WARNING APPLICATIONS RESEARCH

image of a flooded house

Aggregate hailstone approximately six inches in diameter. The Warning Applications group works on systems that help forecasters detect, monitor and predict the occurrence of severe weather.

 

Related links:

home » about nssl » research divisions » warning r&d » warning applications research » mesocyclone detection algorithm (mda)

Mesocyclone Detection Algorithm (MDA)

NSSL's Radar Algorithm Display System (RADS) showing MDA output overlaid on top of a supercell storm in Southeast Oklahoma during the 11 May, 1992 outbreak.

This is an image of NSSL's Radar Algorithm Display System (RADS) showing MDA output overlaid on top of a supercell storm in Southeast Oklahoma during the 11 May, 1992 outbreak. Larger image

  • Thin yellow circles represent incipient or weak 3D vortex detections.
  • Thick yellow and red circle represents a 3D vortex which has been classified as a mesocyclone, with its base detected at the 0.5 degree elevation sweep of the volume scan.
  • Trend windows show trends of various attributes associated with vortex ID #22.
  • The Mesocyclone Table shows vortex ID #22 as being the second strongest within the radar domain (the other vortices are not shown at this magnification).

The NSSL Mesocyclone Detection Algorithm (MDA) was developed to address the shortcomings of the WSR-88D Mesocyclone algorithm. Initial development addressed the high false alarm rate of the WSR-88D algorithm, but in doing so, the probability of detection lowered.

Work was undertaken late in 1993 to redesign the MDA. The first step was to make the algorithm more robust in its various levels of analysis. In the 1D phase, or azimuthal shear pattern vector analysis, the strength thresholds were lowered by 1/3 their original value. This allowed for weaker pattern vectors to be analyzed later. For the 2D feature analysis, a more robust method to isolate the "core" regions of azimuthal shear had to be developed to avoid detecting regions of azimuthal shear not associated with vortex centers.

Vortex classification is now done at the 3D level. Here, the MDA must classify a plethora of weak-to-moderate-to-strong vortices of varying strengths and dimensions (depth, diameter, etc.). The weakness of the WSR-88D algorithm is that this classification was done by strength trehsolding at the 1D pattern vector level. This is now done at the 3D level. Also, this allows the algorithm to classify additional types of vortices, such as low-level (gustnado), or low-topped storm vortices (common in supercells east of the Mississippi).

Classification of vortices doesn't stop there. All the output from the MDA is being used to train a neural network to determine whether a vortex is tornadic or non-tornadic.

Critical Success Indices (CSI) showing the skill of each algorithm to correctly identify tornadic mesocyclones:

More information: An Analysis of MDA and TDA Data by Caren Marzban