The Royal Institution for the Advancement of Learning/McGill University

Published purpose

There is mounting evidence that the magnitude and frequency of flash floods are expected to increase due to an increased likelihood of heavy precipitation events in a future warmer climate. To develop future projections of such events, transient climate change simulations from both Global and Regional Climate Models (i.e. GCMs and RCMs) are routinely used. While GCMs are constrained by their coarse spatial resolution, RCMs are computationally very expensive for generating super-resolution (~ 250 m and finer) simulations, which are required to understand flash flooding dynamics in urban areas. In this study, a physically consistent super-resolution emulator based on deep learning frameworks will be developed to generate super-resolution information of heavy precipitation events from coarser resolution simulations for future flood hazard assessment, for selected urban regions.