Identification of impactful storms in the UK using machine learning

Every year, the UK is impacted by major storms brought by the jet stream and associated Atlantic storm track. Exceptionally severe storms are well remembered by the public, including the ‘Burns’ Day Storm’ of January 1990, the ‘Boxing Day Storm’ of December 1998, and the ‘Great Storm’ of October 1987, all of which had impacts from extreme wind gusts. More recently, the UK experienced an exceptionally long period of stormy weather over the winter of 2013/2014 that has been described as the stormiest season in the UK since 1871.

Since 2015, impactful storms have been named in the UK by a storm naming group made up of the UK Met Office, the Dutch weather service KNMI, and the Irish weather service Met Éireann. The decision to name a storm is based on its forecast impact, not meteorological conditions, to allow for clear messaging around public warnings. This contributes to the complexity in analysing storms, as there are no definitive criteria for classifying a weather event as a storm. Additionally, trends in the frequency of storms in the UK can not be robustly examined because the record of named storms does not extend far enough back in time. 

Climatological analysis and comparison of storms is complex due to the number of variables involved: mean wind speed, maximum wind gust, wind direction, storm duration, spatial extent and storm track, all of which could have a bearing on how one storm could be judged to be “worse” than another.

This work aims to address that by examining the historical record with machine learning techniques. I explore the use of supervised classification models for identifying storms in the historical record pre-2015 based on station observation data and using the named storms as a guide. The model uses patterns in daily windspeed, rainfall, and MSLP across the UK to probabilistically classify each day in the test period as having a storm or not. This expands the catalogue of known storm dates in the UK, allowing further analysis of how storm frequency and severity have changed through time.