Automated Classification of Storms Based on Radar-Derived Storm Properties

1
Automated Classification of Storms Based on
Radar-Derived Storm Properties
Valliappa Lakshmanan, Travis Smith, Robert
Rabin University of Oklahoma National Severe
Storms Laboratory, Norman OK, USA
i. Goal
To identify the storm type (supercell, linear,
pulse storm or non-organized) in real-time.
ii. Why?
1. Automated classification can be used to create
climatology of storms across CONUS 2. The
climatology can be used to create guidance for
probabilistic warnings.
iii. Technique
1. Some of the spatial grids input into the storm
type algorithm these are derived from
multi-radar 3D grids created in real-time for all
WSR-88D in CONUS
2. Pixels in the reflectivity composite field are
clustered to find storms at different scales
(20km2, 160km2, 480 km2). Properties are
extracted from grids on left at these scales.
Reflectivity near ground
Reflectivity _at_ 11km
Reflectivity _at_ -20C
Prob. Of Significant Hail
Step 2
Az Shear 0-3km
VIL
5. Storm type algorithm running in real-time. The
results are shown visualized using Google Earth
4. Train decision tree on data
3. Human-training of storm-type algorithm,
classifying storms into 4 types supercell, line,
pulse, unorganized
1. More categories of storms 2. A broader, more
diverse training set 3. Build climatology in
collaboration with NCDC
iv. Future Plans
Yes, you can! Download the software from
http//www.wdssii.org/ and run w2segmotionll
v. Can I try this on my data?
Please do stop me if you see me in the hallway!
Id love to address any questions or comments.