Storm-scale Vortex Detection and Diagnosis

1
Storm-scale Vortex Detection and Diagnosis

• Real-Time Mining of Integrated Weather
  Information Meeting
• 20 September 2002
• gstumpf_at_ou.edu

2
INTRODUCTION

• The WSR-88D system has two independent algorithms
  for detecting rotation in severe thunderstorms
• Mesocyclone Detection Algorithm (MDA Fall 2003)
• Tornado Detection Algorithm (TDA current).
• Current NWS requirements include goals toward
  improving the probability of detection, false
  alarm rate, and lead time for tornado warnings.

3
INTRODUCTION

• This talk focuses initially on our reasons for
  changing automated vortex detection techniques.
• Description of current techniques
• Examples of why a change is needed
• NSSLs proposed solution is a Vortex Detection
  and Diagnosis Algorithm (VDDA).

4
OLD RADAR PARADIGM

• Early studies (1970s, 1980s) used the only
  available Doppler radar data at the time (mostly
  Central Oklahoma).
• Mesocyclones and Tornado Vortex Signatures (TVS)
  are rotating columns of air in thunderstorms with
  specific spatial and strength criteria.
• How do we define a mesocyclone or TVS? What
  is operationally-significant?

5
WSR-88D NETWORK COMPLETED

• We have gathered a plethora of storm data from
  around the country.
• Not all storms are like those observed in Central
  Oklahoma in the 70s and 80s.
• Many storm and storm-scale vortex types can be
  associated with tornadoes
• Field projects (e.g., VORTEX) have observed
  interesting new things too.

6
NEW ALGORITHMS DEVELOPED IN THEMID 1990s

• On the radar, a Big/strong vortex does not
  always Tornado, and vice versa.
• The future algorithm should detect many
  storm-scale vortices of many sizes and strengths
  (including those lt 1 km).
• Decision makers and future algorithms should
  integrate all available information.

7
Mesocyclone Detection Algorithm

• Designed to detect storm-scale (1-10 km diameter)
  3D vortices.
• High POD
• Can track and trend incipient vortices through
  maturity, on to demise for complete time history.
• Next, the NSSL MDA diagnoses and classifies the
  vortices.
• To determine which are operationally significant.
• Includes probabilistic output from Neural
  Networks.

8
Mesocyclone Detection Algorithm

• Detects 2D features by searching for azimuthal
  shear cores
• Ranks 2D features based on vortex strength
  thresholds
• Vertically-associates 2D feature centroids to
  produce 3D detections.
• Time-associates 3D centroids to produce 4D
  tracks, trends, and extrapolates for forecasts.
• Classifies 4D detections as mesocyclones based
  on thresholds for base, depth, and strength.

9
Storm-relative Velocity
Reflectivity
330o
330o
Mesocyclone
Mesocyclone
100 km
100 km
-4.5
-20.0
22.0
-20.5
17.5
21.5
19.5
99.75
23.5
23.0
-22.5
-9.0
-22.0
20.0
22.5
99.50
24.5
-25.5
24.0
-12.0
-22.0
24.0
20.5
99.25
27.0
-26.5
26.0
27.0
-20.5
-25.0
21.0
99.00
Shear Segments
28.5
-27.0
24.5
28.0
-15.0
-25.0
20.5
98.75
Range (km)
21.5
29.5
-23.5
-7.5
-18.5
30.5
21.0
98.50
29.0
20.5
-23.5
19.5
28.0
-8.5
-19.5
98.25
28.5
-23.0
14.5
27.5
-5.5
-19.5
20.0
98.00
27.5
15.5
26.5
-20.5
-5.5
-11.0
20.0
97.75
335.5o
334.5o
333.5o
332.5o
331.5o
330.5o
329.5o
Azimuth
10
Vertical Association
11
Time Association
Search Radii
current detections
a
b
c
first guess
previous position
12
Time Association
Search Radii
Associated current position
previous position
13
Time Association
Search Radii
Associated current position
5-min Forecast position
previous position
14
Tornado (TVS) Detection Algorithm

