Evaluation of aerosol and cloud properties in three climate models using MODIS observations and its corresponding COSP simulator, as well as their application in aerosol–cloud interactions

The evaluation of modelling diagnostics with appropriate observations is an important task that establishes the capabilities and reliability of models. In this study we compare aerosol and cloud properties obtained from three different climate models (ECHAM-HAM, ECHAM-HAM-SALSA, and NorESM) with satellite observations using Moderate Resolution Imaging Spectroradiometer (MODIS) data. The simulator MODIS-COSP version 1.4 was implemented into the climate models to obtain MODIS-like cloud diagnostics, thus enabling model-to-model and model-to-satellite comparisons. Cloud droplet number concentrations (CDNCs) are derived identically from MODIS-COSP-simulated and MODIS-retrieved values of cloud optical depth and effective radius. For CDNC, the models capture the observed spatial distribution of higher values typically found near the coasts, downwind of the major continents, and lower values over the remote ocean and land areas. However, the COSP-simulated CDNC values are higher than those observed, whilst the direct model CDNC output is significantly lower than the MODIS-COSP diagnostics. NorESM produces large spatial biases for ice cloud properties and thick clouds over land. Despite having identical cloud modules, ECHAM-HAM and ECHAM-HAM-SALSA diverge in their representation of spatial and vertical distributions of clouds. From the spatial distributions of aerosol optical depth (AOD) and aerosol index (AI), we find that NorESM shows large biases for AOD over bright land surfaces, while discrepancies between ECHAM-HAM and ECHAM-HAM-SALSA can be observed mainly over oceans. Overall, the AIs from the different models are in good agreement globally, with higher negative biases in the Northern Hemisphere. We evaluate the aerosol–cloud interactions by computing the sensitivity parameter ACICDNC=dln⁡(CDNC)/dln⁡(AI) on a global scale. However, 1 year of data may be considered not enough to assess the similarity or dissimilarities of the models due to large temporal variability in cloud properties. This study shows how simulators facilitate the evaluation of cloud properties and expose model deficiencies, which are necessary steps to further improve the parameterisation in climate models.