Generating weather pattern definitions over South Africa suitable for future use in impact-orientated medium-range forecasting

This work aims to define a set of representative weather patterns for South Africa that can be utilized to support impact-based forecasting of heatwave events. Sets of weather patterns have been generated using k-means clustering on daily ERA5 reanalysis data between 1979 and 2020. Different pattern sets were generated by varying the clustering atmospheric variable, the spatial domain and the number of weather patterns. These weather patterns are evaluated using the explained variation score to assess their ability to represent the variability of the maximum daily 2m temperature (T_max,2m). The results indicate that a set of 30 weather patterns generated using mean sea-level pressure, with a clustering domain in the range 15°–34°E and 21°–36°S, provides a reasonable representation of T_max,2m variability across South Africa. The implementation of an appropriate weather pattern set into a medium-range forecasting tool has the potential to extend the prediction of high-impact weather events in South Africa, such as heatwaves, and also highlight specific impacts on the population, for example, food and water insecurity, heat exhaustion or energy and transport impacts.