An investigation into the behaviour of North Atlantic cyclones has been conducted and a GIS-based windstorm hazard model developed that can predict patterns of damage across the UK. The underlying rationale is based on the perceived increase in the frequency of severe windstorms in parts of Europe over the past few decades. Examples include the ‘hurricane’ of October 1987, which devastated southern England and resulted in an insurance pay-out of ~£1.2bn, and the gales of January to March 1990 which produced average wind speeds exceeding 32.7m/s. The storm of 25th January 1990, in particular, produced wind speeds greater than 45m/s, and killed 47 people in the UK alone. Subsequent winters have been characterised by storms of this nature with high wind speeds and heavy precipitation. Examples include the severe winter storm of Christmas 1997 in the UK; and the series of storms including Lothar and Martin that affected northern France in December 1999.

Two objective cyclone climatologies of the North Atlantic, spanning a total of 39 winters October – March 1958-1996, have been constructed from mean sea level pressure data from the ECMWF and NCEP/NCAR reanalyses. A comparison of the two climatologies shows some clear differences. In particular, the NCEP/NCAR model does not replicate as many weak systems as the ECMWF model, especially over Greenland.

A detailed investigation of cyclone distribution across the North Atlantic has been carried out on the NCEP/NCAR climatology using Principal Component Analysis. This has identified a general northward shift in the dominant SW/NE orientated cyclone track across the region over time. The relationships between major atmosphere-ocean mechanisms such as the NAO, AO, the position of the Gulf Stream and the Southern Oscillation has also been explored.

Building on the cyclone climatology, a GIS-based windstorm damage model has been developed. This is based on domestic buildings insurance claims data for the UK at the postcode district level and is initialised with damage information from the October 1987, January 1990 and February 1990 storms. The predictor variables include wind speed, mean elevation of the postcode district, and distance of the postcode district from the storm track and also from the point of landfall. As a result of uncertainties regarding the reliability of the February 1990 claims data, it has not been possible to construct a generic damage model. However, ‘one storm’ models for the October 1987 and January 1990 storms have been constructed which can realistically be applied to future storms with characteristics similar to either of these two events.