Modelling strategies for performing convection-permitting climate simulations

The computational cost still remains a limiting factor for performing convection-permitting climate simulations. Choosing a model set-up with the lowest computational cost without deteriorating the model performances is, therefore, of relevance before starting any decadal simulations at convection-permitting scale (CPS). In this study three different strategies that aim at reducing this computational cost are evaluated. These strategies are (1) excluding graupel in the microphysical scheme, (2) reducing the nesting steps to downscale from ERA-Interim scale to CPS and (3) reducing the domain size. To test these strategies, the COSMO-CLM regional model was integrated over a four-month summer period for Belgium and evaluated using both radar and rain-gauges precipitation data. It was found that excluding the graupel parametrization at CPS induces a dry bias, but that excluding the graupel parametrization in the parent nest of the CPS simulation does not impact daily accumulated precipitation. In addition, it was also found that the best downscaling strategy is to use two nesting steps, in our case 25 km and 2.8 km. The 7 km nest was found to be redundant. Finally, it was found that a minimum distance of ~150$\sim 150$ km between the evaluation domain and the lateral boundary is needed for daily precipitation to converge towards observed values. This indicates that the domain size must be large enough for the model to spin-up convective precipitation and in our case a domain size of 180×180$180 \times 180$ grid-points was found to be necessary. Our recommendations for CPS simulations at lowest computational cost are therefore (1) to include graupel parametrization at CPS but not in the parent nest, (2) to use two nesting steps to downscale from ERA-Interim to CPS and (3) to use a domain size large enough to allow for 150 km spatial spin-up.