Enhancing Cloud Droplet Number Concentration Representation in Global Climate Simulations

Aerosol-cloud interactions are currently the largest uncertainty in climate simulations of how Earth's radiation budget is changing. Consequently, significant effort has gone into improving the representation of these interactions in weather and climate models alike. One of the most critical controlling variables for aerosol-cloud interactions is cloud droplet number concentration (CDNC). Here we develop the representation of CDNC in the UK Met Office Unified Model (UM) global model and explore how simulated CDNC should be evaluated using satellite retrievals.

In unified weather and climate prediction systems, it is desirable to represent cloud microphysics using consistent code across scales. Consequently, we implement the UM’s regional double-moment microphysics scheme into the global model, which by default uses a single-moment scheme. Although this increases computational cost, the double-moment scheme can be evaluated more precisely against pixel-level satellite retrievals in high-resolution regional simulations, and it enables a more robust treatment of cloud droplets. Therefore, we describe how we develop and assess it in the global UM.

However, the evaluation of cloud droplets in a global model is difficult as comparisons must be made with satellite-derived, and typically masked, cloud top CDNC. We address this issue through the analysis of various masking strategies, along with evaluating the performance of the Cloud Feedback Model Intercomparison Project (CFMIP) Observation Simulator Package (COSP) satellite simulator as implemented in the UM. In doing so, we improve protocols for global model-satellite comparisons of CDNC.

Through the implementation of double-moment cloud microphysics into the UM’s global model, we find systematic improvements in simulated cloud droplet representation. The annual root mean squared error (RMSE) decreases by 4 cm^-3 globally, with a substantially larger reduction of 16 cm^-3 in the tropics, as the enhanced representation of microphysical processes (e.g., accretion and autoconversion) is particularly beneficial for convective systems. Outside of the tropics, a low droplet bias exists regardless of the microphysics scheme. We find this bias is partially explained by a ~ 50% underprediction of simulated aerosol concentration when compared with in situ measurements from the NASA Atmospheric Tomography (ATom) Mission. Applying an aerosol scaling factor reduces this droplet bias by half, showing that errors in aerosol and activation are comparable.

We further find that COSP improves model agreement with remotely sensed cloud optical depth (τ_c) and effective radius (r_e). However, COSP-derived CDNC has an RMSE 16 cm^-3 higher than that of CDNC calculated directly from the microphysics scheme, as biases in τ_c and r_e propagate through the simulator. Overall, we expect that our improvements to the representation of CDNC in the UM’s global model can meaningfully reduce the uncertainty in simulated aerosol-cloud interactions, and by extension, improve radiative forcing estimates.