Observational constraint of drizzle properties and processes in large-eddy simulations from two models with size-resolved microphysics

Abstract:

From the CAP-MBL long-term measurement campaign over the Azores, two low-cloud periods are selected via analysis of ISCCP cloud property matrices. Each is a persistent instance of a low-cloud weather state that the GISS climate model severely underpredicts both over the Azores and globally. Meteorologically, one period is characterized by shallow cumulus clouds in a cold-air outbreak behind a cold front, and the other by overcast stratocumulus clouds in a region dominated by a high-pressure system. Observations and MERRA reanalysis are used to derive specifications used for large-eddy simulations (LES) of each period. For both periods, 6-hour and longer simulations with periodic boundary conditions and uniform large-scale forcings over 10 to 100-km domains capture the structural differences between the two low-cloud periods, only modestly influenced by domain size. Simulations of both periods underestimate sub-mesoscale variability, but represent distributions of cloud top, cloud base, and drizzle extents reasonably well in comparison with observations. Simulated mean Doppler velocities as a function of reflectivity also reproduce observed relationships reasonably well; the most notable deviation is lesser-than-observed Doppler velocity where reflectivity exceeds about –10 dBZ, which occurs only in the cumulus case. However, the range of simulated spectral skewness is substantially underestimated in both cases. To help identify likely causes, the overcast stratocumulus case is also simulated with an independent LES code with size-resolved microphysics, which exhibits some similar biases. Although unavailable from CAP-MBL, in situ measurements of drizzle size distribution from two other campaigns are used to gain insight into the biases. Finally, the significance of Doppler moment biases is evaluated in part by deriving autoconversion and accretion parameterization expressions from simulations with varying collision-coalescence kernel assumptions.