Unseen or unrealistic? Using ensemble simulations to explore unseen weather extremes

Weather extremes cause high socio-economic impacts globally and are projected to become more frequent in the future due to climate change. Quantifying and explaining the effect climate change has already had on climatic extremes is of high importance but is restricted by the brevity and sparsity of observed meteorological records. Furthermore, some policy makers are interested in the worst plausible events for decision making, and even the longest observed records (~100 years) might not capture such “unseen” events. In this work, large ensemble simulations are employed as numerous alternative realizations of the real world to quantify the likelihood of climate extremes and explain their nonstationary behaviour beyond what is possible from observed records. This research follows the UNprecedented Simulated Extreme ENsemble (UNSEEN) approach, an emerging asset that has yet to be fully exploited. The applicability of the method and its potential are explored. Statistical tests are developed to evaluate UNSEEN. Furthermore, a novel UNSEEN-trends approach is developed, facilitating detection of changes in 100-year precipitation values over short multi-decadal periods. A case study for Svalbard reveals a rise in 3-day precipitation extremes, such that the 100-year event estimated in 1981 occurs with a return period of around 40 years in 2015. The method is furthermore tested on floods in the Amazon using a hydro-climatological modelling framework. Flood magnitudes far beyond observed values are detected, but a rare bias-correction phenomenon unrealistically altering flood simulations is found. This result indicates that, besides statistical tests, performing physical credibility checks might uncover otherwise 'hidden' modelling errors that may lead to unrealistic extreme events. These findings are incorporated into an UNSEEN protocol, including an open and transferable workflow to enhance the uptake of UNSEEN. This new workflow for example demonstrates that the 2020 March-May Siberian heat wave, which led to infrastructure failure and permafrost thawing, was captured by one of the UNSEEN members. Through the various case studies presented in this thesis, it is discussed that UNSEEN can be unrealistic, because the approach hinges on the realism of the models underlying the ensemble simulations and can be sensitive to user decisions. Yet, with the right protocols and evaluation metrics, UNSEEN can provide insights into low-likelihood high-impact weather events and helps to 1) detect non-stationarity of extreme events; 2) review design values; and 3) put memorable events in context. Future model advancements and research on multimodel and multi-method combinations may further increase the spectrum of events for which plausible extremes can realistically be anticipated. The demonstrated strength of UNSEEN warrants more research and implementation of the approach across different types of extremes, regions, seasons, and spatial-temporal scales, and calls for further use-cases and implementation in risk analyses across sectors.