SAR Polarimetry for Sea Ice Monitoring

1
SAR Polarimetry for Sea Ice Monitoring

• Wolfgang Dierking1,2, Henning Skriver2, and
  Preben Gudmandsen3
• 1 Alfred Wegener Institute for Polar and Marine
  Research, Germany
• 2 Ørsted-DTU, Dept. of Electromagnetic Systems,
  Technical University of Denmark
• 3 Technical University of Denmark

2
ESA POLSAR ProjectSea Ice Study

• Our presentation focusses on two questions
• Do we gain anything by utilising polarimetric
  data (intensities phase differences) in sea ice
  classification ?
• What is the optimal strategy for sea ice
  classification ?

3
Data Sets
We utilised airborne SAR imagery from the flight
campaigns listed below

• AIRSAR Beaufort Sea (March 1988), LC-band
• EMISAR Greenland Sea (March 1995), C-band
• EMISAR Baltic Sea (March 1995), LC-band (EMAC
  campaign)
• (Winter conditions, ice regimes at the 3
  sites are different from one another.)

4
Polarimetric Phase ?HHVV

• Including ?HHVV and ?HHVV in a sea ice
  classification scheme does not significantly
  improve the classification accuracy. (Rignot and
  Drinkwater, 1994)
• Thin ice areas often reveal a ?HHVV
  significantly different from zero. Is ?HHVV
  linked to the thickness of thin ice ?
  (Winebrenner et al., 1995 Thomsen et al.,
  1998a,b Thomsen, 2001)
• A fully polarimetric model (intensity and phase)
  has been used to retrieve thin ice thickness by
  means óf a neural network. Result C-band data
  seems to work for thicknesses 0-10cm, L-band is
  less sensitive. Contribution of individual
  polarization coefficients ? (Kwok et al., 1995)

5
Intensity R-?0XP, G-?0HH, B-?0VV
Co-polarisation Ratio ?0VV / ?0HH
Phase ?HHVV Versus Intensity Example
1 Greenland Sea C-Band
Depolarisation Ratio ?0HV / (?0HH ?0VV)
Phase Difference ?HHVV
6
Intensity R-?0XP, G-?0HH, B-?0VV
Co-polarisation Ratio ?0VV / ?0HH
Phase Difference ?HHVV
Phase ?HHVV Versus Intensity - Example 2
Greenland Sea, C-Band
7
Phase difference ?HHVV Observations 1
8
Phase difference ?HHVV - Observations 2
9
Thin Ice Radar Signatures C-band
scatterometer measurements over growing ice in a
cold room at CRREL Backscattered intensity at
like- and cross- polarisation increased by 6-10
dB as the ice thickened from 3cm to 11cm. (Nghiem
et al., 1997)
TH1
TH2
TH3
TH4
10
Phase Difference ?HHVV

• Improves classification discrimination thin ice
  open water
• Linked with thin ice thickness ?
  (classification, heat and salt fluxes)
• Needs further research what determines the
  magnitude of ?HHVV ?
  (brine inclusions, anisotropic volume
  scattering from the ice-water interface,
  dielectric profile)

11
Sea Ice Classification

• Our choice is a
• hierarchical scheme (knowledge-based approach)
• WHY ?
• Results of measurements and theoretical modelling
  of sea ice radar signatures, and the experience
  gained from field campaigns can be considered.
• Decision boundaries at the individual levels in
  the hierarchy can be determined by means of
  statistical methods.
• (A similar procedure is applied at the ASF,
  Kwok et al., 1991)

12
Methodology Step 1

• We determined typical values of various
  polarimetric parameters for different ice types
  visually (subjectively) identified in the
  radar images.
• Polarimetric parameters
• Covariance matrix intensities VV, HH, XP,
  correlation and phase
• difference HHVV, co- and depolarisation
  ratio, symmetry.
• Decomposition (coherency matrix) entropy,
  alpha, anisotropy.

• Visual classification on the basis of
• a 3-layer image format representing only
  intensities (R-HV, G-HH, B-VV)
• complementary data (photos, videos, in-situ spot
  measurements, meteorological data).

13
Greenland Sea
Baltic Sea
Polarimetric parameters of different ice types -
Example
Beaufort Sea
14
Methodology Step 2

• For the classification scheme, polarimetric
  parameters have been selected for which distance
  between ice type data clusters is largest

15
Methodology Step 3
We devised classification rules for each test
site and radar band
Classification Rules for Greenland Sea Ice
(Winter)
16
Classification, Greenland Sea, C-Band
Hierarchical Approach, ISODATA thresholds
Intensity R(HV) G(HH) B(VV)
17
Classification, Greenland Sea, C-Band
Used Parameters ?0HV, ?HHVV ?0VV / ?0HH ?0HV /
(?0VV ?0HH) Classes 1 Ridged ice (39) 2 MY
ice 1(141) 3 MY ice 2 (98) 4 Thin ice 1-3 (101) 5
Thin ice 4 (49) 6 Open Water (33)
Confusion matrix for hierarchical
classification 1 2 3
4 5 6 1 100.0
0.0 0.0 0.0 0.0 0.0
2 0.0 98.5 0.7 0.7 0.0
0.0 3 0.0 7.9 92.1
0.0 0.0 0.0 4 0.0
0.9 13.4 84.8 0.9 0.0
5 0.0 0.0 0.0 9.4 90.6
0.0 6 0.0 0.0 0.0
0.0 0.0 100.0     Confusion
matrix for ISODATA with hierarchical
classification as initialisation 1
2 3 4 5 6
1 100.0 0.0 0.0 0.0
0.0 0.0 2 0.0 99.3 0.0
0.7 0.0 0.0 3 0.0
4.5 93.3 2.2 0.0 0.0
4 0.0 0.9 11.6 86.6
0.9 0.0 5 0.0 0.0 1.9
5.7 92.5 0.0 6 0.0
0.0 0.0 0.0 0.0 100.0
18
Classification Beaufort Sea
19
Classification Baltic Sea
20
Classification Decomposition

• Indicates scattering
• mechanisms from sea ice
• Z9, Z6 surface scattering
• with an increasing amount
• of secondary scattering
• contributions
• Z8, Z5 volume scattering
• from inclusions with a
• decreasing correlation of
• their orientation
• Z2 noise-like scattering
• from randomly oriented
• scatterers

21
Sea Ice Classification

• Regional and seasonal differences in ice cover
  characteristics require different optimal
  sequences of classification rules (are they
  stable for a particular region and season ?).

22
Which frequency, which polarisations ?
X-band (intensity) is very similar to C-band
23
Thank you for your attention !