Kilometer-resolution climate models: prospects and challenges

Currently major efforts are underway toward refining the horizontal grid spacing of climate models to about 1 km, using both global and regional models. Such resolutions have been used for about a decade in limited-area numerical weather prediction applications and have demonstrated significant improvements in the representation of convective precipitation events (thunderstorms and rain showers). There is the well-founded hope that these benefits carry over to climate models, as the approach enables replacing the parameterizations of moist convection and gravity-wave drag by explicit treatments.

In this presentation, we will review three areas of km-resolution climate modeling. First, consideration will be given to an ensemble of km-resolution simulations from the CORDEX-FPS program on convection-permitting climate modelling, with a computational domain covering the greater Alpine region. This addresses the occurrence of short-term heavy precipitation events including their impacts such as flash floods, hail, and lightning. Results demonstrate the benefits of high computational resolution, in particular for the representation of short-term heavy events of severe weather. Second, we will present recent results from the projects trCLIM / CONSTRAIN carried out over the tropical and subtropical Atlantic, with the goal to assess the potential of the methodology to constrain estimates of the equilibrium climate sensitivity. It will be argued that km-resolution is a highly promising approach for constraining uncertainties in global climate change projections, due to improvements in the representation of tropical and subtropical clouds that goes along with an improved representation of the intertropical convergence zone (ITCZ).

Third, technical aspects of developing km-resolution global models will be addressed. Developing this approach requires a concerted effort between climate and computer sciences. Key challenges are the exploitation of the next generation hardware architectures using accelerators (e.g. graphics processing units, GPUs), the development of suitable approaches to overcome the output avalanche, and the consistent maintenance of the rapidly-developing model source codes on a number of different compute architectures. Despite these challenges, it will be argued that km-resolution GCMs with a capacity to run at 1 SYPD (simulated year per day), might be much closer than commonly believed. However, as the computational load of CMIP-style simulations is tremendous, alternative ways to exploit these models will be needed.