Probabilistic hybrid modular structure for multisite and multivariable statistical downscaling

Probabilistic regression approaches for downscaling daily temperature and precipitation are very useful. They provide the whole conditional distribution at each forecast step which leads to a better representation of the temporal variability. The question addressed in this study is: how to extend probabilistic regression approaches in multisite and multivariable downscaling tasks. To this end, this study describes a probabilistic hybrid modular structure which merges the probabilistic regression component with a multivariate randomisation component. Thereby, the aim is to develop new statistical models for multisite precipitations and temperatures downscaling following the described modular structure. These new models are based on statistical tools of increasing interest in the hydro-meteorological literature during the last years, including multivariate tools such as copulas, and probabilistic regression approaches such as quantile regression and vector generalized linear models.

About the speaker:
Mohamed Ali Ben Alaya is a post-doctoral research fellow at PCIC working on the Canadian Network for Regional Climate and Weather Processes (CNRCWP) project. He will be presenting work from his Ph.D. in Water Sciences at the INRS-ETE in Quebec. His research interests include statistical climate downscaling and flood risk assessment in a changing climate.