Hydrometeor Classification Using Polarimetric Radar

1
Hydrometeor Classification Using Polarimetric
Radar

• Kyle C. Wiens
• AT741
• 15 April 2004

2
Why me?

• This was my project last time AT741 was taught.
• I wrote my own algorithm from scratch.
• I use it in my research.
• The concepts and application of fuzzy logic are
  very simple, so...
• If you don't understand, it's because I'm not
  explaining it well enough.

3
Outline

• Brief review of polarimetric variables.
• Combining these variables to get information
  about hydrometeor types
• Methods of combining these variables
• Fuzzy logic description
• Examples from 29 June 2000 supercell

4
Polarimetric Variables

• ZH Size, concentration
• ZDRShape, orientation, (liquid or solid)
• KDPAmount of liquid water, size of drops
• LDROrientation, canting, melting (wet hail)
• ?HVCorrelation (mixture of types/melting)

5
Hydrometeor Identification

• Combine variables and define sub-ranges over
  which specific hydro types are expected

Straka et al. (2000)
6
Hydrometeor Identification

• Extend this concept to the six-dimensional space
  (5 radar variables plus temperature)

Straka et al. (2000)
7
Combining the variables

• In a nutshell
• We have inputs (radar variables)
• We have some decision process
• We get a result (hydrometeor type)

Basically, we want the hydrometeor type that best
fits the inputs.
8
Combining the variables (the old way)

• Look-up table (decision tree), e.g.,
• IF ZHgt55
• AND (-1ltZDRlt0.5)
• AND (-0.5ltKDPlt1)
• AND (LDRgt-24)
• ...
• THEN hydrometeor Large Hail
• But this is not a good way to do it because
• Sub-ranges for each hydro type are not mutually
  exclusive (i.e., they overlap)
• Very inefficient and not comprehensive

9
Combining the variables(the fuzzy logic way)

• Define functions for each input variable and each
  hydrometeor type.
• These functions describe to what degree each
  variable is a member of the hydrometeor type
  family.
• Another way to think of this is that each of
  these functions gives a score to each input
  variable. The higher the score, the greater the
  membership value of that variable to that
  hydrometeor type, i.e., the more likely it is
  that type.

10
Types of Membership functions

• Trapezoid Membership Function Zrnic et al. (2001)

2-D TMF for Moderate Rain
11
Types of Membership functions

• Membership Beta Function Liu and Chandrasekar
  (2000)

2-D MBF for rain
12
Combining the variables

• So, each variable (x) gets a score (?) based on
  where it falls on the membership function for
  each hydrometeor type.
• Score ?(x,m,a,b)

The score is like a probability that the radar
measurement is due to that specific hydro type.
13
Liu and Chandrasekar (2000)
14
Rain!
Hail!
15
Combining the variables

• Now just combine the scores for all the variables
  to get a total membership or truth value (µ)
  for each hydro type.
• Combine as a weighted sum
• Or as a product
• Or some combination of sums and products (which
  is what I do).

16
Combining the variables(My method)

• Take weighted sum of MBFs for polarimetric
  variables
• Then multiply this by MBFs for ZH and temperature
  to get total truth value

17
The Final Result

• So, now we have total truth values (?j) for all
  hydro types.
• The hydrometeor type with the highest total
  truth value wins.

18
Schematic of Fuzzy Logic Process
Liu and Chandrasekar (2000)
19
Examples from 29 June 2000 Supercell

• This storm
• Produced large hail
• F1 tornado
• Well-defined BWER
• It was also very well observed by 3 radars (two
  of which were polarimetric)
• Also had T28 aircraft penetrations which found
  hail.

20
Height 3km
21
Height 7km
22
(No Transcript)
23
LDR Cap
Zdr column
Big drops
24
Hail counts
Hail size
T28 Comparison