From Weather to Climate: Quantifying Long-term Trends in Extreme Weather using Medium-range Weather Forecasts

Olivia Vashti Ayim

The frequency and intensity of some extreme weather events, such as heatwaves, are increasing due to climate change; however, short observational records limit our ability to assess these changes with statistical confidence. In this study, a large ensemble of high-resolution re-forecast simulations from the European Centre for Medium-Range Weather Forecasts' Integrated Forecasting System was used to examine how extreme temperatures respond to local, regional, and global warming, with a focus on the Pacific Northwest. At longer lead times, forecasts are less dependent on initial conditions and more influenced by large-scale climate drivers, making them well-suited for estimating the likelihood of extreme events. Results show that the re-forecast ensemble captures approximately the same day-to-day variability of daily maximum temperatures as ERA5 reanalysis but provides a larger sample size, thus allowing more robust estimation, especially in the tails, reflecting both observed events and plausible alternatives. Compared to the ERA5 and CMIP6 models, the re-forecast data set provides a physically consistent and robust estimate of how the probability of exceeding extreme temperature thresholds changes with warming. We identify statistically significant relationships between the re-forecast probability of local extreme temperatures in the domain of interest and both local and regional temperature anomalies. For this variable, no significant relationship is found with global temperatures, and sensitivity varies spatially across the domain. These findings demonstrate a potential use of re-forecast data in assessing the impact of large-scale climate trends on changing extreme event probabilities, overcoming limitations of short observational records and biases in conventional models. We also note the considerable additional value that would be provided by extending re-forecast ensembles to the past 40 years rather than the current practice of limiting re-forecast ensembles to 20 years.