An algorithm based on Steiner et al. (1995) was used to separate areas of convection from stratiform regions for each volume of reflectivity. This algorithm was developed by Tom Rickenbach at JCET/NASA GCFC. This technique examines the spatial uniformity and intensity of a low level map of radar reflectivity to delineate areas of horizontally variable precipitation (convective rainfall) from weaker, horizontally uniform precipitation (stratiform rainfall). A constant-altitude map of reflectivity (2 km AGL) is examined for local maxima in the reflectivity field. Any local maximum greater than 40 dBZ was identified as a convective cell core. If a local maximum did not exceed this threshold, it was considered to be a convective cell core if the reflectivity value exceeded the mean linear reflectivity in an 11 km radius circle surrounding the point by a constant threshold value corresponding to roughly a factor of two rainfall difference (4.5 dB, Churchill and Houze 1984). Next, a circular "cell" surrounding the convective point is placed in the convective category to simulate a convective cell. The radius of this cell was directly proportional to the reflectivity value of the core, and varies between 1 km and 5 km. Once a reflectivity map at the 2 km height level is partitioned, all points in the column above each convective grid point is considered to be convective. All doppler velocities collocated with convective reflectivity points are incorporated into the convective air motion vertical profiles. All velocities not considered convective were placed into the stratiform category.
Subjective impressions from using the algorithm on a variety of precipitating systems suggests that it does a good job at isolating the intense convective cells; however, the stratiform category may contain a mixture of decaying convective elements as well as stratiform precipitation. For this reason, it may be more appropriate to view the partitioning categories as convective and non-convective. Also, because of the fact that the thresholding is performed at low levels and extrapolated upwards, the algorithm may produce spurious results in regions of large vertical shear.