igarss11_mw_v2.ppt
TRANSCRIPT
July 29, 2011 IGARSS 2011, Vancouver 1
A SIMULATION STUDY OF TOPOGRAPHIC EFFECTS ON POLSAR CLASSIFICATION OF
FORESTS AND CROPS
M. L. Williams1 and T. L. Ainsworth2
1Cooperative Research Centre for Spatial Information2Remote Sensing Division, Naval Research Laboratory
July 29, 2011 IGARSS 2011, Vancouver 2
Methodology• Assess the Effects of Underlying Topography on
Polarimetric Forest Backscatter Using Simulated Data• Estimate the Polarimetric Returns not Corrected by
Orientation Angle Compensations– Compare orientation-angle corrected data against forest returns
simulated over flat terrain
• Simulated Imagery Allows One to Tease Apart Forest Scattering Mechanisms – Direct canopy backscatter, ground-trunk interaction, canopy
attenuation, etc.
• Goal: Use the Analysis of Simulated Data to Develop Better Forest Scattering Models for PolSAR Imagery
July 29, 2011 IGARSS 2011, Vancouver 3
Forest Simulations Start With a Tree
Bare Vertical Trunk
Dead Primary Branches
Upper Crown
Model of a Scots Pine, ~18m in heightBased on Detailed Measurements
Lower Crown
Detail of the Crown
Primary and Secondary Branch Structures
July 29, 2011 IGARSS 2011, Vancouver 4
To Make a Forest
• Tree Branches Grow to Edge of Canopy Envelop – Upper Crown – Live Branches– Lower Crown – Dry Branches– Add Needles / Leaves
• Generate a Set of Unique Trees• Semi-Randomly Plant the Forest:
July 29, 2011 IGARSS 2011, Vancouver 5
Scene Generation
View from Radar Platform at Mid-track
• A realization of a random, sloped ground surface generated according to surface roughness, soil moisture and slope parameters.
• Ground surface covered by layer of random vegetation.
• A forest stand realized as a circular area populated by trees.
• Trees have a distribution of heights around the supplied mean value.
• Tree positions are determined by shuffling from an initially uniform grid, with a crown-overlap cost function.
• Red envelopes mark “living” pine crown, grey layer is “dry” timber.
• Each tree is unique and realized in detail for SAR simulation.
July 29, 2011 IGARSS 2011, Vancouver 6
SAR Simulator – PolSARproSim• Basic Simulations Performed with Stand-alone Version
of the PolSARproSIM Software– Coherent, polarimetric SAR simulator – Methodology documented within PolSARpro
• Five Basic Scattering Components – Signal attenuated by all vegetation layers
Direct Ground
Direct Short Vegetation
Direct Forest
Ground-Forest Backscatter
Ground / Short- Vegetation Return
July 29, 2011 IGARSS 2011, Vancouver 7
Scene Summary• Primary Characteristics
– Scots Pine forest - ~2600 trees• Relatively sparse canopy with large trunk structure
• Mean tree height: 18m
• Mean canopy radius: 2.6m
• Canopy attenuates radar signals
– Low vegetation ground cover• Weak scatterer, weak signal attenuation
– Moderately rough, moist ground surface
• Range and/or Azimuth Ground Slopes– Independent variables in this initial study
• Coupled Radar Scattering Components – Attenuation, Ground Slopes & Multiple Scattering
July 29, 2011 IGARSS 2011, Vancouver 8
Polarimetric Analysis• A Select Set of Polarimetric Descriptors:
• Orientation Angle: θ– Related to range(γ) and azimuth(ω) slopes– The complex phase of the RR-LL correlation (circular basis)
∀ α-angle: Average scattering mechanism – Surface: 0°, Dipole: 45°, Dihedral: 90°
• Double Bounce Scattering Strength– Dominant scattering mechanism in our (current) simulated forest– Arises primarily from ground-trunk interactions
• Circular Basis Correlations – |ρ4|: Reciprocal of orientation angle variance
• Magnitude of the RR-LL correlation • Independent of the underlying scattering mechanism
– |ρr|: Reciprocal of the shape variation • Relates to the variance of the effective scatterer shape• Depends explicitly on the scattering mechanism
( ) ( )( ) ( ) ( )φφγ
ωθsincostan
tantan
+−−=
July 29, 2011 IGARSS 2011, Vancouver 9
A SAR Image Realization L-band, 5 Meter Resolution
RGB Image: HH, HV, VV
• Forest covered central region• Low vegetation area surrounds forest• Range increases from left to right
HH HV VH VV
July 29, 2011 IGARSS 2011, Vancouver 10
α-angle Histograms
Range Slope: 0°, 1°, 2° & 4°
Rapid decrease of high α components between 0° & 2°
Azimuth Slope: 0°, 6°, 8° & 10°
No change until ~6° slope then a shift to lower α values
July 29, 2011 IGARSS 2011, Vancouver 11
With azimuth slope set to zero, the orientation angle does not depend on range slope.
