92nd American Meteorological Society Annual Meeting (January 22-26, 2012)

Thursday, 26 January 2012: 11:30 AM
Automated Identification of Enhanced Rainfall Rates for Radar-Based QPE Using the near Storm Environment and Vertical Reflectivity Profiles
Room 352 (New Orleans Convention Center )
Heather Moser, CIMMS/Univ. of Oklahoma, Norman, OK; and J. Zhang and K. Howard

The National Mosaic and Multi-Sensor QPE rainfall product, or Q2, uses a tropical R(Z) relationship for estimation of rainfall where warm rain processes predominantly determine the DSD structure. The tropical R(Z) generally provides more accurate rainfall estimates than the National Weather Service convective R(Z) in tropical cyclones and convection forming in maritime airmasses. It is based on the mean VPR of each radar and the vertical gradient of reflectivity below the freezing level (i.e., increasing reflectivity with decreasing height implies that condensation and warm rain microphysical growth processes are occurring). On the individual storm scale this approach is appropriate, but ambiguities can arise when the area over which the VPR is computed contains multiple rainfall regimes or different airmasses. Furthermore, extreme rainfall rates have been observed that are underestimated by both the tropical and convective R(Z)s.

Thermodynamics and moisture content of the ambient environment have often been studied as important factors in the development of very heavy rainfall rates and are key discriminators between maritime and continental airmasses. Thus, they may be useful predictors of where tropical R(Z)s are most appropriate vs. where even higher rainfall rates are needed. In order to delineate these two areas, a supervised classification model using support vector machines has been developed that examines both the near-storm environment from the 20-km Rapid Update Cycle hourly analysis and vertical structure of radar reflectivity. Classification and verification was based on comparisons of the tropical and convective/stratiform R(Z)s to collocated hourly rain gauges. The classification model was trained through multiple iterations using different predictor combinations and random sampling to determine both robustness of the results and relative skill of the predictors.

Supplementary URL: