Seasonal forecasting of tropical cyclones
With 2004 seeing Pacific typhoons and Atlantic hurricanes causing economic losses totalling US$40 billion, and Florida being hit by four hurricanes last year and another in July 2005 (Figure 1), it is hardly surprising that
Figure 1. On July 11th, 2005, Hurricane Dennis became the fifth hurricane to strike the state of Florida in eleven months. Category 3 (at landfall) Dennis was the earliest fourth named tropical storm and held the record for the most powerful early season on record for nine days, before having it stolen by Hurricane Emily. Image: courtesy NOAA.

improved short-term prediction of tropical cyclones remains a high priority for a number of research groups. While hurricane forecasts for the Atlantic Basin are becoming increasingly skilful, Eric Blake and William Gray⁷ of Colorado State University, make the point that large gaps remain in our knowledge and understanding of observed variations in the distribution of activity within the hurricane season. The authors are particularly concerned with forecasting the year-to-year variability of August tropical cyclone (TC) activity in the Atlantic, a period that spans the first third of the climatologically most active segment of the hurricane season. In a paper in Weather and Forecasting, they show that 55 – 75 percent of the variance of August TC activity can be hindcast using a combination of just four or five global predictors (figure 2). These are chosen from a pool of 12 predictors
Figure 2. Simplified conceptual summary of the key features in an idealised summer pattern prior to increased August tropical cyclone activity in the Atlantic. Courtesy: Eric Blake.

previously shown to have a predictive association with TC activity. The most prominent predictor appears to be the equatorial July 200mb wind off the west coast of South America. When this is anomalously strong and blowing from the northeast, Atlantic TC activity the following month is enhanced. Blake & Gray, stress the importance of an August-only forecast, as predicted TC activity during this month has a significant relationship with the incidence of US August TC landfall events.

Clearly, it is the forecasting of landfalling hurricanes that is of most concern, and this is an issue tackled in the journal Nature by Mark Saunders and Adam Lea⁵⁵ of UCL’s Benfield UCL Hazard Research Centre. In a similar manner to Blake and Gray, Saunders and Lea use July wind anomalies; in this case to predict, with significant skill, the wind energy (ACE = Accumulated Cyclone Energy) of US landfalling hurricanes for the following main hurricane season (August to October). The Saunders and Lea model is based upon the identification of six regions over North America and over the east Pacific and North Atlantic Oceans, where July wind anomalies exhibit a significant link to the energy of – specifically – landfalling hurricanes during the subsequent hurricane season (figure 3). The reason for the connection, they suggest, is that wind anomalies in these particular regions are indicative of atmospheric circulation patterns that either favour or hinder evolving hurricanes striking the US coast. Highlighting the importance of their study for loss prediction, the authors point out that hindcasts using their model are significantly linked to both annual economic and insured losses over the past half century. Figure 3. Tropospheric height-averaged wind anomalies linked significantly to above-median seasonal US landfalling hurricane activity 1950-2003. Wind data for July is shown in the top panel, and for August-October in the bottom panel. Plotted is the difference in vector wind anomalies (averaged between 925 and 400 mb height) between those subset years when US ACE Index is in its upper and lower quartiles. The significance of the difference in wind magnitude (P-value) between these subset years is shown by the colour bar. Seasonal predictability of the US ACE Index is assessed using a July height-averaged wind index using the six regions marked by white boxes. Courtesy: Mark Saunders.

Palaeotempestology
Going back further than the historical record, important information on storm frequencies and trends can also be acquired by looking for geological features associated with storm activity. The study of such features forms the basis of the science of Palaeotempestology, which seeks to increase the size of temporal length of the storm catalogue through the interpretation of geological proxy evidence at affected coastlines. In a new volume on Hurricanes and Typhoons: Past, Present and Future, edited by R. Murname and K. B. Liu⁴⁵, Kam-biu Liu³⁷ of Louisiana State University provides an excellent introduction to the subject, and its application to past hurricane frequency along the US Gulf Coast. Here sand deposits contained in coastal lake and marsh sediments provide a proxy record of major hurricanes going back some 5,000 years. The record suggests a return period of around 300 years for a catastrophic hurricane (category 4 or 5). It also supports a relatively quiet period of low hurricane activity over the past millennium, preceded by a ‘hyperactive’ episode between 1,000 and 3,400 years ago.