0 0.2 0.4 0.6 0.8 10
200
400
600
800
1000
1200
1400
Ort Dispersion (ρ4)
\s_cov_p000_p000
\s_cov_m006_p000
\s_cov_m004_p000
\s_cov_m002_p000\s_cov_p002_p000
\s_cov_p004_p000
\s_cov_p006_p000
-40 -20 0 20 400
200
400
600
800
1000
Orientation (deg)
\s_cov_p000_p000
\s_cov_m006_p000
\s_cov_m004_p000
\s_cov_m002_p000\s_cov_p002_p000
\s_cov_p004_p000
\s_cov_p006_p000
Range Slopes in Grassy RegionsOrientation Angle & Correlation |ρ4|
Low orientation angle correlations without a strong dependence on range slope. Slightly lower for positive slopes.
July 29, 2011 IGARSS 2011, Vancouver 12
Grass Region Forest
Small range tilt reduces double bounce power in forest, but not in the grassy regions.
Range Slopes in Forest AreaDouble Bounce Power & Ground-Forest Component
Ground-Forest Component
The ground-forest multiple scattering accounts for the rapid drop of the double bounce backscatter.
-30 -25 -20 -15 -10 -5 0 50
500
1000
1500
2000
Double Bounce Power (dB)
\s_cov_p000_p000\s_cov_m006_p000
\s_cov_m004_p000
\s_cov_m002_p000
\s_cov_p002_p000
\s_cov_p004_p000\s_cov_p006_p000
-30 -25 -20 -15 -10 -5 0 50
200
400
600
800
1000
1200
Double Bounce Power (dB)
\gfs_cov_p000_p000\gfs_cov_m006_p000
\gfs_cov_m004_p000
\gfs_cov_m002_p000
\gfs_cov_p002_p000
\gfs_cov_p004_p000\gfs_cov_p006_p000
All tree trunks are perfectly vertical – not a good approximation.
July 29, 2011 IGARSS 2011, Vancouver 13
The orientation angle basically matches changes in azimuth slopes.
Azimuth Slopes in Grassy RegionsOrientation Angle & Correlation |ρ4|
Low correlation values imply poor estimates of orientation angles.
0 0.2 0.4 0.6 0.8 10
200
400
600
800
1000
1200
Ort Dispersion (ρ4)
\s_cov_p000_p000
\s_cov_p000_p002
\s_cov_p000_p004\s_cov_p000_p006
\s_cov_p000_p008
\s_cov_p000_p010
-40 -20 0 20 400
200
400
600
800
1000
Orientation (deg)
\s_cov_p000_p000
\s_cov_p000_p002
\s_cov_p000_p004\s_cov_p000_p006
\s_cov_p000_p008
\s_cov_p000_p010
July 29, 2011 IGARSS 2011, Vancouver 14
-30 -25 -20 -15 -10 -5 0 50
200
400
600
800
1000
1200
1400
1600
Double Bounce Power (dB)
\s_cov_p000_p000
\s_cov_p000_p002
\s_cov_p000_p004\s_cov_p000_p006
\s_cov_p000_p008
\s_cov_p000_p010
Grass Region Forest
Small azimuth tilts allow ground-trunk scattering, only larger slopes reduce double bounce power in forest.
Azimuth Slopes in Forest AreaDouble Bounce Power
Ground-Forest Component
The ground-forest multiple scattering component accounts for double bounce changes with azimuth slope.
Tree trunk forward scattering is not limited to specular directions.
-30 -25 -20 -15 -10 -5 0 50
200
400
600
800
1000
1200
Double Bounce Power (dB)
\gfs_cov_p000_p000\gfs_cov_p000_p002
\gfs_cov_p000_p004
\gfs_cov_p000_p006
\gfs_cov_p000_p008\gfs_cov_p000_p010
July 29, 2011 IGARSS 2011, Vancouver 15
The orientation angle basically matches changes in azimuth slopes.
