On 11 October, after prolonged rainfall the River Cherwell (along with many other rivers in the English Midlands) broke its banks and the railway station at Banbury was flooded to a depth of two feet. Crowds of people turned up to see the trains passing through the water. Nine days later the station had flooded again, with water rushing between the platforms in a huge stream. By 21 October the water was once more two feet over the rails. On 7 November the station was once flooded again, and after further prolonged rainfall, by 14 November the water was as high as on any of the previous occasions.

The year was 1875. No one thought to blame the persistent flooding on climate change.

The surface of the northern Atlantic Ocean preserves its summer warmth into October and consequently frontal systems can carry far more rain than in the depths of winter. During the second half of the 19th century - in November 1852, November 1875 and November 1894 - there was widespread flooding across England after a series of depressions arrived and parked themselves over the country. There have been fewer such episodes during the second half of the 20th century and then, at the beginning of the 21st century, the situation returned. Floods deluged northern, western and south eastern England, resulting in an insurance loss approaching £500m. Is this then a symptom of climate change? Should catastrophe modellers be abandoning the use of history as the means to calibrate the recurrence of extreme events? Should reinsurers be rebalancing their portfolios to accommodate a new hazardous world?

One measure is certain: the globe has been getting warmer. The last decade has been the warmest for 1,000 years. The year 1998 was the warmest year on record. Some of the consequences of a rise in temperatures are straightforward. In going short on a Chicago Winter Heating Degree Day temperature derivative put contract, basing a pricing analysis on the mean temperatures over the past 40 years, is likely to lose you money because recorded temperatures have risen over this period.

However, most atmospheric catastrophic loss events are not simply a consequence of a rise in temperatures, but reflect some derivative of this rise. Perhaps the simplest of these derivative responses concerns precipitation. Warmer temperatures lead to increased evaporation and consequently raised precipitation. In the US, precipitation amounts have risen on average by around 10% over the past century. However, the affect is not uniform: there has been a significant increase in winter rainfall over Scotland during the 1980s, but no demonstrable increase in extreme rainfalls in Wales. In Germany, winter rainfall has increased with the arrival of more westerly weather systems. Along the Rhine in Germany and Holland there were two “50-year return period” flood events during the mid 1990s. Even with flooding, the global warming picture is not so simple: in England, the largest inland flood during the 20th century occurred in 1947, the result of the rapid thaw of a snowpack accumulated over two months of extreme low temperatures. This is just the kind of event that in a warmer world is very unlikely to recur.

In using historical flood statistics, it is important to recognise that the catchment into which the rain falls is not the same as it was 100 years ago. Increased development on flood plains, changes in farming and urbanisation that increase run-off, combined with channel dredging and embankments that reduce the friction of the river-channels all tend to increase the height of flood-waves as they pass downstream. For all these reasons flood risk is rising and consequently river flood catastrophe loss is likely to provide an increased component of the risk borne by re/insurers in many countries where flood is a general insurance coverage.

What of other atmospheric catastrophes with the potential to impact reinsurance? In making the connection between temperature and precipitation, the situation is mediated by changes in climatology. The most important large-scale climatological process on the planet is the ENSO (El Niño Southern Oscillation) phenomenon, in which warm ocean water spreads across the equatorial Pacific. The most striking feature of global climate over the past 25 years has been the increased incidence of El Niño warming events, raising the eastern equatorial Pacific Ocean temperatures. Over the past century, there have been an approximately equal number of El Niño events and its opposite La Niña (when the Eastern equatorial Pacific is cooler than ususal on average). However since 1976 there have been seven El Niño events and only three La Niñas. Also in this period there have been the two biggest El Niños on record. The first in 1982-3 was exceeded by the strength of the 1997-8 El Niño, when sea surface temperatures over wide areas of the central and eastern tropical Pacific were more than 50C higher than usual. The warming extended widely over the northern and southern hemispheres: global temperatures in February 1998 were further above average than in any previous month.

