Targeted bulk microphysics scheme improvements through cloud-resolving and limited-area model intercomparison with observations
Authors
Adam Varble — Pacific Northwest National Laboratory
Edward Zipser — University of Utah
Ann M. Fridlind — NASA - Goddard Institute for Space Studies
Ping Zhu — Florida International University
Andrew Ackerman — NASA - Goddard Institute for Space Studies
Jean-Pierre Chaboureau — University of Toulouse, France/CNRS
Jiwen Fan — Pacific Northwest National Laboratory
Adrian Hill — UK Meteorological Office
Ben Shipway — UK Meteorological Office
Christopher R Williams — University of Colorado, Boulder
Category
Precipitation
Description
Errors in cloud-resolving model (CRM) and limited-area model (LAM) simulations that result from various bulk microphysics scheme assumptions can be reduced by making scheme alterations guided by comparisons with each other and comparisons with available observations. This is the approach being used for a suite of CRM and LAM simulations of the 23–24 January 2006 mesoscale convective system (MCS) event during the Tropical Warm Pool-International Cloud Experiment (TWP-ICE). There are many differences between CRM and LAM simulations due to very different boundary conditions and forcing methodologies. For example, stratiform area is much lower in the LAM simulations than the CRM simulations. In terms of radar reflectivity and rainfall statistics, the CRM simulations perform better than the LAM simulations, which is not surprising because the CRM forcing is partly derived through radar-derived rain-rate observations. With the inclusion of analysis nudging in coarser domains, the LAM simulations are able, however, to generate cyclonic flow associated with the MCS whereas the CRMs inherently cannot.
Despite these differences, there are many important similarities between CRM biases and LAM biases related to bulk microphysics assumptions. Across all simulations, convective area is too high, simulated convective radar reflectivity aloft is too high, and stratiform rain rates are too low. Despite this general agreement in bias, there still exists substantial spread in model output. Differences between simulations are highly correlated with differences in assumed hydrometeor size distribution properties and the number of prognostic moments of the hydrometeor size distributions. Specific scheme components that appear related to the biases and have possibilities for future improvement are the rain droplet breakup parameterization, the density (and hence fall speed) of rimed precipitating ice, the assumed mass-dimension relationship for snow, and the assumed gamma size distribution shape parameter for rain. These will be explored in upcoming sensitivity tests.