Potential Assessment of SAR in Compact and Full Polarimetry Mode for Snow Detection
Gulab Singh, Yoshio Yamaguchi, Sang-Eun Park Gopalan Venkataraman
Niigata University, Japan IIT Bombay, India
Outline• Introduction
• SAR Measurements
• Snow monitoring methods
• Study Area: Part of Himalayan Snow and Glacier Covered Region
• Summary
[1] J. C. Souyris, et. al., “Compact polarimetry based on symmetry properties of geophysical media: The π/4 mode,” IEEETGRS, vol. 43, no. 3, pp. 634–646, Mar. 2005.
[2] R. K. Raney, “Dual polarized SAR and Stokes parameters,” IEEE GRSL., vol. 3, no. 3, pp. 317–319, Jul. 2006[3] R. K. Raney, “Hybrid-polarity SAR architecture”, IEEE TGRS, vol 45, no. 11, pp. 3397-3404, 2007.[4] P. Dubois-Fernandez, et. al., “Compact polarimetry at low frequency”, IEEE TGRS vol. 46, no. 10, pp. 3208–3221, 2008Applications in land parameters estimation over flat terrain /region[5] M. Lavalle, “Full and Compact Polarimetric Radar Interferometry for Vegetation Remote Sensing”, Ph.D. Thesis,
Université de Rennes 1, France, 2009.[6] T. L. Ainsworth, J. P. Kelly and J.-S. Lee, “Classification comparisons between dual-pol, compact polarimetric and
quad-pol SAR imagery”, ISPRS Journal of Photogrammetry and Remote Sensing, 64, pp. 464-471, 2009.[7] F.J. Charbonneau, B. Brisco, R.K. Raney, H. McNairn, et. al., “ Compact Polarimetry overview and applications
assessment”, Can. J. Remote Sensing, vol. 36, no. S2, pp. S298-S315, 2010.****************************************************************************************************************************************[8] S. R. Cloude, Polarisation: Applications in Remote Sensing. London, U.K.: Oxford Univ. Press, 2009The compact assumptions in [1],[4]-[6] do not apply to scattering from sloped terrain [2],[3],[7] ,hybrid system3-dB loss in the radar signal , mismatching the transmitter and receiver polarization basisthe system and theoretical justification issues
*****************************************************************************************************************************************
out of several land parameters ……… snow……………
Introduction: previous studies
Snowfall SASE HQ19-01-2006
Snow parameters inmountain areas areparticularly sensitive tochanges in environmentalconditions.
SASE Observatory at Solang, Himachal
Timely information about snow parameters andtheir temporal and spatial variability representsa significant contribution in climatology, localweather, avalanche forecasting and for thehydropower production in high mountainousareas.
Ground-based method represents onlyexact location measurements of fieldobservations which may not berepresentative of a large area or basin.
Due to the strong spatial and timedependent dynamics of snow cover,frequent observation cycles are necessary.
Snow covered : gentle slope
Snow free : Steep Slope
Snow covered: River (Solang Nala) Bank, Himachal
23-01-2009
20-01-2009
SAR
Air/snow interface
snow
snow/ground interface
Ground
SAR interaction with snowpack
m3
m3
m3
m3
Dry snow
ε's>> ε''s
05
101520253035404550556065707580859095
0.5
1.5
2.5
3.5
4.5
5.5
6.5
7.5
8.5
9.5
10.5
11.5
12.5
13.5
14.5
Snow Wetness (Ws in %)
Pe
ne
tra
tio
n D
ep
th (
δp
in
cm
)1.27 GHz
5.6 GHz
9.6 GHzat snow density at 300 kg/m3
ε's>> ε''s
SAR measurements
Single Polarization (ERS-1/2,JERS/PALSAR, Radarsat-1/2, ASAR, TSX)
Dual -Polarization (ASAR, PALSAR, TSX, Radarsat-2)
Quad Polarization (PALSAR, TSX, Radarsat-2)
Compact Polarization (MiniSAR/Chandrayaan-1)
a few satellites are planned by leading space agencies for earth observations
Snow/ice monitoring ??
