Authors:
Danny Williamson | University of Exeter | United Kingdom
Fleur Couvreux | CNRM, Météo-France | France
Frederic Hourdin | laboratoire de meterologie dynamique | France
Romain Roehrig | CNRM, Météo-France | France
Nadja Villefranque | Université de Toulouse | France
Catherine Rio | CNRM, Météo-France | France
James Salter | University of Exeter | United Kingdom
Victoria Volodina | Alan Turing Institute | United Kingdom
Wenzhe Xu | University of Exeter | United Kingdom
Climate models solve a discretized version of the Navier-Stokes equations on the rotating sphere whilst respecting the laws of thermodynamics and energy conservation. However, the scale of the feasible discretizations means that processes essential to plausible evolution of the climate, such as convection and the formation of clouds, are not resolved and so must be modeled using `parameterizations’ that contain a number of free parameters. Traditional climate model development involves encoding our best physical understanding into separate parameterizations and ensuring the processes are well represented, then coupling the parameterizations and finally `tuning’ the free parameters in the coupled model so that the simulations satisfy certain targets (Hourdin et al. 2017). UQ has been proposed for aiding the tuning phase (Williamson et al. 2017, Salter et al. 2019), yet the task is difficult in isolation due to the number of parameters and demanding computation of the coupled model. In this paper we re-imagine climate model development and embed UQ (specifically emulation, history matching and sensitivity analysis) at the process modeling stage. This philosophy is adopted for the development of the 2 French climate models (CNRM-CMand IPSL-CM) specifically in the development and testing of convection schemes to improve model cloud physics. In this talk I will review the methods and some of the key benefits to climate model development through examples with the French models.