Our algorithm is relatively simple and straight forward. At first, we attempt to fit several wind field models, including modified Rankine combined vortex (MRCV), locally around each point. Because each wind field model has different number of parameters, we use AIC to choose a best suitable wind field. Calculated parameters are outputted if parameters satisfy criteria for mesoscale vortex, divergence, or convergence. We perform this process for PPI observations in a duration. Here, we do not have complicated shear-segment processing as found in most of the past method. Secondly, we compare the calculated parameters with specific thresholds to categorize each target signature. We conduct spatial and temporal continuity checks to remove false detection in this stage. Finally, we make quality check and vortex/divergence identification with reflectivity data and with motion of detected signatures. Motion of each signature can be calculated from the outputted data.
We examined the algorithm performance with simulated MRCV data with random noises and voids. The result shows that errors of signature's location, core radius and strength are sufficiently small for MRCV as expected. Random noise and randomly distributed void data (up to 20 %) affect little. It is noted that range dependency of the detection performance, which is often problematic on a detection and evaluation of distant mesocyclone, seems small. We also examined it with real data in the cases of supercell storms and gust fronts. For the supercell cases, detected mesocyclones are in accordance with objectively analyzed ones. However, for the strong gust front cases, false mesocyclones are detected near the gust front because of significant azimuthal shear.
We plan to add some wind field models, including gust front and non-axisymmetric vortices/divergence, to improve detection performance.