Tampa BSides - Chef's Tour of Microsoft Security Adoption Framework (SAF)
Tomo_20.pdf
1. P-Band Penetration in Tropical and
Boreal Forests: Tomographical Results
Stefano Tebaldini, Mauro Mariotti d’Alessandro, Ho Tong Minh Dinh, Fabio Rocca
Politecnico di Milano
Dipartimento di Elettronica e Informazione
2. Introduction
Longer wavelength SARs precious tool for forestry remote sensing
• Under foliage penetration capabilities
• Mitigate saturation in backscatter vs forest biomass law
From Le Toan et al., 2004
3. Introduction
Longer wavelength SARs precious tool for forestry remote sensing
• Under foliage penetration capabilities
• Mitigate saturation in backscatter vs forest biomass law
Sensitivity to the whole forest structure many different scattering mechanisms
• Back scatter from the canopy
• Back scatter from the ground (Bragg)
• Trunk-Ground forward scatter
• Canopy-Ground forward scatter
Signal interpretation requires physical models
• One passage coherent or incoherent polarimetric decomposition
• Two passages PolInSAR (i.e: RVoG)
4. Introduction
Multi-baseline SAR Tomography Direct imaging of the forest vertical structure
Tomogram - HH
Track n cross 60
range 50
40
Height [m]
Reference
Track (Master) 30
π/2 20
10
Track 1 0
θ -10
slant 400 600 800 1000 1200 1400
range Slant range [m]
elevation
Tomography is a fundamental tool to:
• investigate the phenomenology of Radar scattering from
forested areas
• help physical modeling to be used with non
interferometric data and single baseline data
5. Introduction
Multi-baseline SAR Tomography Direct imaging of the forest vertical structure
Tomogram - HH
Track n cross 60
range 50
40
Height [m]
Reference
Track (Master) 30
π/2 20
10
Track 1 0
θ The BIOMASS Tomographic phase
-10
slant 400 600 800 1000 1200 1400
range Features: Slant range [m]
elevation • 55 days (3% of mission lifetime)
• ≤ 4 day repeat pass time
Main goal:
Help improve forest biomass and height retrieval methods
by addressing three questions:
• What are the main scattering mechanisms (SMs) at forest
and ground level
• How do the SMs vary as a function of polarization
• How do the SMs vary over the global forest biomes
6. Investigated sites
BIOSAR 2007
Site Remningstorp, Southern Sweden
Period Spring 2007
Scene Semi-boreal forest
Topography Flat
Carrier frequency P-Band
Vertical resolution 10 m (near range) to 40 m (far range)
BIOSAR 2008
Site Krycklan, Northern Sweden
Period Fall 2008
Scene Boreal forest
Topography Hilly
Carrier frequency P-Band and L-Band
Vertical resolution P-Band: 20 m (near range) to 80 m (far range)
L-Band: 6 m (near range) to 25 m (far range)
TROPISAR – data courtesy of ONERA
Site Paracou, French Guyana
Period August 2009
Scene Tropical forest
Carrier frequency P-Band
Vertical resolution ≈15 m
7. Investigated sites
BIOSAR 2007
Site Remningstorp, Southern Sweden
Period Spring 2007
Scene
BIOSAR 2007,forest
Semi-boreal
BIOSAR 2008: Vertical resolution ≥ forest height
Topography Tomographic imaging: Capon spectrum
Flat
Carrier frequency • Greatly
P-Band enhances vertical resolution
Vertical resolution Requires multilooking
10•m (near range) to 40 m (far range) horizontal resolution loss
• Not radiometrically accurate BIOSAR 2008
Quantitative measurements by assuming ground + volume scattering
Site Krycklan, Northern Sweden
• Parametric models Period Fall 2008
Scene Boreal forest
• Algebraic Synthesis
Topography Hilly
Carrier frequency P-Band and L-Band
Vertical resolution P-Band: 20 m (near range) to 80 m (far range)
TROPISAR: Vertical resolution < forest height L-Band: 6 m (near range) to 25 m (far range)
Tomographic imaging: coherent focusing at pixel level
TROPISAR
Site
• no need for multilooking