In a second paper in the volume, Jeffrey Donnelly of the Woods Hole Oceanographic Institution and Thompson Webb III¹⁵ of Brown University examine geological evidence for intense (category 3 and above) hurricane landfalls in the north-eastern US. The storm surges associated with powerful hurricanes remove sediments from beach and near-shore environments and dump them in normally quiescent back barrier environments that are typically occupied by marshes or lakes. The record of storm sediments preserved in these environments, the authors point out, provides the best potential for developing long-term records of intense hurricane activity. Amongst other conclusions, Donnelly and Webb’s studies reveal that at least six intense hurricanes made landfall in southern New England in the last 700 years, yielding a return period of around 116 years and an annual probability of about 0.9 percent. This, they stress, is higher than that obtained by simply looking at the storm record over the past century.

Historical studies of storminess
A special volume of the journal Marine Geology devoted to the theme of Storms and their Significance in Coastal Morpho-sedimentary Dynamics (edited by G. W. Stone and J. D. Orford⁵⁸) provides a wealth of papers on the history of storm activity in the North Atlantic region. Worthy of particular note is a wide-ranging study by Barry Keim³³ of Louisiana State University and colleagues, which focuses on the spatial and temporal variability of coastal storms in the North Atlantic Basin. Keim and his colleagues recognise a decadal-scale variability with respect to both tropical and extra-tropical storms, neither of which can yet be linked conclusively to natural or anthropogenic forces. The authors note that tropical storm frequencies have declined over the past few decades, perhaps – they suggest – as a consequence of recent intense and prolonged El Niño – Southern Oscillation (ENSO) events. They also point out that there is a strong suggestion that extratropical storm systems have declined generally over the past 50 – 100 years, but that there is an increase in frequency of very powerful storms, particularly at higher latitudes. Both the frequency and tracks of such storms are associated with ENSO and the North Atlantic Oscillation (NAO) conditions.

In another study of historical storminess in the North Atlantic region, published in the same volume of Marine Geology, Alastair Dawson¹³ of Coventry University and co-authors identify a link with what they refer to as climate ‘see-saws’. Such climate variations are characterised by the coincidence of severe winters in western Greenland and mild winters in northern Europe, and vice versa. This temperature ‘sea-saw’ is also reflected, say the authors, in records of historic storminess from Scotland, NW Ireland and Iceland. Over the past 150 years, the stormiest winters in these regions have occurred when western Greenland temperatures have been significantly below average, while less stormy winters have been associated with mild conditions in western Greenland. Looking back further, Dawson and colleagues propose that winter storminess in the regions examined was at its lowest during the Medieval Warm Period, which ended in the early 15^th century. In contrast, the period from around 1420 to the present has been characterised by sustained winter storminess across the North Atlantic.

European windstorms
Two papers in the aforementioned Marine Geology volume address storminess in the eastern Atlantic region. In the first, I. Lozano³⁸ of the University College Cork and co-authors analyse storm records to provide a measure of storminess and vulnerability along the Atlantic coastlines of Europe, including those of the UK, Ireland, France, and northern Spain and Portugal. The study examines the evolution in the occurrence of extratropical cyclones that have affected the Atlantic margin since 1940 and their relationship with the NAO. Lozano and colleagues note that the total number of cyclones crossing the North Atlantic has undergone a slight fall over the period 1965-1995 (figure 4), which is consistent with a longer term pattern of decreasing Atlantic storm numbers since 1900. For the UK and Ireland, the authors recognise the tendency towards a two-season annual storminess pattern, with quieter summers followed by more stormy winters, and attribute this to a northward displacement in the main North Atlantic cyclone track. Looking ahead, Lozano and colleagues forecast that by around 2060, the northern part of the study region, covering Ireland and Scotland, will have a tendency to be struck by fewer but more intense storms.

