of tornadoes can, at least partially, be attributed to changes in the environmental conditions.

Trapp (2014) showed statistically that significant tornadoes (rated (E)F2 or greater) and outbreaks of tornadoes (a day with at least 20 tornadoes) are likely to be part of multi-day events. This analysis does not specifically define clusters in space, but spatial clusters of tornadoes were the subject of the spatial analysis of Elsner et al. (2014). They confirmed that during outbreaks, tornadoes tend to occur in clusters and showed that the number of tornadoes per unit area of the cluster is actually increasing.

In recent years there are, therefore, more tornadoes per cluster of similar area. Because clusters increase the spatial and temporal variability, the observed changes in the density of tornadoes within clusters are consistent with the also observed increase in the volatility of the annual tornado count mentioned above.

Some of these concepts are summarized in Figure 1 which shows that (E)F1 tornado days are decreasing, while the "big days", with at least 30 (E)F1+ tornadoes, are increasing. These trends are quite strong and it is unlikely that they are due exclusively to under-reporting or other data issues. The consequence of these trends is that the variability or volatility of tornadoes is increasing. If the number of days with more than 30 (E)F1+ tornadoes is increasing while the average number of tornadoes is roughly constant, the change in the number of tornadoes per day increases.

Figure 1: Tornado Days

Days with at least one (E)F1+ tornado over the continental US (black dots), days with at least 30 (E)F1+ tornadoes (grey dots). The solid lines show the 10 year average.

Source: Brooks et al.

Figure 2 shows the yearly value of standard deviation of the daily counts of (E)F1+ tornadoes. The red line represents the linear trend and the scatter around it implies that a large part of the variability depends on other factors. Nonetheless, the trend is statistically significant and therefore a robust feature of the observations.

Figure 2: Standard deviation of daily counts 1955-2014

Standard deviation of the daily counts of (E)F1+ tornadoes (black dots). The red line represents the linear trend. 1974 and 2011 reflect the largest number of tornadoes, although the yearly counts of tornadoes are stable.

Source: Aspen Re

Seasonal Predictability

Tornadoes and hail are notoriously hard to predict on short timescales. The average warning time for tornadoes issued by the Storm Prediction Center of 14 minutes underlines this point.2 However, at the seasonal scale, researchers have concentrated on the relationship between tornadoes and hail, which are local phenomena, and the large scale environment in which they develop. This work is effectively providing the foundations of seasonal forecasts as the large scale environment is better known and weather and climate models simulate it with greater skill than the small scale characteristics of convective storms. There is an interesting development in this on-going research; Allen et al. (2015) demonstrated that lower tornado and hail activity in the central U.S. occurred during strong El Niño events and, conversely, higher tornado and hail activity occurred during strong La Niña events. Since El Niño winter conditions tend to persist into the spring, this relation can be used to forecast spring time hail and tornado activity for the central U.S already in winter. More work needs to be completed before a useful forecast emerges but, coupled with the current forecasting ability for El Niño, it suggests encouraging progress.

Tornado Risk in Urban Areas

Metropolitan areas, given the concentration of insured values, are of concern for the (re)insurance sector. The interaction between urban areas and weather has been the subject of research for several years. For example, it is now well known that urban areas and their associated Urban Heat Islands (UHI) interact with the environment with increased precipitation and lightening above and downwind of metroareas. Cusack (2014), once again using Local Storm Reports, demonstrated that more tornadoes per unit area occur in metro areas than the surrounding non-metro areas. Indeed metro areas, defined as a circle with a 40 km circumference centred on the city centre, received roughly 50 percent more F1+ rated tornado touchdowns per unit area than the nonmetro areas, defined as a ring around the metro area with an inner radius of 60 km and an outer radius of 100 km. 

Low population density can lead to under-reporting of tornadoes. However this cannot explain the lower incidence of tornadoes in the non-metro areas, because their population density exceeds the known thresholds for such underreporting. Two meteorological processes are more likely to explain the difference. First, the interaction of the UHI with the environment as the UHI can provide additional heat to an upcoming storm. Secondly, the taller buildings in metro areas slow down the wind and this increases the wind shear, i.e. the wind difference between the surface and the levels above. This can help storms to rotate favoring tornado genesis. While these results still need to be confirmed, for example by using tornado paths rather than touchdowns, they are generally robust and constitute a benchmark for the validation of cat model event sets.

Model Validation

The latest research shows quite clearly that tornado volatility has been increasing for the last few years, explaining at least in part the high year-to-year volatility of convective stormrelated losses. However, there is encouraging progress in seasonal forecasting of tornadoes and hail which, in the next few years can help in estimation of tornado and hail risk a few months ahead of the main convective season. The Aspen R&D team continues to monitor these developments and, when relevant, incorporate them in its validation of vendor models.

References

  1. http://www.swissre.com/media/news _ releases/Insured _ losses _ from _ disasters _ below _ average _ in _ 2014.html
  2. http://www.noaa.gov/factsheets/new%20version/Tornadoes_web_version_final.pdf

Bibliography

Allen, J. T., M. K. Tippett, and A. H. Sobel, 2015: Influence of the El Niño/Southern Oscillation on tornado and hail frequency in the United States. Nat. Geosci., 1–6, doi:10.1038/ngeo2385. http://www.nature.com/doifinder/10.1038/ngeo2385.

Brooks, H. E., 2013: Increased variability of tornado occurrence in the US in recent years. European Conference on Severe Storms. 

Cusack, S., 2014: Increased tornado hazard in large metropolitan areas. Atmos. Res., 149, 255–262.

Elsner, J. B., S. C. Elsner, and T. H. Jagger, 2014: The increasing efficiency of tornado days in the United States. Clim. Dyn., doi:10.1007/s00382-014-2277-3.

Tippett, M. K., 2014: Changing volatility of U.S. annual tornado reports. Geophys. Res. Lett., 41, 6956–6961, doi:10.1002/2014GL061347.

Trapp, R. J., 2014: On the Significance of Multiple Consecutive Days of Tornado Activity. Mon. Weather Rev., 142, 1452–1459, doi:10.1175/MWR-D-13-00347.1. http://journals.ametsoc.org/doi/abs/10.1175/MWR-D-13-00347.1.

 

Back to articles

The above article/opinion reflects the opinion of the author and does not necessarily represent Aspen's views. The article reflects the opinion of the author at the time it was written taking into account market, regulatory and other conditions at the time of writing which may change over time. Aspen does not undertake a duty to update these articles.