Paracou, French Guyana
Period
• model free
August 2009
Scene • radiometrically
Tropical forest accurate
Carrier frequency P-Band
Vertical resolution ≈15 m
8. Results from BIOSAR 2007
Campaign BioSAR 2007 - ESA
System E-SAR - DLR
Period Spring 2007
Site Remningstorp, South Sweden
Scene Semi-boreal forest
Norway spruce, Scots pine, Birch
Topography Flat
Tomographic 9 – Fully Polarimetric
tracks
Carrier 350 MHz
frequency
Slant range 2m
resolution
Azimuth 1.6 m
resolution
Vertical 10 m (near range) to 40 m
resolution (far range)
9. BIOSAR 2007
HHVV phase
HHVV coherence +180°
2
slant range [Km]
• Phase: 1.6
0 1.2
Forest: φHH - φVV ≈ 80° 0.8
Open areas: φHH - φVV ≈ 0° 0.4
-180°
1 2 3 4 5
HHVV coherence amplitude
1
slant range [Km]
2
• Amplitude: 0.8
1.6
Forest: |γHHVV | ≈ 0.45 0.6
1.2
0.4
Open areas: |γHHVV | ≈ 0.8 0.2
0.8
0.4
0
1 2 3 4 5
Mean reflectivity - HH
Amplitude Stability Analysis 2
slant range [Km]
• Presence of a high number of 1.6
amplitude stable points in the 1.2
co-polar channels 0.8
0.4
1 2 3 4 5
azimuth [Km]
10. BIOSAR 2007 – Tomographic profiles
Tomographic reconstruction
of an azimuth cut: Reflectivity (HH) – Average on 9 tracks
50
azimuth [m]
Reflectivity (HH) – Average on 9 tracks 40
30
20
10
slant range
200 600 1000 1400 1800 2200
Capon Spectrum - HH
60
azimuth 50
height [m]
The analyzed profile is almost totally forested, 40
30
except for the dark areas 20
10
0
HH: -10
200 600 1000 1400 1800 2200
Dominant phase center is ground locked
Vegetation is barely visible Capon Spectrum - HV
60
LIDAR Terrain Height
50
LIDAR Forest Height
height [m]
40
Similar conclusions for VV 30
20
10
HV: 0
-10
Dominant phase center is ground locked 200 600 1000 1400 1800 2200
Vegetation is much more visible slant range [m]
11. BIOSAR 2007 – Tomographic profiles
Tomographic reconstruction
Physical interpretation:
of an azimuth cut: Reflectivity (HH) – Average on 9 tracks
• Scattering from ground level is determined by an imperfect dihedral
50
azimuth [m]
Reflectivity (HH) – Average on 9 tracks 40
contribution from ground-trunk interactions, perturbed by understory and
30
20
topography oscillations 10
slant range
Possible presence of canopy-ground interactions600
200 1000 1400 1800 2200
• Scattering from above the ground, due to canopy backscattering, HH
60
Capon Spectrum - is
extremely weak
azimuth 50
height [m]
The analyzed profile is almost totally forested, 40
30
except for the dark areas 20
10
0
HH: -10
200 600 1000 1400 1800 2200
Dominant phase center is ground locked
Vegetation is barely visible Capon Spectrum - HV
60
LIDAR Terrain Height
50
LIDAR Forest Height
height [m]
40
Similar conclusions for VV 30
20
10
HV: 0
-10
Dominant phase center is ground locked 200 600 1000 1400 1800 2200
Vegetation is much more visible slant range [m]
12. Results from BIOSAR 2008
Campaign BioSAR 2008 - ESA
System E-SAR - DLR
Site Krycklan river catchment,
Northern Sweden
Scene Boreal forest
Pine, Spruce, Birch, Mixed stand
Topography Hilly
Tomographic 6 + 6 – Fully Polarimetric
Tracks (South-West and North-East)
Carrier P-Band and L-Band
Frequency
Slant range 1.5 m
resolution
Azimuth 1.6 m
resolution
Vertical resolution 20 m (near range) to >80 m (far range)
(P-Band)
Vertical resolution 6 m (near range) to 25 m (far range)
(L-Band)
13. BIOSAR 2008 – Tomographic profiles
Tomographic reconstruction of P-Band SW - HV
30
an azimuth cut:
Height [m]
20
Polarization: HV
10
Method: Capon Spectrum
0
• Results are geocoded onto the same ground
range, height grid -10
2000 2500 3000 3500 4000 4500 5000
P-Band NE - HV
• All panels have been re-interpolated such that 30
the ground level corresponds to 0 m
Height [m]
20
• Loss of resolution from near to far range, 10
especially at P-Band (Δz > 80 m at far ranges) 0
-10
• Relevant contributions from the ground level 5000 4500 4000 3500 3000 2500 2000
below the forest are found at P-Band L-Band SW - HV
30
30
LIDAR DEM
250
Height [m]
20
20
Height [m]
Height [m]
10
10
200
00
-10
-10 2000
2000 2500
2500 3000
3000 3500
3500 4000
4000 4500
4500 5000
5000
150 Ground range [m]
2000 2500 3000 3500 4000 4500 5000 Ground range [m]
Ground range [m]
14. BIOSAR 2008 – ground/volume decomposition
Ground to Volume Ratio:
P-Band SW P-Band SW
Ratio between the HH HV
backscattered powers
associated with ground-only
and volume-contributions 15 15
0 0
-15 -15
L-Band SW L-Band SW
HH HV
15 15
0 0
-15 -15
15. BIOSAR 2008 – ground/volume decomposition
HV GVR vs. LIDAR H100 HV GVR vs. Terrain slope
P-Band SW P-Band SW
• At both wavelengths it is 15
15 15
observed that the HV GVR 10
10 10
HV GVR [dB]
HV GVR [dB]
decreases with forest height,
HV GVR [dB]
55 5
consistently with the 00 0
enlargement of volumetric -5
-5 -5
structures. -10
-10 -10
-15
-15 -15
• HV GVR exhibits a 10 10 15 15 20 20 25 25
Forest Height [m]
30 30 0 5 10 15
dependence on terrain slope L-Band SW L-Band SW
15
15 15
at P-Band but not at L-Band
10
10 10
This result indicates that HV
HV GVR [dB]
HV GVR [dB]
HV GVR [dB]
55 5
ground contributions are
00 0
due to double bounce
-5
-5 -5
contributions at P-Band, but
-10
-10 -10
not at L-Band
-15
-15 -15
1010 1515 2020 2525 3030 0 5 10 15
Forest Height [m] Absolute Ground Slope [deg]
LIDAR [m]
16. Results from TropiSAR
Campaign TropiSAR- ESA
data courtesy of ONERA
System Sethi- ONERA
Period August 2009
Site (among Paracou, French Guyana
others)
Scene Tropical forest
estimated 150 species per hectare
Dominant families:
Lecythidaceae, Leguminoseae,
Chrysobalanaceae, Euphorbiaceae.
Tomographic 6 – Fully Polarimetric
tracks
Carrier P-Band
frequency
Slant range ≈1 m
resolution
Azimuth ≈1 m
resolution
Vertical 15 m
resolution
17. Processing of TropiSAR
Goal: generation of a stack of multi-layer SLC SAR images out of a stack of multi-baseline
SLC SAR images
height
Tomographic
Processor
Slant range
azimuth
Layer N
SAR Tomography
resolution cell
SAR resolution Layer 1
cell
18. TROPISAR – Tomographic profiles
Tomographic reconstruction of two azimuth cuts:
Polarization = HH - azimuth bin = 455
60
Method: coherent focusing
Height [m]
40
20
All panels have been re-interpolated 0
such that the ground level corresponds
400 600 800 1000 1200 1400
to 0 m Polarization = HV - azimuth bin = 455
60
Height [m]
40
20
HH 0
Visible contribution from the 400 600 800 1000 1200 1400
ground level beneath the forest Slant range [m]
Polarization = HH - azimuth bin = 1455
60
Vegetation is well visible
Height [m]
40
20
0
HV
400 600 800 1000 1200 1400
Poor contributions from the Polarization = HV - azimuth bin = 1455
ground level beneath the forest 60
Height [m]
40
20
Vegetation is well visible
0
400 600 800 1000 1200 1400
Slant range [m]
19. TROPISAR – Tomographic sections
Tomographic reconstruction of radar scattering from four
different heights
Ground level Ground level + 10 m
Method: coherent focusing 20 20
15 15
Polarization: HH
Slant range
Slant range
10 10
5 5
• The strongest dependence on 0 0
terrain topograpy is found at the
-5 -5
ground level
• The most uniform tomographic Azimuth
-10
Azimuth
-10
layer is found at about15-20 m
above the ground Ground level + 20 m Ground level + 35 m
20 20
• Highest layers exhibit a
15 15
dependence on terrain topography,
Slant range
Slant range
similarly to the ground layer 10 10
5 5
0 0
-5 -5
Tomographic data exhibit a more
-10 -10
complex dependence of terrain Azimuth Azimuth
topography than traditional SAR data.
