Winter storms and hurricanes, all of an afternoon

The Willis Research Network workshops and seminars are always interesting events, and this year’s London seminar was no exception. Held on the 22nd of September, we focussed on wind extremes, specifically seasonal hurricane forecasting and simulating European winter-storm footprints. We need little reminder of the importance of this science, so soon after Hurricane Matthew tore a path through Haiti and eastern Cuba, and across the Bahamas to affect Florida and the east coast of the U.S.

Probabilistic hurricane forecasting

Dr Gabriel Vecchi from the Geophysical Fluid Dynamics Laboratory walked us through the science of hurricane forecasting. We are used to seeing the short-term forecasts describing the details of individual events as they develop and progress, but he explained some of the key differences between this and one of his fields of research – long range seasonal hurricane forecasts.

In a long-range seasonal forecast we should not think deterministically, but rather probabilistically

He explained how in a long-range seasonal forecast we should not think deterministically, but rather probabilistically. A forecast given by a long-range climate model differs to that of a short-range weather model, as it expresses the likelihood of all possible outcomes rather than just one solution.

Seasonal forecasts can be tricky to interpret in a day-to-day context, largely because of the difference in the type of information offered when compared to normal weather forecasts. Seasonal hurricane forecasts express this spread of possibilities by phrases such as “above-normal”, “below-normal” or “near-normal”, with “normal” referring to the average count of storms.

They also indicate a range for the number of storms predicted, rather than a single number. This is because of the internal variability within the climate system, that is, the ways in which the climate varies of its own accord to hurricanes, through such phenomena as the El Niño Southern Oscillation. This range is a way of describing the level of natural variability and uncertainty within each forecast.

By the very nature of this probabilistic approach, a seasonal forecast should be wrong sometimes, otherwise you are probably not doing it correctly. If there are very strong signals for one end of the distribution over another, then on balance this is where the smart money should be—but it is not guaranteed.

Extremes should be expected to be part of the makeup of any distribution of events and any seasonal forecast

But when there are no strong signals, then probabilities are more evenly spread, and a forecaster will often tend to lean toward the climatological average (normally taken from a 30-year period of historical data).

Additionally, extreme seasons are, of course, expected to occur occasionally. Dr Vecchi commented that “weird things happen all the time”, when explaining how extremes should be expected to be part of the makeup of any distribution of events and any seasonal forecast.

Dr Vecchi discussed the limitations of seasonal predictability, and where the latest models are finding improved skill. His new model “HiFLOR” shows great promise and may potentially be able to show usable skill at predicting category 4- and 5-strength storms in some regions of the tropics, which hints at a possibility of landfall seasonal forecasts, as opposed to currently produced basin-wide activity forecasts.

hiflor

European winter-storm season is coming

The second half of the seminar switched the focus to Europe. While hurricane season rumbles on with attention recently focussed on Hurricane Matthew, it can be easy to forget that we’re heading into a new European winter-storm season. After an active season last year, this second session of the seminar was designed to look into the WRN projects going on at the University of Exeter with our WRN fellow Dr Ben Youngman. He presented his innovative work on stochastic generation of windstorm footprints.

European storm footprints

The stochastic generation of windstorm footprints model can simulate extreme windstorm events using extreme value theory and geostatistics

His statistical model can simulate extreme windstorm events using extreme value theory and geostatistics. Ben explained how his novel method objectively sets a threshold to identify extremes at each location based on the observed wind gust data, fits an extreme value statistical model to each location and constructs an estimate of the spatial dependence between each site. This is all done in a computationally efficient manner by fitting a kind of statistical surface, a spline, that can deform to fit the observations, in some ways it’s a bit like vacuum forming plastic.

This process is designed to build a stochastic event set for European windstorms, based on historical data and robust statistical manipulation of the data both in space, using geostatistics.

stochastic-generation

We’d like to thank again everyone who attended the event, and we’ll look forward to hosting the next seminar. As ever, we hope the content was relevant and insightful. For anyone interested in the topics described in this blog, the slides are available to view here. If you have any questions on the content, then you are more than welcome to get in touch.

About WRN

Events like these are designed to publicise some of the latest science supported through the WRN and show how we are applying them to our business users. For 10 years, since the foundation of the WRN, we have been holding events that bring together industry analysts and leading scientific experts to share knowledge and experiences. There is a perennial need to examine the cutting edge of science and explore new ways to applying it to our industry. To mark the WRN’s ten-year anniversary, we have put together a brochure to highlight some of the research projects that we have supported over the last decade.

About Geoffrey Saville

Geoffrey Saville is a member of Willis Towers Watson's Analytics Technology Team, having joined the company in 2013…
Categories: Natural Catastrophe, Property, Reinsurance, Weather risk | Tags: , ,

Leave a Reply

Your email address will not be published. Required fields are marked *