• Very similar techniques to those used by MDA for
  2D, 3D, and 4D detection and classification.
• The main differences
• 2D feature detection based on gate-to-gate
  azimuthal shear
• Choice of classification thresholds designed for
  stronger and more intense vortices

15
Storm-relative Velocity
Reflectivity
330o
330o
TVS
100 km
100 km
-4.5
-20.0
22.0
-20.5
17.5
21.5
19.5
99.75
23.5
23.0
-22.5
-9.0
-22.0
20.0
22.5
99.50
24.5
-25.5
24.0
-12.0
-22.0
24.0
20.5
99.25
27.0
-26.5
26.0
27.0
-20.5
-25.0
21.0
99.00
Shear Segments
28.5
-27.0
24.5
28.0
-15.0
-25.0
20.5
98.75
Range (km)
21.5
29.5
-23.5
-7.5
-18.5
30.5
21.0
98.50
29.0
20.5
-23.5
19.5
28.0
-8.5
-19.5
98.25
28.5
-23.0
14.5
27.5
-5.5
-19.5
20.0
98.00
27.5
15.5
26.5
-20.5
-5.5
-11.0
20.0
97.75
335.5o
334.5o
333.5o
332.5o
331.5o
330.5o
329.5o
Azimuth
16
MDA/TDA Limitations

• Operates using single-radar radial velocity data.
• Volume scan output 5-6 minutes older than first
  elevation scan (nearest surface). Reduces
  lead-time.
• Very sensitive to artifacts in radar data
• Dealiasing errors within storm and non-storm echo
  (anomalous propagation, ground clutter, chaff,
  clear air return, first trip ring)
• Beam broadening, cone-of-silence, radar horizon
• Vortex radius to beam width ratio, beam center
  offsets.

17
MDA/TDA Limitations

• Reflectivity data used crudely to filter
  velocities associated with non-storm echo (single
  0 dBZ threshold).
• Heuristic threshold-based rules.
• Azimuthal shear is associated with rotation, but
  also associated with boundaries (aligned parallel
  to radar radials) and other phenomenon.

18
THE VDDA

• The next-generation Vortex Detection and
  Diagnosis Algorithm (VDDA) will be designed
  detect a broader spectrum of storm-scale vortices
  (including TVSs as well - merging MDA and TDA).
• The VDDA will integrate new ideas learned from
  the new radar data, and data from radar
  reflectivity and other sensors (e.g. near-storm
  environment, satellites, etc.).

19
VDDA Considerations

• Current algorithm paradigms are that a TVS is a
  gate-to-gate signature, and that a mesocyclone is
  not.
• These assumptions do not hold true, especially at
  near and far ranges.
• A mesocyclone at far ranges can exhibit
  gate-to-gate shear.
• A tornado (or tornado cyclone) at near ranges can
  be sampled across more than two adjacent radar
  azimuths.

20
VDDA Considerations

• The radar characteristics of the signature are a
  function of
• Vortex core radius to beamwidth ratio (i.e.,
  distance and diameter of vortex).
• Rotational Velocity
• Offset between the radar beam centroid and the
  vortex centroid.

21
VDDA Considerations

• Observational studies have shown that a variety
  of vortex scales (0 to 10 km) can be associated
  with tornadoes, tornado cyclones, and
  mesocyclones.
• Not all tornadoes are associated with the classic
  definition of a supercell
• Supercells with small horizontal dimensions
  (mini-supercell).
• Supercells with small vertical dimensions
  (low-topped supercells).
• Tropical-cyclone mesocyclones (TC-mesos).
• Bow echo tornadoes (along the leading edge).

