Quantile-based downscaling of rainfall extremes: Notes on methodological functionality, associated uncertainty and application in practice

Local characteristics of extreme rainfall quantiles, manifested through Intensity-Duration-Frequency (IDF) curves, are key to infrastructure design. Due to climate change, rainfall extremes are subject to changes, it is, therefore, crucial to explore the potential impacts these changes will have on design storms. A new strain of methodologies, quantile-based downscaling approaches, have recently been proposed to exclusively downscale extreme rainfall quantiles obtained from Global Climate Models (GCMs). These approaches, however, have not been systematically intercompared and the uncertainties related to assigning future design storms are poorly understood. This study evaluates the functionality of three quantile-based downscaling methods during the historical and future periods in Montreal, Canada. Results show that the performance of quantile-based downscaling approaches in reproducing observed extreme quantiles can be divergent. At lower return periods, however, differences between the three schemes are not significant. Similar performances for reproducing historical rainfall extremes, however, does not necessarily imply similar future projections due to the different functionalities of the three approaches in mapping GCM projections into finer scales. Despite these uncertainties, the total projection range of future rainfall extremes are, in many cases, comparable to the confidence interval of the parametric probability distribution when fitted to the observed annual maximum rainfall series. A risk-based approach to accommodate this uncertainty in vulnerability assessments through evaluating potential alterations in historical rainfall extremes using an ensemble projection coming from multiple downscaling approaches is suggested. This allows for the selection of design storms based on the acceptable level of risk and given budgetary and operational restrictions.