Quality Control of Canadian Radar Reflectivity Data

1
Quality Control of Canadian Radar Reflectivity
Data
Valliappa Lakshmanan, Jian Zhang, Carrie
Langston University of Oklahoma National Severe
Storms Laboratory, Norman OK, USA
What we did
New version
88D version
Raw data
We modified the WDSS-II Quality Control Neural
Network (QCNN) so that it would be able to QC
Canadian radar reflectivity data
XDR June 6, 07
Why Canadian data?
So that we can include Canadian data into our
4-dimensional real-time reflectivity mosaics.
These mosaics are used by both severe weather
algorithms and by precipitation estimation
algorithms at NSSL.
New version
88D version
Raw data
XDR June 6, 07
Why Adapt QCNN?

• Operates in real-time, on virtual volumes so that
  elevations are cleaned tilt-by-tilt.
• Excellent preprocessing filters
• Speckle removal
• Entropy check to remove radar test patterns
• Sun-strobe check to remove radial contamination
• Neural network trained to classify range gates
• Based on texture features computed from
  reflectivity, 3D volumetric features of
  reflectivity, velocity and spectrum width
• Objective and data-driven technique
• Post-processing based on region growing provides
  high (99.9) accuracy

New version
88D version
Raw data
WGJ Oct 1, 07
The changes we made to QCNN and reason for change
Change to 88D Algorithm (to handle Canadian Data) Reason for change (characteristic of Canadian data)
Train reflectivity-only neural network on truthed 88D data Velocity data not collected at same time as reflectivity
Use tilt at physical height (3-5km) instead of next higher tilt when computing 3D features Several scans at low tilts (0.3, 0.5) subject to AP errors
Can I try QCNN on my radar data?
Yes, you can! Download the software from
http//www.wdssii.org/
Please do stop me if you see me in the hallway!
Id love to address any questions or comments.