Azimuth Slopes in Forest AreaOrientation Angle & Correlation |ρ4|
Higher correlation values in forest area imply good estimates of the orientation angles.
0 0.2 0.4 0.6 0.8 10
500
1000
1500
2000
Ort Dispersion (ρ4)
\s_cov_p000_p000\s_cov_p000_p002
\s_cov_p000_p004
\s_cov_p000_p006
\s_cov_p000_p008\s_cov_p000_p010
-20 -15 -10 -5 0 50
200
400
600
800
1000
1200
1400
Orientation (deg)
\s_cov_p000_p000\s_cov_p000_p002
\s_cov_p000_p004
\s_cov_p000_p006
\s_cov_p000_p008\s_cov_p000_p010
July 29, 2011 IGARSS 2011, Vancouver 16
Full SAR Simulation
Azimuth Slopes in Forest AreaOrientation Angle
Orientation angle estimates from full SAR simulation closely matches the orientation angles of the strong ground-forest scattering component.
The ground-forest induced orientation angle shows low variance (high correlations) independent of azimuth slope.
The orientation angle variance from the full simulation increases with increases in the azimuth slope.
-20 -15 -10 -5 0 50
200
400
600
800
1000
1200
1400
Orientation (deg)
\s_cov_p000_p000\s_cov_p000_p002
\s_cov_p000_p004
\s_cov_p000_p006
\s_cov_p000_p008\s_cov_p000_p010
-20 -15 -10 -5 0 50
500
1000
1500
2000
Orientation (deg)
\gfs_cov_p000_p000\gfs_cov_p000_p002
\gfs_cov_p000_p004
\gfs_cov_p000_p006
\gfs_cov_p000_p008\gfs_cov_p000_p010
Ground-Forest Component
July 29, 2011 IGARSS 2011, Vancouver 17
Full SAR Simulation
Azimuth Slopes in Forest AreaScatter Shape Variation
The Shape Variation reflects inherent changes to the polarimetric scattering.
Even after orientation angle compensation, once the azimuth slopes exceed ~6°, the ground-forest scattering component changes character.
The double bounce scattering, probably generated from ground-trunk interactions, is greatly reduced.
This change in Shape Variation is consistent with the α-angle changes due to azimuth slopes shown earlier.
Ground-Forest Component
-1 -0.8 -0.6 -0.40
500
1000
1500
2000
2500
3000
3500
Shape Variation (ρab)
\gfs_cov_p000_p000
\gfs_cov_p000_p002
\gfs_cov_p000_p004\gfs_cov_p000_p006
\gfs_cov_p000_p008
\gfs_cov_p000_p010
-1 -0.5 0 0.5 10
200
400
600
800
1000
Shape Variation (ρab)
\s_cov_p000_p000
\s_cov_p000_p002
\s_cov_p000_p004\s_cov_p000_p006
\s_cov_p000_p008
\s_cov_p000_p010
July 29, 2011 IGARSS 2011, Vancouver 18
Grassy Region
Major Forest Scattering Components
Forest scattering, at least for our simulated Scots Pine forest, is almost completely specified by the Direct Forest (blue) and the Ground-Forest (green) scattering components.
The cyan curve, the combination of Direct Forest and Ground-Forest scattering, almost exactly matches the Full Scattering curve (red).
Forest Returns
-30 -25 -20 -15 -10 -5 0 50
500
1000
1500
Double Bounce Power (dB)
.\dfs_cov_p000_p000.tla
.\gfs_cov_p000_p000.tla
.\s_cov_p000_p000.tla
Mixed
July 29, 2011 IGARSS 2011, Vancouver 19
Major Forest Scattering Components
The Full Orientation Angle Correlation matches the Correlation formed from the Direct Forest and Ground-Forest scattering components.
Similarly, the Full Shape Variation reflects the combination of the Direct Forest and Ground-Forest scattering.
