Intercomparison of simulations using 4 WRF microphysical schemes with dual-Polarization data for a German squall line

The ability of a 2.8 km version of the WRF model to reproduce accurately the microphysical structure of a squall line is evaluated by comparing WRF simulations using four different microphysical schemes to detailed observations gathered by the POLDIRAD dual-polarization radar located near Munich, Germany. A squall line event which occurred on August 12, 2004 is simulated with each version of the model. PPI and RHI scans showing reflectivity and hydrometeor mixing ratios from both the radar and derived from model output using a polarimetric radar forward operator are compared. The Ebert-McBride contiguous rain area method of verification is tested on the reflectivity output from the simulations. The dual-polarization data and the use of the Ebert-McBride technique permit verification of fields that have generally been ignored in the past.

The comparison of results focused on a 280 by 280 km domain having 2.8 km grid spacing, nested within a larger domain having 8.4 km grid spacing. No convective parameterization was used on either grid. The four microphysical schemes used included the Lin et al, the Thompson, the WSM6 and WSM 5 schemes. Only the WSM 6 version seemed to have the correct timing for when the squall line was best organized, intense, and oriented north-south, around 17 UTC. The other three versions were too fast by roughly 1 hour. The stratiform region was well-developed in the radar data by 19-20 UTC, but it evolved differently in all four simulations. Subjectively, the best agreement with observations in the general location of the stratiform rain took place at 20 UTC in the Thompson and WSM5 runs, but in the WSM 6 run, the best agreement was with the 20 UTC model output and 19 UTC observations, and in the Lin et al. run, best agreement was at 19 UTC. None of the models showed as much organization to the stratiform region as observations indicated.

Results indicate that over the full 2.8 km grid model domain, only the WSM6 did not have a problem with a high bias for reflectivities above the 25 dBZ threshold used to define the convective system in the Ebert-McBride scheme. Biases were largest for the WSM5 and Lin et al. schemes. Maximum reflectivities in all 4 simulations overestimated the observed values, but all were within about 5 dBZ of those observed. The worst overestimate was in the Thompson scheme.

The Ebert-McBride technique was applied both at 18 UTC, not taking into account temporal errors, and then again allowing for time errors. In the 18 UTC verification, the best domain-averaged correlation coefficient and equitable threat scores were in the WSM 6 simulation. Looking only at the squall line itself, as identified by the Ebert-Mcbride scheme, the Lin et al. microphysics scheme seemed to do best, with the lowest mean square error and highest correlation coefficient after displacement errors were taken into account at around 18 UTC, when the system was at its mature stage. A hydrometeor classification scheme applied to the model and radar data revealed that all of the parameterizations had difficulties in distinguishing between snow and graupel and the stratiform and convective parts of the storm.