Medium range probabilistic prediction of severe weather warnings

For some applications, the issuing of a severe weather ("flash") warning constitutes an event that impacts a user - independently of the accuracy of the particular warning. Such warnings are generally issued only at short lead times and thus tend to be accurate. In addition, as these warnings are issued by the official meteorological service, not taking precautionary action has regulatory and liability implications for some users. The cost of having to take precautionary action in response to a severe weather warning might be mitigated if a probabilistic forecast of the likelihood of such a warning being issued was available in the medium range. Currently the National Severe Weather Warning Service issues warnings if the probability of a Flash warning exceeds 60%, but some users have cost-loss ratios much lower than C/L=0.6, so warnings at lower probability thresholds will be of value to them. An advanced mitigating action which itself has a low cost-loss ratio, might lead to a reduction in the cost and/or loss associated with taking precautions at the last minute. For example, if there is a 20% chance of flooding in your area in 3 days time you probably wouldn't start filling sand-bags - but you might plan to be at home in 3 days time just in case. A model output statistic approach, combined with "best member dressing" is presented for generating probabilistic forecasts of severe weather warnings issued by the Met Office.