Figure 4. The paths of winter cyclones with a minimum wind speed record of 40 knotts affecting European Atlantic coastal zone 1 (the UK and Ireland) between 1965 and 1994. Over this period, the total number of cyclones crossing the North Atlantic underwent a slight fall. Courtesy: Robert Devoy.

In the second paper, N.L. Betts⁴ and colleagues at Queen’s University, Belfast, examine the synoptic climatology of recent extreme coastal storms in the South- Western Approaches to SW England and northern France. The authors adopted cluster analysis to reveal discrete cyclone track regimes linked to upper airflow patterns as being responsible for the formation of intense storms (central pressure at sea-level _‹ 990 mb) that also promote severe (_› 60 cm) surges along the French coast of the South-Western Approaches. Betts and co-workers note that fluctuations in storminess are strongly influenced by the southward intrusion and strengthening of the jet stream in the mid-Atlantic. Such occurrences are often linked with negative sea-surface temperature (SST) anomalies near Newfoundland and the strengthening of the thermal gradient across the Atlantic well to the south of its normal position. The authors show that the most influential variables in promoting storms characterised by severe surge events in the South-Western Approaches are trans-Atlantic SST gradients and particularly the prevailing west-east SST gradient during the month of the storm.

While most windstorm losses in any single year are typically dominated by tropical cyclone activity, the larger footprints of extra-tropical storms can result in serious damage over a considerable area. Europe is particularly susceptible, as evidenced by economic losses due to windstorm averaging US$2.9 billion a year between 1990 and 1999. A paper published in the journal Weather by Paul Blackmore and Elina Tsokri⁶ provides an interesting insight into how the UK’s Building Research Establishment (BRE) collates information on windstorm damage to buildings and structures, and how it classifies windstorms on a damage basis. Data is gathered primarily from newspaper reports, supplemented by information supplied directly by local authorities and other owners of substantial building stocks. Storms are categorised in terms of the numbers of locations (cities, towns or villages) at which damage is reported (TABLE 1); a severe gale being a windstorm that causes reported damage at more than 500 locations. Estimates of the financial cost of wind damage are based upon precedent, which arrives at an average of £380 per damage incident (determined from claims data). For the year 2002, which forms the focus of the paper, this gives an estimated repair cost of £122 million, although consequential and secondary losses will probably double this figure. Over the 41 years during which data has been collected, the average number of damage incidents is approximately 164,000 per year. In 2002, the total number of incidents – at 367,500 – was more than twice the long-term average. Also unusual was the fact that wind damage was reported for every month of the year.

Table 1. Windstorm events which caused damage in the UK over the period 1962 to 2002

Other noteworthy papers
An approach, which is becoming increasingly commonplace in weather forecasting in general, involves the use of so-called ensemble forecasts. These are produced by statistically evaluating a large number of forecasts, derived either from a single numerical model run with different initial conditions, or from a portfolio of independent numerical models. The resulting forecast is superior in quality to forecasts made using the individual numerical models. A special issue of the journal Tellus was devoted to this forecasting technique; focusing on studies arising out of the EU-funded Project DEMETER (Development of a Multi-model Ensemble System for Seasonal to Inter-annual Climate Prediction). The volume, edited by Tim Palmer⁴⁸, contains 21 papers, many of which would be worth accessing for anyone with in an interest in state-of-the-art seasonal forecasting and its applications.

While multi-model ensemble approaches may be improving forecasting, there is still some way to go. The UK Met Office, for example, has traditionally rated its next-day forecasts as being 85 percent accurate. Although this sounds impressive, forecasts made simply by assuming that the weather tomorrow will be the same as today, achieve 77 percent. So much uncertainty surrounding weather forecasting, reports Jim Giles²¹ in Nature, has led to the growth of many numbers of companies – large and small – who aim to provide a forecasting service, some of which are used by the insurance sector. There is, however, no guarantee that the products of some of these companies are worth the paper they are written on. In order to bring some order to the chaos and limit the impact of rogue companies, the UK’s Royal Meteorological Society is developing an accepted set of metrics designed to rate the accuracy of different forecasting companies. The Society recognises that this may take some time, but the end product should be beneficial to insurers and reinsurers with an interest in taking advantage of the range of forecast supply options available.

«back to top«