Features, December 1999/January 2000 The appliance of science Academic researchers are pushing back the frontiers of climate forecasting. The insurance industry is watching closely, reports Sumit Paul-Choudhury They can't tell you if it'll be sunny next week - but they may be able to tell you how likely you are to be blown away by a hurricane next year. Academic meteorologists are making great strides in their efforts to push back the horizons of weather forecasting. Last November, researchers at the Benfield Greig Hazard Research Centre (BGHRC) of University College London (UCL) published one of the longest-range forecasts of US hurricane strikes yet announced - stretching forward a full year, to the 2000 hurricane season. For the record, 2000 will see slightly more storms in the Atlantic and making landfall in the US than an average year, they predict. The UCL team has also published long-range forecasts for the likelihood of winter gales in Britain. If forecasts like these turn out to be reliable, they may offer companies a rare insight into the impact that weather could actually have on their businesses in the future - and the chance to do something about it in advance. The insurance industry, in particular, has been closely involved with the work. The catastrophic losses of the early 1990s did much to wake insurers up to the fact that traditional techniques for estimating the risks associated with a natural catastrophe were not enough. Since then, the industry has adopted an increasingly quantitative, rigorous approach which attempts to anticipate, rather than react to, catastrophic events. Accurate long-term forecasts, together with computer simulations of catastrophic events, could help to bring about this shift in emphasis from retrospective to prospective. Although long-range forecasting has been tried before, it has mostly lacked scientific credibility. This time, the work has been endorsed by the UK Meteorological Office and is based on a full-blown climatic model, rather than being entirely based on patterns deduced from historical records. "We're confident, based on hindcasting back to the mid-1980s, that our results are statistically significant," says Mark Saunders, head of the UCL team. Hindcasting (also known as back-testing) is the process whereby a theoretical model is fed historical information and its output compared with what actually happened. In December 1998, for example, the same researchers - Saunders and his colleague Paul Rockett - predicted that an above-average total of four tropical storms, two hurricanes and one intense hurricane would make landfall in the US during the 1999 Hurricane season. The actual total: five, three and one, respectively. Saunders and Rockett also predicted that there would be 12 Atlantic tropical storms and seven hurricanes, a good fit for the 12 tropical storms and eight hurricanes actually observed. "A year ahead, we can predict about 30% of the variability in the total number of hurricanes, and about 15% among in terms of the number of storms that actually hit land," says Saunders. "That represents a fair bit of skill compared with what you would expect by chance." Why have such forecasts only become possible now? One part of the answer is increased computational power, which makes the analysis possible; another is the application of non-traditional mathematical techniques. Another part is cultural: a realisation that the received wisdom on climatic systems ­ that they are so complex that the forecasts can only be made reliably up to a week or so into the future - is not universally true. "For more than 30 years, 'chaos' has been the dominant paradigm to explain the limits of weather forecasting," wrote Jagadish Shukla, a professor at George Mason University in the US, in a 1998 article in the journal Science. "It has now been demonstrated that there are important exceptions to this chaotic behaviour and that certain aspects of climate are far more predictable than previously thought." The Atlantic storm and winter gale forecasts produced by the UCL team fall into one of these exceptional areas. Both forecasts use sea temperatures in the Atlantic as predictors, which change much more slowly than other climatological variables, imparting significant predictability to atmospheric development. The forecasts do not predict specific events; that is, they don't claim to give timings (or, in this case, locations) for the storms. Rather, they offer information about how the likelihood of storms in a given season compares with the historical average. The UCL group and other researchers hope that it may be possible to make similarly probabilistic forecasts of variables such as winter and summer temperatures on land. Limited though this information might at first seem, it may nonetheless prove extremely valuable to businesses affected strongly by the weather - which is to say a large segment of most economies. Most obvious among the potential users of this work are those who ultimately end up managing the risk. In some cases, these are concerns such as energy and agricultural companies, whose businesses are directly exposed to the weather. The largest concentrations of risk, however, reside with those who ultimately provide insurance to such companies. Those might be traditional insurers and reinsurers, or the speculative investors in new financial instruments, such as "catastrophe bonds", whose payments are linked to the occurrence of catastrophes (see Environmental Finance October 1999, page 28) and weather derivatives, whose pay out is linked to prespecified weather conditions. Vast sums are at stake - it has been estimated that severe hurricane in Florida could generate as much as $60 billion of insurance claims - so even a slight improvement in the risk estimate could be extremely valuable. FOR THE FULL STORY EVERY MONTH, SUBSCRIBE TO ENVIRONMENTAL FINANCE, OR CONTACT info@environmental-finance.com FOR OUR BACK ISSUES SERVICE