Technical Interchange Meeting – ROC / NSSL / NCAR ROC / NSSL / NCAR TIM Boulder CO 11 May 2005...

Post on 18-Jan-2016

219 views 0 download

Tags:

Transcript of Technical Interchange Meeting – ROC / NSSL / NCAR ROC / NSSL / NCAR TIM Boulder CO 11 May 2005...

Technical Interchange Meeting – ROC / NSSL / NCAR

ROC / NSSL / NCAR TIM

Boulder CO

11 May 2005

Real-time time-series implementationof the Radar Echo Classifier (REC)

for clutter detection in ORDA

Mike Dixon

NCAR

Goal

The goal was to develop and test a version of the REC with the following properties:

• Fast and efficient for real-time operation, suitable for use in the

ORDA.

• Works with time series data, so that the algorithm has access to the

spectral domain.

• Is suitable for detecting clutter and AP.

Beam processing sequence

Beam 1 Beam 2 Beam3 Beam 4 Beam 5

IN

OutBeam Queue

ComputeMoments

ComputeMoments

ComputeMoments

ComputeMoments

ComputeMoments

ComputeREC

FilterClutter

FilteredMoments out

REC for clutter or AP detection

• Kernel: 5 deg wide, 2 km along the beam.

• Uses the following fields:

– TDBZ - DBZ texture: squared change in dBZ from one gate to the next,

in range, averaged over the kernel.

– SPIN - DBZ ‘spin’: measure of how frequently the trend in reflectivity

along a beam changes with range. Averaged over the kernel.

– VEL: velocity at the gate.

– SDVE: standard deviation of velocity over the kernel.

– WIDTH: spectrum width at the gate.

– CLUTPROB: clutter probability, based on ratios of power near 0 m/s to

power in rest of spectrum.

Membership functions

0 45 1000

1

0

TDBZ SPIN

0 50 100

0

1

0 3.2

1

0

WIDTH

0 0.7

SDVE

0

1

VEL

-2.3 0 2.3

0

1

0 3 15

CLUTPROB

0

1

Data sets

This implementation of the REC was developed to handle time-series data in LIRP format. It was tested on the following data sets:

• KJIM, stratiform situation, non-phase-coded.

• SPOL at Boulder, convective situation, phase-coded.

• SPOL at NAME, Mexico, convective situation, non-phase-coded,

alternating-pulse dual-polarization.

KJIM Case

• Non-phase-coded data

• Stratiform rain to NW

• Ground clutter

KJIM dBZ

KJIM Vel

KJIM WIDTH

KJIM TDBZ

KJIM SPIN

KJIM SDVE

KJIM Clutter Probability

KJIM REC

KJIM Clutter Flag

KJIM dBZ

KJIM dBZ filtered

KJIM VEL

KJIM VEL filtered

KJIM WIDTH

KJIM WIDTH filtered

KJIM filter everywhere

KJIM dBZ

KJIM dBZ filtered everywhere

KJIM vel

KJIM vel filtered everywhere

SZ Case - SPOL

• SZ864 decoding

• Strong mountain ground clutter

• Convective weather situation

SZ dBZ

SZ VEL

SZ WIDTH

SZ Trip flags

SZ TDBZ

SZ SPIN

SZ SDVE

SZ REC

SZ REC Clutter Flag

SZ Clutter found

SZ dBZ

SZ dBZ filtered

SZ VEL

SZ VEL filtered

SZ WIDTH

SZ WIDTH filtered

Dual Polarization Case – SPOL at NAME

• Alternate-pulse dual polarization

• Strong ground clutter

• Some sea clutter at times

• Convective weather situation

Additional REC fields for Dual Pol

The following fields were added to the REC for the dual

polarization case:

– RHOHV – value at the gate.

– SD-ZDR – standard deviation of ZDR in range,

computed for the single beam only, no azimuth

averaging.

– SD-RHOHV – standard deviation of RHOHV in range,

computed for the single beam only, no azimuth

averaging.

Membership functions – Dual Pol

0 0.8 0.95

1

0

RHOHV

0 2 3

1

0

SD-ZDR

0 0.02 0.03

1

0

SD-RHOHV

Dual-pol dBZ

Dual-pol VEL

Dual-pol WIDTH

Dual-pol TDBZ

Dual-pol SPIN

Dual-pol SDVE

Dual-pol ZDR

Dual-pol SDZDR

Dual-pol RHOHV

Dual-pol SD-RHOHV

REC – no dual-pol fields

REC with dual pol fields

REC FLAG – no dual-pol fields

REC FLAG with dual pol fields

Dual-pol Clutter found

Dual-pol dBZ

Dual-pol dBZ filtered

Dual-pol VEL

Dual-pol Vel filtered

Dual-pol WIDTH

Dual-pol WIDTH filtered

Relative performance considerations

Some tests were carried out on the computer time taken by the REC and clutter filtering as compared to moments estimation.

The following table shows the number of seconds taken to compute moments, REC and filter clutter for a single PPI for each of the cases. The test machine was a 2.8GHz Pentium IV.

These numbers are useful to show the relative costs of each operation.

KJIM, filter only where

REC > thresh

KJIM, filter clutter everywhere

SZ, filteronly where REC > thresh

Dual Pol, filter only where REC > thresh

Moments estimation

3.18 3.18 9.58 8.23

REC computation

0.53 0.53 1.16 0.54

Clutter filtering

2.37 9.89 2.89 1.35

Further work

• Clutter filtering – handling residual power.• SZ clutter filtering.• Tuning the REC for dual polarization data.• Pattern matching for spectra in range (NESPA,

NIMA).

Thank you