Towards convection-resolving climate modeling

Abstract:

Moist convection is a fundamental process in our climate system, but is usually parameterized in climate models. The underlying approximations introduce significant uncertainties and biases, and there is thus a general thrust towards the explicit representation of convection. For climate applications, convection-resolving simulations are still very expensive, but are increasingly becoming feasible. Here we present recent results pertaining to the development and exploitation of convection-resolving regional climate models. We discuss the potential and challenges of the approach, highlight validation using decade-long simulations, explore convection-resolving climate change scenarios, and provide an outlook on the use of next-generation supercomputing architectures.

Detailed results will be presented using the COSMO model over two computational domains at a horizontal resolution of 2.2 km. The first covers an extended Alpine region from Northern Italy to Northern Germany. For this domain decade-long simulations have been conducted, driven by both reanalysis as well as CMIP5 model data. Results show that explicit convection leads to significant improvements in the representation of summer precipitation, and to substantial differences in climate projections in terms of precipitation statistics.

The second domain covers European (with 1536x1536x60 grid points) and the respective simulations exploit heterogeneous many-core hardware architectures. Results demonstrate realistic mesoscale processes embedded in synoptic-scale features, such as line convection along cold frontal systems, or the triggering of moist convection by propagating cold-air pools. Currently a 10-year simulation using this set up is near completion. To efficiently use GPU-based high-performance computers, the model code underwent significant development, including a rewrite of the dynamical core in C++. It is argued that today’s largest supercomputers would in principle be able to support – already now – global convection-resolving climate simulations, provided the respectively refactored codes would be available.