Abstract. Optimising the model performance to reduce model biases is a challenging task in global and regional climate modelling, especially relevant for free-running climate change simulations. This challenge is addressed in the present study through a systematic regional climate model tuning strategy using a novel methodology, which includes an iterative update of the reference configuration and combines expert judgement with objective tuning using a Linear Meta-Model optimisation (LiMMo) to derive an optimised model configuration. We applied this methodology to the regional climate model ICON-CLM setup over Europe at 12 km grid size (EURO-CORDEX domain) in order to reduce, e.g., the overestimation of incoming solar radiation and too low 2 m temperature. During this process, the sensitivity of the model to changes of 29 model parameters and their physical consistency was tested and investigated. Comparing the results of optimisation by expert judgement with those of LiMMo showed that the latter not only confirmed the expert judgement by focusing on a priori known highly sensitive parameters, but also allowed for fine-tuning of the model configuration with explicit control over the tuning process, making parameter combinations more efficient. With reference to the default ICON numerical weather prediction configuration, the model optimisation yielded significant improvements for a real climate mode simulations use case. For example, biases in incoming short wave radiation could be reduced by 30 %, latent heat flux biases by 15 %, by tuning cloud parameters in combination with surface flux parameters. Furthermore, the new optimised configuration could only be reached by using updated, higher-quality external datasets, including transient aerosols. Based on the community-based coordinated parameter tuning, we recommend an ICON-CLM model configuration for the EURO-CORDEX domain that is already being used for the downscaling of global CMIP6 simulations.
Geyer et al. (Wed,) studied this question.