An open workflow to gain insights about low‐likelihood high‐impact weather events from initialized predictions

Low-likelihood weather events can cause dramatic impacts, especially when they are unprecedented. In 2020, amongst other high-impact weather events, UK floods caused more than £300 million damage, prolonged heat over Siberia led to infrastructure failure and permafrost thawing, while wildfires ravaged California. Such rare phenomena cannot be studied well from historical records or reanalysis data. One way to improve our awareness is to exploit ensemble prediction systems, which represent large samples of simulated weather events. This ‘UNSEEN’ method has been successfully applied in several scientific studies, but uptake is hindered by large data and processing requirements, and by uncertainty regarding the credibility of the simulations. Here, we provide a protocol to apply and ensure credibility of UNSEEN for studying low-likelihood high-impact weather events globally, including an open workflow based on Copernicus Climate Change Services (C3S) seasonal predictions. Demonstrating the workflow using European Centre for Medium-Range Weather Forecasts (ECMWF) SEAS5, we find that the 2020 March–May Siberian heatwave was predicted by one of the ensemble members; and that the record-shattering August 2020 California-Mexico temperatures were part of a strong increasing trend. However, each of the case studies exposes challenges with respect to the credibility of UNSEEN and the sensitivity of the outcomes to user decisions. We conclude that UNSEEN can provide new insights about low-likelihood weather events when the decisions are transparent, and the challenges and sensitivities are acknowledged. Anticipating plausible low-likelihood extreme events and uncovering unforeseen hazards under a changing climate warrants further research at the science-policy interface to manage high impacts.