In recent years, thanks to developments in statistical learning methods, particularly neural networks, regional climate model (RCM) emulators have emerged as a promising and maybe indispensable solution for accurately studying the impacts of climate change at local scales. The main objective is to learn a downscaling relationship between a “large-scale” description of the meteorological situation (low resolution, middle/upper atmosphere) and one or more high-resolution surface variables (the one of the RCM) at a given time step. Once estimated, this downscaling function can be applied to any simulation from a low-resolution global model to produce high-resolution information.

Thanks to the very low computational cost of statistical methods, large ensembles can be created to properly study local climate change, i.e., by integrating the differences that are sources of uncertainty. In this presentation, I will introduce and validate the emulator concept developed at CNRM, then illustrate an example of its application.

Lieu
École normale supérieure – PSL
24 rue Lhomond – aile Erasme
salle Claude Froidevaux – E314