Atlantic hurricane season is officially here; it started yesterday and will continue through November 30. This is the time when most tropical cyclones in the Atlantic form, but not all. Take for instance, Tropical Storm Arlene, which was short-lived, weak and didn’t make landfall; however, she was the second named storm to occur in April since we’ve used satellites to observe ocean-bound storms.
Hurricane Season 2017 Forecast
Many centers around the world issue forecasts for the Atlantic hurricane season using a variety of methodologies that give us an idea of the number of storms to expect. Ranges are typically used, as formation of tropical cyclones is complex and small changes in atmospheric or ocean conditions can be multiplied by the non-linear processes involved in tropical cyclone intensification.
Every year, our Willis Research Network fellow at the National Center for Atmospheric Research, Dr. James Done, gathers a variety of forecasts, and produces a document to describe the various factors informing different predictions. His latest summary was released earlier this week.
The general view is we’ll see near-normal or slightly above normal hurricane activity. The average of the forecasts reviewed in Dr. Done’s document includes 11.9 named storms, 5.2 hurricanes and 2.2 major hurricanes. This is based largely on an expectation for a weak El Niño phase of the El Niño-Southern Oscillation (ENSO) during hurricane season, which normally makes conditions less favourable for hurricane formation.
However, in the fast moving world of forecasting, new views are released frequently. In fact, since Dr. Done’s summary was released, Tropical Storm Risk has already updated their numbers and increased their prediction slightly to be more in line with NOAA’s figures.
Much of the reasoning behind the forecast is down to the expectation that ENSO influence will be weak this year, leaving the door open for other factors such as the Atlantic Multi-decadal Oscillation (AMO), the Madden-Julian Oscillation (MJO) and the Gulf of Mexico loop current, to have a greater influence. A balancing act of different influences makes seasonal forecasting a challenge, and an understanding of the range of uncertainty of probabilistic forecasts is essential.
Seasonal Forecast make-up
Seasonal forecasts work by predicting a shift in the distribution of possible hurricane activity, based on influences such as ENSO, Sea Surface Temperature in the Atlantic, the MJO and many other factors. Some forecasts utilise statistical analogues, i.e. knowledge of similar conditions in past years, to make predictions, while other forecasts are grounded in outputs from general circulation models (GCMs). These outline the physical dynamics of the atmosphere, and are therefore capable of producing conditions not recorded in the relatively short observational record.
Internal variability (the variability arising from the nonlinear interactions and feedbacks within the climate system) accounts for a large portion of variation of annual seasonal hurricane activity. By understanding the dynamics of the climate system, such as ENSO, we can start to develop predictions based on the expected atmospheric conditions, which offer greater skill in forecasting hurricane numbers. The latest GCMs are steadily improving, with long range skill growing despite the ‘Spring Predictability Barrier’ (a lull in ENSO forecasting accuracy). As forecast skill in the seasonal and annual timescale improves, more opportunities to use the guidance available from these forecasts may develop.
Recap on last year?
Last season’s activity was near-average with 15 named storms, 7 hurricanes and 3 major hurricanes, slightly above the averages of 12, 6.4 and 2.7, respectively. There were also a few significant storms for a number of reasons. Hurricane Hermine was the first to make landfall in Florida since Wilma in 2005; Hurricane Nicole was the first major hurricane to hit Bermuda since Fabian in 2003 and Hurricane Otto was the first on record to hit Costa Rica directly. All three storms caused multiple deaths and resulted in hundreds of millions of dollars in damage.
The most destructive storm was Hurricane Matthew, which hit the Greater Antilles, Venezuela and Colombia, and affected much of the east coast of North America. The storm peaked at category 5 on the Saffir-Simpson scale while in the Caribbean Sea, then moved north towards the U.S. It killed around 600 people and caused upwards of $10B in damage. With claims still being settled, estimates of insurance losses in Florida alone are around $928M (according to the Property Claims Service).
Using Season Forecasts
“The pessimist complains about the wind; the optimist expects it to change; the realist adjusts the sails.” – William Arthur Ward.
We have a choice when managing windstorm risk. Consider a scenario in which we see a significant early season storm affecting an area of high exposure. This will cause an insured loss, but as a single event it will be unlikely to consume all of the cover afforded by a reinsurance contract. At this point, we can either do nothing, or react and ‘adjust the sails.’
If we do nothing, we may simply expect no further events to occur, and/or rely on the cover that has been set ahead of the hurricane season to absorb any remaining losses. If we react, we might think about options involving reinsurance reinstatements, or off-setting storm risk via the ILS/cat bond markets through the rest of the season. Either way, decisions rest on tried and tested methods within the reinsurance process based on detailed catastrophe modelling and reinsurance structuring.
However, choosing not to react may make the rest of the hurricane season rather uncomfortable. We can make these decisions easier by utilising our best knowledge of seasonal forecasting to adjust pricing and capital allocation.
To accompany the seasonal forecast, Dr. Done recently delivered a webinar, hosted by Prasad Gunturi (Willis Re), highlighting the latest forecasts and describing some of the research that supports these methods. In the webinar, which can be accessed via passcode WRe2017, he noted research led by Dr. Kerry Emanuel, which tests a hypothetical property insurance portfolio. The paper shows the value of a variety of strategies in which the reinsurance price varies according to active or quiet expectations for hurricane season, as opposed to the same price every year. Continued use of seasonal forecasts is shown to have significant value after around a decade.
Whatever this season brings, it’s worth remembering that it only takes one major hurricane landfall in vulnerable or highly exposed areas to cause significant impact. Seasonal forecasting can be useful for understanding the background environment, perhaps through the use of indexes such as a version of the cyclone damage potential index, as developed via the Willis Research Network by Dr. Done and team. However, seasonal forecasts will not predict individual storms, and should ideally be applied to a longer term strategy.
By Geoffrey Saville, with contributions from James Done, Prasad Gunturi and Roy Cloutier.