Severe winter windstorms are one of the most damaging extreme events over Europe. Their windspeeds regularly cause high losses by damaging property and infrastructure. The storm frequency over a season, the storm intensity, and the preferred tracks are all important information for insurance, warning systems and the general public to manage and prevent losses. Therefore, a skillful prediction of the upcoming windstorm season is of great value.

Besides the interest of better understanding extreme events like windstorms seasonal time scales are gaining interest. However, extratropical seasonal forecasts are often less predictable than weather forecasts or climate projections, not least because they are a combination of the initial value problem and the boundary value problem. The aim of this study is to investigate the skill and origin of seasonal forecasts of extreme winter windstorms.

These windstorms are tracked with an impact-based algorithm which uses the 10m-windspeed and a local threshold of the 2% highest wind speed. The extreme events are tracked in observational re-analysis and the GloSea5 seasonal forecast system of the UK Met Office. We investigate storms in the core winter from the model simulations initialised in November with its prediction for December to February. The forecast is an ensemble with multiple realisations to cover the range of potential states of the atmosphere.

The first part of the study aims to understand the performance of the forecast model itself. This study shows that the seasonal windstorm frequency is predictable, especially over the British Isles and the North Sea. This study shows for the first time that also the accumulated intensity over a winter season, based on a track-based severity index of extreme wind events, can be skilfully predicted on a seasonal scale.

For the second part of this study, we ask, why are these storm characteristics predictable and why over some regions more than others? We use the three most dominant large-scale atmospheric variability patterns over Europe: The North Atlantic Pattern (NAO), Scandinavian Pattern (SCA), and East Atlantic Pattern (EA). A statistical multi-linear regression model has been built with these large-scale patterns to understand how much of the forecast skill is explained by each. The NAO is known as the most dominant and influential pattern for European weather, but in connection to windstorms is a gap of explanatory variance seen over the North Atlantic. This gap can be filled by the other 2 weather patterns. Altogether they explain up to 80% of the interannual winter variability of windstorm frequency and up to 60% of windstorm intensity.

As it seems that these large-scale patterns are well linked to windstorms, they could also be used to predict the windstorm frequency and intensity. This could mean that end-users don’t need to deal with the direct model output but could use the predicted large-scale pattern indexes and get an information about the upcoming storm season. This has been investigated by statistically re-forecast the windstorm frequency or intensity by using a multi-linear regression out of NAO, SCA and EA. Although it is skilful, the performance of such a prediction does not reach the same skill level as the explicit model performance with the direct forecast storm output.

In the third part of this study, we examined the signal-to-noise paradox. This phenomenon is not fully understood, but it results from the counterintuitive situation of the forecast model ensemble mean being better at predicting reality than its own forecast ensemble members. This paradox is found especially in forecasts of the NAO. We showed in this study that the paradox also exists for windstorm forecasts, and this implies that seasonal forecasts may have potential for more confident forecasts of extreme winter windstorms in future.