(Hybrid C-L)
• With the quad polarization capabilities, newergeneration spaceborne SAR sensors areexpected to lead significant improvements ineasily snow identification based on microwavescattering mechanisms
24-05-2010 AVNIR-2 06-06-2010 PALSAR
• Is SAR acquisition in quad polarizationadvantageous as compared to SAR acquisitionin single, dual and hybrid polarization formonitoring snow cover in mountainous area(Himalayas)?
Date Sensor Polarization Off-nadir
angle (0)
Orbit pass GMT
(hh:mm:ss)
Himalayan
Regions
19/05/2007 ENVISAT-ASAR HH+VV 39.1– 42.8 Descending 04:35:58 Badrinath
10/11/2007 ENVISAT-ASAR HH+VV 39.1– 42.8 Descending 04:37:42 Badrinath
12/05/2007 ALOS-PALSAR Quad-Pol 21.5 Ascending 17:04:40 Badrinath
12/11/2007 ALOS-PALSAR Quad-Pol 21.5 Ascending 17:04:31 Badrinath
22/05/2009 ALOS-PALSAR Quad-Pol 23.1 Ascending 17:13:13 Siachen
ENVISAT-ASAR APS and ALOS-PALSAR SLC data
Siachen
Badrinath
N
126cm – 886cm
Snow Monitoring Methods
Based on
-Single Polarization (temporal changes) -Dual –Polarization (Pol. ratio) -Quad Polarization -Compact (Hybrid CL)Polarization
SAR measurements
PALSAR Backscatter Response
σ0
~10 times lower
Problem with single/Dual Pol. SAR data for snow mapping
AVNIR-2 (06-05-07) Snow Map (ASAR) Snow Map (PALSAR) Snow map (PALSAR)
NO
YES
Generate Polarization Fraction value image
ALOS PALSAR Quad Polarization SLC Data
Multi-Looked (6×1) in (Azimuth × Range)
and make Coherency Matrix (T3)
(HV≈VH)
Generate Eigenvalues Image (λ1, λ2, λ3)
PF >=0.55 && Normalized λ3<0.015
Snow Area
Non-snow feasible Area
Extract Scattering Matrix(S)
Polarimetric Speckle Filtering
13
10321
3
PF
Snow Detection Algorithm
(SDA)
Problem with Single/Dual Pol. SAR Data for Snow Mapping - Resolved by Quad Pol.
SDA based Snow Map
Non-snow
feasible Area Snow Cover Area
Discrimination of snow from
other Bragg scattering dominant
surface may be problematic.
L-band fully polarimetric SAR is
not able to detect shallow-depth
snow
snow cover (magenta) derived
from PALSAR (26-05-11),
overlaid to AVNIR-2 (24-05-11)
Agassizhorn region,
Bernese Alps, Switzerland
Study Area(snow and glacier covered terrain)
Part of Indian Himalaya (place of ice )
Siachen Glacier area Standing snow
1.2-8.8 m (low-high altitude)
SWE Product of AMSR-E
Feb., 2007 Aug., 2007
Length ~73 km
FP vs CPFP [C16] [C9] (monosatic) (refl. sym.)[C5]
m-δ CP[J4] refl. & rot. sym. *C’5]
SDA
Tx=LHC, Rx=H,V
[1]-[8]
FP vs CPFP [C16] [C9] (monosatic) (refl. sym.)[C5]
m-δ CP[J4] refl. & rot. sym. *C’5]
Tx=LHC, Rx=H,VPolSARPro ver. 4.1.5 (ESA)
ver. 2.0
SDA
[1]-[8]
[C9] [C5](SHV=SVH) (SHHS*HV ≈ S VVS*VH ≈0)
(SHHS*HV ≈ S VVS*VH ≈0)
[Cꞌ5]
SDACP [J4] =
Degree of polarization
Relative Phase[1]-[8]
ζ0HV/ζ0
HHPF-λ3 approach (CP)PF-λ3 approach (FP) PF-λ3 approach(FP-RS)
m-δ approach
Data
Approach
Dual-Pol
σ0HV/σ0
HH
CP
m-δ
CP
PF-λ3
FP-RS
PF-λ3
FP
PF-λ3
Non-snow feasible area
(%)
71.38 67.21 56.56 45.52 40.98
Snow area (%) 28.62 32.79 43.44 54.48 59.02
FP : Full Polarimetry
FP-RS : Full Polarimetry with Reflection Symmetry condition
CP : Compact Polarimetry
Non- snow feasible area
22-05-2009 SD126-886 cm (low-high altitude)
Summary • Importance of snow studies
• PALSAR backscattering coefficient response for variousfeatures
• Comparisons between single, dual, compact and quadpolarization data for snow detection
• Identification of suitable polarimetric descriptors fordiscriminating the snowpack– PF and normalized λ3
• Results with single polarization SAR (C-&L-band) for snow discriminationnot good.
