Quality Control of Radar Data with a focus on Polarimetric Applications

1
Quality Control of Radar Data with a focus on
PolarimetricApplications
1
2

• One of the most beneficial aspects of
  polarimetric radar is that quality control of
  anomalous propagation, ground clutter, second
  trip, etc. can be accomplished much easier than
  with traditional Doppler radars (where only Zh,
  Vr, SW, and possibly NCP, or normalized coherent
  power, and the vertical and temporal structure of
  reflectivity are available)
• The distinct signatures of non-meteorological
  echo in the polarimetric fields can often be used
  to remove the above from data of interest
• In an operational setting (i.e. for estimating
  rainfall), QC can provide a big headache in
  generating real-time products (where hand-editing
  is not feasible), and polarimetric QC can be used
  in an automated sense

3

• Non-Polarimetric QC options
• Hand editing via solo (not feasible for large
  datasets)
• Threshold reflectivity lt some value
• Threshold Vr near 0 to attempt to remove
  stationary clutter
• Threshold normalized coherent power (NCP) below
  some value
• NCP mean power at each pulse/total power
• Examine horizontal and vertical structure of
  echoes
• Second trip echo will have non-coherent velocity
  structure (I.e. high standard deviation with
  range)
• Despeckling

4
Zh lt 5
NCP lt 0.7
-0.5 lt Vr lt 0.5
5
Second-Trip example
http//apollo.lsc.vsc.edu/classes/remote/lecture_n
otes/radar/conventional/graphics/sec_trip_dzve.gif
6
The purpose of the 1C51 algorithm is to remove
non-meteorological radar echoes that adversely
affect the quality of higher level products, such
as clutter associated with insects, birds, chaff,
wildfires, antenna towers, and anomalous
propagation (AP). Eight adjustable parameters,
three echo height thresholds and five radar
reflectivity thresholds, are used to optimize the
performance of the algorithm. Optimum performance
is time consuming and requires an analyst to
select different sets of parameters on a per
volume scan basis, and on occasions performing
several iterations while adjusting one or more of
the parameters.
http//trmm-fc.gsfc.nasa.gov/trmm_gv/gv_products/G
Vproducts.html
7
Applied NCP, Zh, and Vr filter despeckling
filter
8
These QC methods have their drawbacks as
well NCP - clutter has high values which appear
like precipitation Zh threshold - removes light
precipitation and other features of interest
(I.e. gust fronts, etc.) Vr threshold - ground
clutter has near zero velocity, however swaying
branches and moving leaves may return non-zero
velocities - also removes echo in precipitation
near the 0 m/s isodop Vertical structure of
reflectivity - range sensitive due to resolution
issues, may remove shallow precipitation,
especially at far ranges The moral of the story
is - these methods require either a lot of work
or lots of tuning to get something that is
useful
9
a lot of junk still gets through the QC filters.
10
For hydrologic products, QC remains a big problem
because these false echoes can blow up
reflectivity-based rainfall estimates. - Impacts
river monitoring, flood forecasting and warning
(particularly in complex terrain) Hail
contamination (oscillations in reflectivity due
to Mie scattering) can also cause large errors -
NWS is forced to cap rainfall estimates at Zh lt
53 dBZ However, many QC techniques have been
developed for polarimetric radar, because often
times the consistency of the horizontal and
vertical transmitted pulse falls apart when the
Rayleigh-Gans theory breaks down (i.e. for
non-meteorlogical scatterers) We can use this
information to delete such echo in polarimetric
radar data, and these techniques can be used in
an automated sense.
11
Zh
fdp
?hv
Zh lt 5
NCP lt 0.7
-0.5 lt Vr lt 0.5
12
What polarimetric variables are useful for QC?
13
?hv Distinctly different correlation coefficient
values for clutter vs. meteorological echo
rain
clutter
Zrnic et al. (2006)
14
Rainfall and other hydrometeors are typically
characterized by a correlation coefficient near
unity while clutter and clear air have very low
correlation coefficient. E.g., Rainfall ?hvgt
0.97 clutter and clear air ?hv lt 0.6
typically Therefore, it is possible to chose a
simple threshold in ?hv that separates most
hydrometeor echo from most non-hydrometeor echo.
Even when the correlation coefficient is lowered
due to the presence of large wet hail, large
melting aggregates, or mixtures of hydrometeors
(rain and hail), values are still typically in
excess of 0.8.
15
Value of ?hv for a given hydrometeor type is
dependent on the quality of the radar system
design. The better the polarimetric purity
(e.g., good matching of sidelobe patterns at H/V
and good cross-pol isolation) the higher the ?hv
Must consider quality of radar design and
corresponding data in order to choose the
appropriate threshold. BOTTOM LINE - Know
polarimetric properties of rainfall for your
radar and then conduct sensitivity tests
16
Zh
?hv
Zh (corrected)
Thresholded on ?hv lt 0.85, Despeckled
17
Zh
Fdp
Zh (corrected) From last slide
18
The standard deviation of the differential phase,
s(?dp), in light rain or drizzle is typically a
few degrees or less s(?dp) 2 to 3 deg/km. Again,
depends on quality of radar system design. The
value can be larger (smaller) for a poorly (well)
designed system. For moderate-to-heavy rain,
s(?dp) increases proportionally to specific
differential phase (Kdp) and the rain rate (R)
but rarely ever exceeds 9 deg/km over a path
length of several km (e.g., 3-4 km). For ice
hydrometeors whether ice crystals, aggregates,
graupel, or hail, s(?dp) is typically small (2 to
3 deg/km or less) unless the particles are in the
Mie scattering regime (i.e., not small relative
to the wavelength). Mie effects are typically not
a problem for S-band but can become an issue at
higher frequencies (C-band, X-band etc).
19
The standard deviation of phase is calculated
along each ray of the radar data. Usually some
window length is selected to calculate the
value over, typically 3-5 km depending on the
type of precipitation your are viewing (shorter
length for heavier or more widespread
precipitation). The calculation moves along the
ray with range, and is simply s(Fdp) ( 1/N S
(Fdpi - Fdp)2 ) The values are recorded at the
center of the window within the radar data.
20
s(Fdp)
Fdp
Zh
Zh and Fdp thresholded on s(Fdp) lt 12 and ?hv lt
0.8
21
(No Transcript)
22
(No Transcript)
23
Alas, Even with polarimetric techniques, we still
need to watch our for the occasional glitch.
24

