Explaining discrepancies in passive microwave cloud-radiation databases in microphysical context from two different cloud-resolving models.

Mesoscale Cloud Resolving Models (CRMs) are often used to generate descriptions of the microphysical properties of precipitating clouds for the purpose of guiding precipitation retrieval algorithms designed for satellite-borne passive microwave radiometers. However, CRMs were not originally designed for that purpose. Notably, individual CRMs have adopted different bulk microphysical schemes to optimize the dynamical evolution of storms and accumulated rainfall, rather than optimizing for simulations of radiative properties -- which are greatly affected by the microphysical details and vertical distributions of liquid and frozen hydrometeors. Thus, in principle, the simulated upwelling passive microwave (PMW) brightness temperatures (TBs) and associated precipitation retrievals generated by means of different CRMs with different microphysical parameterizations may be significantly different -- even when the different CRMs prognostically adhere to the main dynamical and precipitation characteristics of a given storm. We investigate this issue for two different mesoscale models run at CRM scales, each using different parameterizations for the ongoing microphysics. These are the University of Wisconsin Nonhydrostatic Modeling System (NMS) and the 5th generation version of the Pennsylvania State University / National Center for Atmospheric Research Mesoscale Model (MM5). These two models are used to simulate the same flood-producing storm that occurred over northern Italy during 24-26 November 2002. Model outputs that best reproduce the structure of the storm, as observed by the Advanced Microwave Scanning Radiometer (AMSR-E) onboard the NASA AQUA satellite, are used to calculate upwelling PMW TBs. The simulated TBs are then used for retrieving the precipitation fields in conjunction with the AMSR-E observations. Finally, the two sets of results are intercompared in order to provide an indication of the expected uncertainties in CRM-based precipitation retrievals due to differing microphysical parameterizations. Results show that the two models are in close agreement insofar as simulating the organizational characteristics of the storm, and that the bulk statistical properties of the two sets of retrieved precipitation rates are in close correspondence. By the same token, although the overall conditional bias is only 0.09, close examination of the two sets of retrievals indicates that rain rates begin to show their most significant differences above ~4.5 mm h-1, with differences larger than 10 mm h-1 occurring just under 2% of the time.