Abstract Using successive versions of a global climate model, we show how convective transport to the free troposphere of the humidity evaporated at the surface or, reciprocally, entrainment of dry air from the free troposphere into the mixed layer, controls surface evaporative cooling and then sea surface temperature. This control is as important as the radiative effect of boundary layer clouds on radiation. Those aspects are shown to be improved when activating a mass flux representation of the organized structures of the convective boundary layer coupled to eddy diffusion, the so‐called “thermal plume model,” leading to an increased near‐surface drying compared to the use of turbulent diffusion alone. Controlling detrainment by air properties from just above the boundary layer allows the thermal plume model to be valid for both cumulus and stratocumulus regimes, improving the contrast in near‐surface humidity between the trade winds region and East Tropical oceans. Using pairs of stand‐alone atmospheric simulations forced by sea surface temperature and of coupled atmosphere‐ocean simulations, we show how the improvement of the surface fluxes that arise from this improved physics projects into an improvement of the representation of sea surface temperature patterns in the coupled model, and in particular into a reduction of the East Tropical Ocean warm bias. The work presented here led to the bias reduction in sea surface temperature in the Institute Pierre Simon Laplace coupled model, IPSL‐CM6A, developed recently for the 6th phase of the Coupled Model Intercomparison Project, CMIP6.

Abstract This study presents the version of the LMDZ global atmospheric model used as the atmospheric component of the Institut Pierre Simon Laplace coupled model (IPSL‐CM6A‐LR) to contribute to the 6th phase of the international Coupled Model Intercomparison Project (CMIP6). This LMDZ6A version includes original convective parameterizations that define the LMDZ “New Physics”: a mass flux parameterization of the organized structures of the convective boundary layer, the “thermal plume model,” and a parameterization of the cold pools created by reevaporation of convective rainfall. The vertical velocity associated with thermal plumes and gust fronts of cold pools are used to control the triggering and intensity of deep convection. Because of several shortcomings, the early version 5B of this New Physics was worse than the previous “Standard Physics” version 5A regarding several classical climate metrics. To overcome these deficiencies, version 6A includes new developments: a stochastic triggering of deep convection, a modification of the thermal plume model that allows the representation of stratocumulus and cumulus clouds in a unified framework, an improved parameterization of very stable boundary layers, and the modification of the gravity waves scheme targeting the quasi‐biennal oscillation in the stratosphere. These improvements to the physical content and a more well‐defined tuning strategy led to major improvements in the LMDZ6A version model climatology. Beyond the presentation of this particular model version and documentation of its climatology, the present paper underlines possible methodological pathways toward model improvement that can be shared across modeling groups.

Abstract The representation of stratocumulus clouds, and of the stratocumulus to cumulus transitions which are ubiquitous features of marine boundary layer clouds, remains a challenge for climate models. We show how a mass flux representation of boundary layer convective structures combined with an eddy diffusivity scheme, the “thermal plume model,” first developed to represent cumulus clouds, can also adequately simulate stratocumulus and the stratocumulus to cumulus transition in a climate model. To achieve this, the detrainment formulation, in which detrainment increases for increasing negative buoyancy, has to be slightly modified: the buoyancy of a thermal plume parcel of air is computed by comparing the virtual potential temperature θv,th of the parcel with that of the surrounding environment θv,env at a given distance above instead of at the same level. This is consistent with the picture of detrained air parcels that experience some overshoot and reach a final destination at a level lower than the one at which they effectively leave the cloud or organized convective plume. The impacts of this modification are documented both for selected cases of stratocumulus, in comparison with large‐eddy simulations, and in full 3‐D climate simulations, in comparison with satellite observations of cloud cover. The modified scheme provides a uniform treatment of the dry convective boundary layer, of cumulus clouds, of stratocumulus, and of the transition from stratocumulus to cumulus. It is included in the most recent version of the LMDZ atmospheric general circulation model.

Based on a decade of research on cloud processes, a new version of the LMDZ atmospheric general circulation model has been developed that corresponds to a complete recasting of the parameterization of turbulence, convection and clouds. This LMDZ5B version includes a mass-flux representation of the thermal plumes or rolls of the convective boundary layer, coupled to a bi-Gaussian statistical cloud scheme, as well as a parameterization of the cold pools generated below cumulonimbus by re-evaporation of convective precipitation. The triggering and closure of deep convection are now controlled by lifting processes in the sub-cloud layer. An available lifting energy and lifting power are provided both by the thermal plumes and by the spread of cold pools. The individual parameterizations were carefully validated against the results of explicit high resolution simulations. Here we present the work done to go from those new concepts and developments to a full 3D atmospheric model, used in particular for climate change projections with the IPSL-CM5B coupled model. Based on a series of sensitivity experiments, we document the differences with the previous LMDZ5A version distinguishing the role of parameterization changes from that of model tuning. Improvements found previously in single-column simulations of case studies are confirmed in the 3D model: (1) the convective boundary layer and cumulus clouds are better represented and (2) the diurnal cycle of convective rainfall over continents is delayed by several hours, solving a longstanding problem in climate modeling. The variability of tropical rainfall is also larger in LMDZ5B at intraseasonal time-scales. Significant biases of the LMDZ5A model however remain, or are even sometimes amplified. The paper emphasizes the importance of parameterization improvements and model tuning in the frame of climate change studies as well as the new paradigm that represents the improvement of 3D climate models under the control of single-column case studies simulations.

L'objectif principal de cette these est d'ameliorer la representation des couches limites convectives nuageuses dans les modeles de climat et en particulier, de developper un modele statistique de nuage. Pour cela, nous sommes partis du « modele du thermique »: dans ce modele, les panaches ascendants sont representes par un unique panache d'air ascendant alimentes lateralement par de l'air entraine depuis l'environnement et detrainent une partie de leur air dans l'environnement. Le developpement d'une nouvelle parametrisation des taux de melange a pu etre fait grâce a l'utilisation de traceurs passifs emis en surface dans les simulations explicites 3D a haute resolution. Cette parametrisation est ensuite testee dans une version unicolonne du modele de climat LMDZ sur des cas de convection peu profonde et evaluee a partir de simulations haute resolution. Le modele du thermique est ensuite couple a un schema statistique de nuage base une distribution sous-maille bi-Gaussienne du deficit a saturation. L'echantillonnage conditionnel de simulations hautes resolutions permet d'attribuer un des modes de la bi-Gaussienne a l'air transporte par les thermiques et le second a l'air environnant et de valider ainsi la parametrisation des moments d'ordre 1 de la distribution en fonction des variables thermodynamiques calculees dans le modele des thermiques. Certaines des parametrisations evoquees ont ete integrees dans la nouvelle version 3D de LMDZ, utilisee depuis pour effectuer des simulations climatiques dont l'analyse des resultats servira de base au prochain rapport du Groupe d’experts Intergouvernemental sur l'Evolution du Climat (GIEC).