Using observational retrievals to expose precipitation and convective dynamics biases in simulations of a tropical MCS

Realistically simulating convective and stratiform structures of mesoscale convective systems (MCSs) requires proper representation of the large-scale environment and adequate representation of convective dynamics and parameterized microphysics in the simulation. The partitioning and areal coverage of such structures determine the heating distribution of any given system, thus affecting the large-scale dynamic response to the system. This issue is becoming evermore relevant with increasing computing power that will allow numerical weather prediction and global climate models to approach 1-10 km horizontal grid spacing in the coming decades. To test whether mesoscale simulations using 1-km grid spacing properly simulate a large tropical monsoonal MCS, several cloud-resolving model (CRM) and limited area model (LAM) simulations of a Tropical Warm Pool-International Cloud Experiment (TWP-ICE) MCS are compared with several observational retrievals. These comparisons expose a high bias in convective radar reflectivity aloft and a low bias in stratiform rainfall. Combined with errors due to large-scale environmental biases, model setup, and bulk microphysics parameterization assumptions, these biases appear related to overly intense deep convection in the simulations based on comparisons with dual-Doppler retrieved vertical velocity. The difference between simulated and dual-Doppler retrieved vertical velocity is especially large in the upper troposphere. This upper tropospheric difference is partially due to lofting and freezing of large rain water contents in simulations, which leads to large increases in buoyancy through latent heating. Possible reasons for overly intense simulated convective updrafts with large condensate loadings at mid and upper levels are explored.