• Summary for Non-hydrometeor Rejection by
  Polarimetric Methods
• Advantages of polarimetric approach
• Better than traditional approaches, more
  automated
• Simple empirical technique -thresholds in ?hv and
  s(?dp).
• Solid foundation in theory.
• Thresholds determined from theory, quality of
  radar design and data, and sensitivity testing.
• Effective 99.9 of the time.
• Disadvantages of polarimetric approach
• The 0.1 failure rate can be troublesome,
  especially if systematic.
• Some subjectivity in determination of thresholds.
• No exact (perfect) thresholds
• Raise them higher to delete more non-hydrometeor
  echo and you will end up removing hydrometeor
  echo
• Lower them to keep more hydrometeor echo and you
  will allow more non-hydrometeor echo to pass
  through
• Future work investigate variable thresholds as a
  function of Z or power.
• Current method does not eliminate second trip
  echo
• Need to look to other variables (possibly values
  of Vr, LDR or ?dp relative to typical values.

25
Simulating Polarimetric Observables
withT-Matrixand Mueller-Matrix
25
26

• Modeling the Polarimetric Radar Observables
• T-Matrix and Mueller-Matrix
• Radar polarimetric techniques permit the
  measurement of additional target characteristics
  beyond reflectivity and Doppler velocity.
• For precipitation particles, these polarization
  characteristics are related to the distributions
  of size, shape, composition, and spatial
  orientation of the particles filling the radar
  resolution volume.
• Oblate raindrops oscillate, cant off-vertical
  in windy conditions
• Non-spherical hail and graupel tumbles when it
  falls
• Geometrical considerations are also important to
  consider (i.e. radar transmit polarization, radar
  elevation angle, etc.)
• A forward model has been developed to calculate
  the Mie scattering properties of non-spherical
  dielectric bodies (hydrometeors) ? T-Matrix.
• The T-Matrix model calculates the scattering
  properties for an arbitrary particle.
• The Mueller-Matrix uses the results of the
  T-Matrix for an arbitrary particle and introduces
  the aforementioned geometric considerations
  (particle fall modes, radar scanning geometry)
  such that the radar measurables can be estimated.

27

• Consider the following

T-matrix calculates scattering properties about
the symmetry axis, Mueller-matrix rotates an
ensemble of particles into the proper geometry
Radar transmit polarization (H,V)
radar
Vivekenandan et al. (1991)
28
Considering the geometry on the previous slide
(and below), where the symmetry axis N is
oriented at (?,f) to the incident wave, Holt
(1984) showed for an individual particle (note
elevation angle ? p/2 - ?)
Backscattered electric field
Scattering Matrix (S)
Incident field
where a is the canting angle of the particle.
29
Once the scattering matrix S is calculated with
elements fhh, fvv, fvh, fhv are computed for
arbitrary orientation N for the radar and
forward directions, the standard Stokes
parameters (I, Q, U, V) are used to completely
characterize the scattered field. The Mueller
matrix is constructed as
np is the number concentration of particles, and
Sp indicates a sum over the orientation, size,
particle type distributions.
30
We must also take into account propagation
effects. The extinction matrix K is defined in
terms of the forward scattering amplitudes fij
using the following
where
?Solving these equations along the radar beam
path allows us to now calculate the radar
observables from these matricies.
31
Zhh, Zdr, and LDR can be calculated from S
k is the refractive index
?hv and backscatter differential phase (d) can be
calculated as
and Kdp can be calculated from
32

• Particle orientation and oscillation
• f is usually prescribed as random (usually a
  reasonable assumption unless strong wind shear or
  electric fields cause a deterministic angle)
• q, which represents canting or drop oscillations
  are either modeled as a
• quasi-Gaussian with parameters (?-, s)
• simple harmonic oscillator model with ?? given
  by (where ?- and ?m are the mean and amplitude)

33
0.5

• Model sensitivity tests
• Zdr vs. radar elevation angle
• Assume
• 51 prolate ice spheroid
• D4 mm
• ?8 mm
• Simple harmonic distribution of ?
• (?- p/2, ?m0.5 30.5)
• Common fall modes of needles or ice columns
• Why are the curves different?

30.5
34
Same particle as previous slide. Black curves
are ?m0.5 Dashed _30 curvesare ?m0.5
35
Same particle as previous slides, except is is
oriented in an oblate fashion (?- 0). Possible
when the particle is oriented in a strong
electric field. Fluctuating Zdr behavior due to
Mie effects at this wavelength. Leads to
positive Kdp in such regions.
36
The figures show differential phase shift, Z,
Zdr, and LDR for two successive scans around 30
seconds apart. It can be seen that in this short
interval the differential phase shift flips from
around -5 degrees over 5 km to 10 degrees. The
most likely explanation for this sudden change in
orientation is that there was a lightning strike
that instantly dissipated the growing electric
field, after which the crystals fell back to
their preferred horizontal orientation. There is
a clear increase in LDR when the crystals are
vertically aligned, but no corresponding
signature in Zdr.
150227
150258
http//www.met.rdg.ac.uk/radar/stoppress/negative_
kdp.html