• Results with dual polarization SAR measurements better than single pol.But it does not care of unwanted topographic distorted area.
• Full polarimetry SAR technique SDA has produced promising results.
• SDA
……takes care of unwanted topographic distorted area
...... suitable for CP too.
****CP shows capability ̴ 15% less than FP****
Summary
Quad Pol.>Compact Pol.>Dual Pol.>Single Pol.
PF-λ3 > m-δ > σ0HV/σ0
HH
Snow Practicability
m-δ Decomposition3-CSPD (Pseudo)3-CSPD (True)
May 22, 2009
0
10000
20000
30000
40000
50000
60000
70000
80000
-3.14 -1.57 0 1.57 3.14
Odd-bounce Scatterers
Peak
Even-bounce Scatterers
Peak
Volume scatterers
Peak
m-δ approachζ0HV/ζ0
HHPF-λ3 approach
Left Circular Transmission (LC)
Rotation: Anti - Clockwise
Wave Vector: H + j • V
Scattering Vector:HH + j • HV 1
j • VV + HV 2
LH
=
LV
H - j V
2
H - j V
2
H +j V
2
ReceptionH
V
Recep
tio
n
Transmission
Badrinath Region
ASAR
SNOW FREE SNOW COVER
Wet Snow Cover in month of May 2006
SNOW FREE SNOW COVER
Wet Snow Cover in month of September 2006
Badrinath Region
ASAR
PF Images
FP ̴3% SCA more than CP
RSI based snow cover mapping
-35
-30
-25
-20
1 31 61 91 121 151
Bac
ksca
tter
ing
Co
effi
cien
t (i
n d
B)
Distance in Pixels
HV VH
-25
-20
-15
-10
-5
0
1 51 101 151 201
HV VH
DEBRIS COVERED GLACIER
SNOW COVERED AREA
Distance in Pixels
Bac
ksca
tter
ing
Co
effi
cie
nt
(in
dB
)Himalayan Region
PALSAR image on 12-05-2007
Symmetry Test (Noise test)
FRA
Slope
Mapover part
of Himalaya
Penetration Depth at Snow Cover
Terrain
Penetration depth can be written with some approximations
ε’ >> ε” as
Dry Snow = ice particles + air + no liquid water
εds' = 1 + 1.7 ρs + 0.7 ρs2
(Tiuri et al. 1984)
εds ' = 1 + 1.5995 ρs + 1.861 ρs3
for ρs <0.45 gm/cm3
(Matzler 1996)
εds" = εice" ( 0.52ρs + 0.62 ρs2 )
where εice" = 0.008 (Matzler 1988)
Wet Snow = ice particles + air + contents of liquid water
Ice
Liquid waterAir Grain Boundary
(Matzler 1987)
f0= 10 GHz