Predictability of temperature extremes at interannual and decadal timescales

Temperature extremes, with significant impacts on health, infrastructure, agriculture, and ecosystems, represent a critical challenge. This thesis addresses the predictability of temperature extremes across different forecast horizons, offering insights into the potential for actionable climate information. Through two distinct studies, this research evaluates the skill of climate prediction systems in forecasting changes in the statistics of temperature extremes and investigates the underlying sources of predictability. The first study assesses the decadal predictability of European temperature extremes. In addition to standard grid-point metrics, the study also introduces a spatio-temporal analysis of extreme events. Results show substantial skill in predicting summer heatwaves in Southern and Central Europe and winter cold spells in Eastern and Northern Europe, with much of the skill attributable to the long-term trends. Nevertheless, after detrending the data, residual predictive skill is found in some regions, such as Scandinavia, suggesting that there are potential contributions to predictability from signals beyond long-term trends. The study also examines the models’ ability to reproduce large-scale circulation patterns linked to extremes, finding reasonable spatial agreement but underestimation in the intensity of key features, such as high-latitude blocking. The second study focuses on the interannual predictability of temperature extremes across different forecast seasons. The analysis reveals significant predictive skill across many areas, with skill generally decreasing with forecast time but remaining significant in several regions even into the second forecast year. After removing the externally forced signal, skill persists primarily during the first year, especially in tropical regions, indicating a contribution from internal climate variability. The role of ENSO is found to be particularly important though other sources of predictability may also play a role. Together, these studies advance our understanding of the predictability of temperature extremes at seasonal to decadal timescales. The findings highlight both the potential and limitations of current prediction systems.