El Niño years provoke more vigorous tropical cyclone activity in the central and eastern Pacific (in particular off the Pacific coast of Mexico), but reduce the activity in the Atlantic. Therefore global warming appears to have increased the proportion of years with a small number of Atlantic hurricanes, although there has been some offsetting increase in the number appearing in non-El Nino years, back closer to the levels seen in the middle part of the 20th century. Nevertheless, Hurricane Andrew in 1992 is a reminder that an El Niño year is not a watertight guarantee of the absence of a major catastrophic hurricane loss.

For Atlantic-origin extra-tropical cyclones (the windstorms that affect Western Europe) the picture is also far from simple. These storms form along the Polar front where the cold Arctic and warm equatorial air masses meet. The stronger the temperature contrast at the front, the greater the potential for storm intensification. Yet as Arctic winter temperatures have risen faster than those at middle latitudes, the temperature contrast might be predicted to fall. The decade from 1981-1990 was a very active period for intense extra-tropical cyclones arriving in western Europe, but during the years of record annual temperatures through the 1990s, windstorm activity in Europe was unexceptional. Wave-height observations show that windspeeds over the north east Atlantic increased from the 1970s through the 1980s, along with the population of extreme storms, but during the mid 1990s wave heights decreased again. Activity picked up in the final month of the decade, throwing three small and intense storms, Anatol, Lothar and Martin, into western Europe. Are these storms once again part of a rising trend as witnessed during the 1980s, or is this just an isolated outbreak?

To observe trends over longer time periods one can employ pressure observations, reliably measured for more than 200 years. If storms are getting stronger then one would expect that the minimum pressures reported at key locations across Europe would have occurred during the past few decades. However, across north west Europe the minimum pressure records were almost all set before 1900: in Belfast 928mb in 1886; Shetland 922 mb in 1839; Bergen 937mb in 1822; Perth, Scotland 925 mb in 1884; Chester 935mb in 1735; London 948 mb in 1821; Paris 951mb in the same storm; and Orléans, France 959mb in 1896. At Hamburg, the 1972 Lower Saxony storm beat the previous record set in 1825 by 1mb.

There is certainly no simple correlation between a warmer climate and the occurrence of extreme windstorms impacting Western Europe. Historically, many of the most destructive windstorms have occurred at periods when the general climate was colder, including some of the most intense known windstorms that occurred during the Little Ice Age, in 1836 and 1839, 1703 and 1581. At colder climatic periods the Arctic front will tend to be further south than its current average location between Scotland and Iceland.

So while droughts, periods of extreme heating and flash floods are all widely predicted to be on the increase, the picture for the most catastrophic hurricanes and extra-tropical storms remains cloudy. Not that this will prevent reinsurers for Denmark and France raising rates after the December 1999 windstorms. Yet before assuming that catastrophe risk has increased by some unknown amount from the long-term baseline, we need to study statistically the nature of the evidence for an increase in the extremes, against a well-researched historical record. It is no longer possible to have any unusual atmospheric event without it being claimed by some politician or journalist as “proof” of climate change. As an antidote, it is always worth contemplating some of those events that have not recurred recently, such as the Seine flood in 1910, when four cubic kilometres of water flowed through the streets of Paris and 150,000 people had to be evacuated, or the catastrophic floods on the river Tarn in south west France in 1930 that flooded Montauban to the depth of almost 7m, drowning 200 people, destroying 11 major bridges and putting 500 factories out of action.

While insurance and economic losses provide a convenient way to summarise the impact of disasters, changes in exposure and vulnerability are an order of magnitude greater than any change in the occurrence of the events themselves. The main difference with the experience of October-November 1875 and October-November 2000 in England is that in 1875 the trains continued to run, speeding through the station to prevent the fire boxes being extinguished.

Robert Muir-Wood is head of global risk modelling with RMS, London.