Composition du jury

– Laurence Picon
– Jean-Philippe Lafore
– Robert Plant
– Kerry Emanuel
– Jason Evans UNSW co-Directeur de thèse
– Jean-Yves Grandpeix
– Sandrine Bony co-Directeur de thèse
– Steven Sherwood UNSW Directeur de thèse

Résumé

Convective parameterizations struggle to represent the spatial and temporal variability of convection. This may be because they wrongly assume that convection can be diagnosed from the large-scale state, without knowing the convective history. The concept of convective memory, which states that convection depends on its own history, could be used to overcome this issue. This thesis proposes a new framework for convective memory, with a distinction between microstate (unresolved) memory and macrostate (large-scale) memory.
The thesis uses a hierarchy of models either in Radiative-Convective Equilibrium or under fixed-macrostate conditions, and analyses the recovery to homogenisation perturbations. It exploits three types of models: a Cloud-Resolving Model, a General Circulation Model (GCM) in 1D and in 3D, and a simple predator-prey model. The results show that convective memory plays a role on time scales between an hour and a day. Convective memory in time is dramatically enhanced by convective organisation in space. Microstate memory is found to be mostly stored in boundary layer microstate structures of water vapour and temperature, with a dominant water vapour memory. Furthermore, the convective microstate is shown to be inherently unstable, which confirms that knowledge of the macrostate conditions is not sufficient to predict convection. The standard version of the GCM already shows a reasonable level of convective persistence. A simple modification of the GCM convection scheme, meant to improve cold pools over oceans, improves some aspects of memory but deteriorates others. With this modification, the GCM cold pools become less cold and thus weaker to trigger convection. This leads to more intermittent precipitation, partly correcting a typical GCM bias.
Overall the thesis fosters the idea of introducing prognostic variables into GCMs and suggests ways to do it. It reveals the potential of microstate memory to improve GCM large-scale properties.