Statistical Modelling of Complex Spatial Extreme Phenomena

点击翻译按钮获取中文摘要

Canada has been hit by many large scale natural catastrophes in recent years, such as the 2017 floods in Quebec, the drought that hit the Prairies in 2018--19 or the unprecedented wildfires in British Columbia in the summer 2021. Reports on global warming often focus on the expected increase in mean temperatures, but few highlight the potential increased frequency and intensity of climate extreme episodes. These changes, together with the scarcity of extreme events in historical records, limit our ability to estimate the probability of future disasters. Extreme value theory provides a well- founded framework for studying rare events and its univariate setting is well understood. However, it is often the co--occurrence of the extreme episodes and their aggregation that exacerbates the impact on communities at risk. Many research projects in recent years have therefore focused on multivariate and spatio--temporal extreme modelling. Despite these advances, current spatial models are incapable of handling hundreds of sites simultaneously and result in risk maps with poor spatial resolution. Estimates are also highly uncertain as a result of small sample sizes. This directly affects our ability to mitigate the impact of complex extreme natural hazards. This research agenda proposes strategies to improve risk assessment for large- scale environmental spatial extreme events using the conditional spatial extreme model and focusing on increasing both its scalability and flexibility. Three projects are described to this effect that would permit usage of more sites simultaneously, create more realistic simulations and finally allow us to combine reanalysis model output with historical observations. The projects will use climate reanalysis data from the Pacific Climate Impacts Consortium (PCIC) and historical data from NOAA and Environment Canada. The long- term objective of the research is to be able to tackle applications with thousands of weather stations simultaneously and to combine information from climate models with historical records to improve risk management through a catalogue of potential extreme events simulated from the developed models.