Report on the predictability of weather patterns and regimes of relevance for the case study applications

This report describes a series of investigations undertaken in the SECLI-FIRM project on the use of weather types or regimes in seasonal forecasts. Each section describes a different approach to the construction and application of weather types, based on different data sets and algorithms, and in the context of different Case Studies.

Section 2 describes work using a system of weather types currently used operationally at the Met Office for sub-seasonal forecasting. A methodology is developed to utilize these weather types on seasonal timescales, to forecast quantities relevant to UK winter electricity supply and demand, in support of SECLI-FIRM Case Study 8. It was found that only the weather types representing the North Atlantic Oscillation (NAO) pattern could be forecast skilfully, and furthermore, this only resulted in significant skill for forecasting wind speed and not temperature. There was therefore no benefit in terms of skill to using weather types over a simple NAO index in this context, and in particular they would not be useful for seasonal forecasts of electricity demand based on temperature.

Section 3 was also led by the Met Office, and tests a different weather regime-like method – canonical correlation analysis on a principal component analysis, CCA on PCA – to produce forecasts of significant wave height (SWH) in the North Sea, at long lead times, supporting Case Study 7. It was found that there is scope for forecasting mean SWH in May in the southern North Sea, from 1^st March, using wind speeds as the predictor variable, which could be used to extend the summer operational windows for asset maintenance.

Section 4 investigates the use of a statistical bias correction and downscaling method, ADAMONT, in conjunction with weather regimes used operationally by Météo-France. The ADAMONT method on its own is shown to considerably improve the skill of seasonal forecasts of temperature, wind and precipitation. However, conditioning the ADAMONT downscaling on weather regimes does not yield any improvement in skill, due to the lack of skill in forecasting the weather regime frequencies themselves. Methods are discussed for improving the weather regime forecast skill.

Section 5, led by UL, explores how well weather types based on k-means clustering are able to represent the different wind regimes observed in Spain and Italy, in support of Case Study 4. Although the clusters are able to distinguish different distributions of daily PMSL, and the mean wind speed patterns in each cluster are distinct, the distributions of daily wind speeds in each cluster are very broad, and the clusters do not separate different regimes. Furthermore, as the distributions are very skewed, the mean wind speeds in each cluster are not representative of the most frequent wind speeds in those weather regimes. This means that this method on its own would not be helpful in forecasting within-season wind variability in Spain and Italy, regardless of the skill of seasonal forecast systems.

Overall, this report demonstrates an array of techniques for using North Atlantic/European atmospheric circulation patterns to make probabilistic forecasts of quantities of interest to users at long lead times. Under perfect forecast conditions, weather regimes could benefit forecasts in some cases, for example by allowing better quantification of within-season variability. However, this is not straightforward or universal. Furthermore, the limited skill in seasonal forecasts of weather regimes means that any possible benefits typically fail to be realised in practice. Nevertheless, the door remains open to alternative regime-based approaches in specific cases, tailored to specific user needs.