22
(No Transcript)
23
Low-topped mini-supercells
KLWX Sterling VA 30 April 1994
24
Not all mini-supercells are low-topped!
Cone of Silence
KPUX Pueblo 22 June 1995
25
Tropical Cyclone Mesos (low-topped and mini)
KMLB Melbourne FL 11 Nov 94
T.C. Josephine
26
Heres a TC-meso that is low-topped, but NOT mini!
KEVX Eglin FL 4 Oct 1995
Hurricane Opal
27
Leading-edge tornadoes (shallow, short-lived)
KLSX St. Louis 15 April 1994
28
Leading-edge tornadoes (shallow, short-lived)
KLSX St. Louis 15 April 1994
29
VDDA feature extraction and forecasting

• Develop new 2D and 3D vortex feature extractor
  utilizing testing on analytically-modeled
  vortices (with sampling limitations), boundaries,
  etc.
• Least-squares shear derivatives (LLSD)
• Statistics-based image processing (K-means)
• Advanced motion estimation

30
LLSD

• A linear least-squares fit of radial velocity
  bins in the neighborhood of a gate.
• The number of data bins in the neighborhood
  depends on the range from the radar.
• A constant kernel size means more data bins at
  close ranges with polar grids.
• Fit to a linear combination of azimuth and range
• Coefficient for azimuth is an estimate of
  azimuthal shear or rotation
• Coefficient for range is an estimate of the
  radial shear or divergence/convergence.

31
Linear Least Squares Derivative (simulated data)
32
Linear Least Squares Derivative (actual data)
Actual WSR-88D Velocity
Azimuthal Shear (LLSD)
Radial Shear (LLSD)
Anticyclonic Shear
Meso
TVS
33
Simulated radial velocity data of a variety of
phenomenon

• Follows the method of Wood and Brown (1997).
• Simulate symmetric vortices of varying strength,
  size, and radar sampling
• Varying Ranges 2 to 200 km
• Varying Rotational Velocities 5 to 50 m/s
• Varying Diameters 0.25 to 6 km
• Varying Beam/Vortex center offsets 0.5o to 0.5o
• Other special simulations
• Mesocyclones with rear-flank downdrafts.
• Mesocyclones with embedded TVS.
• Straight gust front boundaries

34
Storm-relative velocity
Straight Gust Front
Meso with Rear- Flank Downdraft
Meso with embedded TVS
Pure Rankine Vortex (Meso)
35
SRV with NSSL MDA 2D features
Straight Gust Front
Meso with Rear- Flank Downdraft
Meso with embedded TVS
Pure Rankine Vortex (Meso)
36
Azimuthal and Radial shear
Straight Gust Front
Meso with Rear- Flank Downdraft
Meso with embedded TVS
Pure Rankine Vortex (Meso)
37
VDDA feature extraction and forecasting

• Combine rotation and divergence fields from
  multiple radars into 3D mosaicked grid.
• Rapidly-updating grid provides greater lead time.
• Use multi-scale statistical texturing techniques
  (e.g., Kmeans) to extract 2D and 3D cores of
  rotation.

38
VDDA feature extraction and forecasting
39
VDDA feature extraction and forecasting

• Diagnose properties of the rotation cores to
  determine probability that they are associated
  with severe weather or tornadoes.
• Instead of centroid extrapolation, use
  statistical motion estimator to forecast vortex
  locations, could provide more lead time.

40
Multiple-source integration

• Use information from multiple-radar reflectivity
  data (vertical profiles or VIL), near-storm
  environment, and IR satellite to filter out
  non-storm echo prior to rotation core extraction
  (instead of just 0 dBZ thresholds).
• Integrate information from BWER, hook echo ID,
  boundary ID (which also uses LLSD), total
  lightning data (CG and IC), and near-storm
  environment data for vortex diagnoses (e.g.,
  Neural Networks).

LLSD Convergence
41
Current VDDA Work

• Testing process to compute LLSD at different
  scales.
• Simulated vortices with random noise (1000
  trials), and various vortex diameters.
• Compared to azimuthal shear.