0 0.2 0.4 0.6 0.8 10
1000
2000
3000
4000
5000
6000
7000
Ort Dispersion (ρ4)
.\dfs_cov_p000_p000.tla
.\gfs_cov_p000_p000.tla
.\s_cov_p000_p000.tla
Mixed
-1 -0.5 0 0.5 10
500
1000
1500
2000
2500
3000
Shape Variation (ρab)
.\dfs_cov_p000_p000.tla
.\gfs_cov_p000_p000.tla
.\s_cov_p000_p000.tla
Mixed
July 29, 2011 IGARSS 2011, Vancouver 20
Preliminary Conclusions (1)
• Results Based on a Single Forest Simulation
– Relatively light canopy attenuation
– Sparse forest – typical of Scots Pine
• Known Defects in the Tree Realizations
– All tree trunks aligned perfectly vertical
– One can design trees with randomly angled trunks
• Only Investigated Polarimetric Variations with Ground Slope
– Wider variety of forest parameters needed
– Adjust relative weighting of canopy attenuation, direct ground scattering, tree architecture, etc.
• Simulations May Produce Artificial Artifacts Not Seen in Scenery
– Blind use or application of software may be misleading
July 29, 2011 IGARSS 2011, Vancouver 21
Preliminary Conclusions (2)• Both Range and Azimuth Slopes Reduce Double Bounce Ground-
Trunk Scattering in this Forest– Both orientation angle correlation and shape variation indicate inherent
changes in polarimetric scattering due to topographic slopes
• Orientation Angle Compensation Does Not Remove All Topographic Effects on SAR Polarimetry
• The Large Number of Forest Parameters May Limit Full Investigation – Tree architecture, relative ground / canopy scattering strength, and
canopy attenuation may be the important variables
July 29, 2011 IGARSS 2011, Vancouver 22
Background
July 29, 2011 IGARSS 2011, Vancouver 23
Volume of Spheroid Scatterers• Assuming a cloud of spheroids for volumetric
canopy scatterers, characterized by independent parameters: size, shape and orientation.
O
X
Y
Z
N)cos,sinsin,cos(sin θφθφθ
φ
θiθ
)cos,0,(sin ii θθ)0,1,0(
)sin,0,cos( ii θθ−
Each elementary scatterer features a body of revolution w.r.t. the symmetric axis, ON.
( )( )
−
− ββ
ββψ
ψββββ
cossin
sincos
0
0
cossin
sincos
b
a
S
S
The projected symmetry axis on the polarization plane is oriented towards angle β;
The local incidence angle w.r.t. the symmetry axis is ψ;
The principal scattering components are Sa and Sb.
July 29, 2011 IGARSS 2011, Vancouver 24
Circular Polarization Representation• Oriented targets can be clearly expressed in the
circular polarization basis– Mean orientation angle from the phase of RR-LL– Similar to circularly polarized weather radar analysis
222
4
1baLLRR SSSS −==
222
4
1baLRRL SSSS +==
β42*
4
1 jbaLLRR eSSSS −−=
( ) ( ) β2**
4
1 jbabaRLRR eSSSSSS −+−=
( ) ( ) β2**
4
1 jbabaRLLL eSSSSSS +−=
If β and ψ are separable,
Advantage: Orientation parameters readily separable as phase terms.
( ) 044
44 4cos βββ ρββ jjj eee −−− ≡−=
*22
*22
2
2
Re2
Re2
baba
baba
ba
ba
SSSS
SSSS
SS
SSCDR
++
−+=
+
−=
022
2
22
*22Im2
ββ ρρρ jr
j
baba
baba
x eeSSSS
SSjSS−− ≡
+−
+−=
Symmetric distribution Orientation
dispersionMean orientation angle
Independent of orientation 0: sphere 1: dipole
Relates to both shape variation and orientation dispersion
July 29, 2011 IGARSS 2011, Vancouver 25
Orientation Distribution• The orientation angle is close to a von Mises
distribution
0,0
cos
0 )(2
1)(
===
βθβκ
κπβ e
If
Established a relation between ρ2 and ρ4.The orientation parameters, mean and dispersion, can be determined and removed, which leaves only scatterer shape information.
July 29, 2011 IGARSS 2011, Vancouver 26
Physical Parameters Retrieval• It is then straightforward to solve the principal
components from orientation compensated data
• Theoretically, the solution works for a single volume scattering mechanism and homogenous targets, resolving parameters:
( )*222Re2 RLRRRRRLa SSSSS ++=
( )*222Re2 RLRRRRRLb SSSSS −+=
( )*22* Im2 RLRRRRRLba SSjSSSS +−=
( )22
ba
*ba
ab
SS
SSeℜ=ρ
2
2
b
a
S
Sr =
We get mean shape and shape variation.
Size
Shape
Orientation