21. Dependence on Topography
A closer look…
This resolution cell gathers contributions from terrain only.
=> Signal intensity in this cell is affected by terrain slope the
same way as in traditional SAR images of bare surfaces
22. Dependence on Topography
A closer look…
This cell is completely within the volume layer,
independently on volume orientation w.r.t. the Radar LOS.
=> Signal intensity in this cell is independent of terrain
slope
This resolution cell gathers contributions from terrain only.
=> Signal intensity in this cell is affected by terrain slope the
same way as in traditional SAR images of bare surfaces
23. Dependence on Topography
A closer look…
The scattering volume within cells at the boundaries of the
vegetation layer depends on volume orientation w.r.t. the
Radar LOS.
=> Signal intensity in this cell is affected by terrain slope
in a similar way as the cell corresponding to the ground
layer.
This cell is completely within the volume layer,
independently on volume orientation w.r.t. the Radar LOS.
=> Signal intensity in this cell is independent of terrain
slope
This resolution cell gathers contributions from terrain only.
=> Signal intensity in this cell is affected by terrain slope the
same way as in traditional SAR images of bare surfaces
32. Tomography @ BIOMASS resolution
Resolution Loss Factor w.r.t. E-SAR = 100/6 •12.5/1.6 > 100 !
•At 30° a 60 x 60 estimation window contains just 5 independent looks ! less robust
statistics
• Slant range resolution loss further causes a spread of the of the backscattered power
distribution, resulting in a vertical resolution loss
E-SAR - HV
Theoretical vertical resolution limit due to pulse 30
bandwidth is ≈ 10 m at θ = 30°
20
Height [m]
10
• Nevertheless, Tomographic profiles
0
provide information about the forest
structure that is consistent with the -10
2000 2500 3000 3500 4000 4500 5000
airborne case BioMass – HV
30
30
20
20
Height [m]
BioMass data-set derived by
Elevation [m]
DLR from BIOSAR 2008 10
10
Pulse Bandwidth = 6 MHz
00
Azimuth resolution = 12.5 m
-10
-10 2000
2000 2500
2500 3000
3000 3500
3500 4000
4000 –4500 5000
LIDAR TOP
4500 5000
Ground range [m] HEIGHT
33. BioMass: Forest Height Retrieval
Forest height has been retrieved
through a direct investigation of the Forest Relative
shape of the retrieved tomographic height error
profiles
Rising trend due to the very large 30 1
variation of baseline aperture resulting
from flight geometry 15 0.5
0 0
Good match with LIDAR
• Standard Deviation < 4 m w.r.t.
2D Histogram Normalized 2D Histogram
LIDAR by exploiting a 1 hectare 30
30 30
30
BioMass Forest Height [m]
BioMass Forest Height [m]
estimation window 25
25 25
25
• No significant bias beyond 10 m 20
20 20
20
SAR [m]
SAR [m]
15
15 15
15
Estimation loses reliability for forest 10
10 10
10
lower than 10 m, consistently with the 55 55
theoretical resolution limit 0
0 00
0
0 5
5 10
10 15 15 20
20 25
25 30
30 00 5
5 10
10 15 15 20
20 25
25 30
30
LIDAR [m]
LIDAR [m] LIDAR [m]
LIDAR [m]
34. Conclusions
Tomography is highly sensitive to forest structure:
• Double bounce contributions from ground-trunk interactions have clearly been observed at the
Paracou site, despite the presence of a tropical forest 40 m high
• Boreal and semi-boreal forest have shown an almost ground-locked vertical structure in both in
co and cross polarization, suggesting specular reflections play a non negligible role at P-Band
Different tomographic layers connect differently to forest biomass
• Best correlation factor observed at 30 m in HV (R = .82 @ 125 m , R = .93 @ 250 m)
• Preliminary biomass inversion results are very encouraging. Final assessment needs accurate
comparison to existing inversion techniques (Intensity, Intensity + PolInSAR height, LIDAR)
Forest imaging @ BIOMASS resolution is a challenging problem.
Measurements from BIOSAR 2007 and BIOSAR 2008 show that:
• Tomographic imaging consistent with the airborne case
• Forest height retrieved within an accuracy of 20% with a 1 ha spatial resolution
• No significant bias observed for forests higher than 10 m, consistently with the theoretical limit
Assessment of tomography capabilities @ BIOMASS resolution in tropical forests is yet to
be done