测绘遥感信息工程国家重点实验室

State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing

Synthetic Aperture Radar Tomography –

practical course
Timo Balz, Stefano Tebaldini, Laurent Ferro-Famil
 TomoSAR_Main.m
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%
% DEMONSTRATIVE TOMOGRAPHIC SAR PROCESSING FOR FOREST ANALYSIS
% AUTHOR: STEFANO TEBALDINI, POLITECNICO DI MILANO
% EMAIL: [email protected]
% TEL: +390223993614
%
% THE FOLLOWING SCRIPT AND ALL RELATED SCRIPTS/FUNCTIONS AND DATA ARE INTENDED AS
% MATERIAL FOR THE TOMOSAR TRAINING COURSE HELD IN BEIJING IN FEBRUARY 2015
% BY LAURENT FERRO-FAMIL AND STEFANO TEBALDINI
%
% THIS SOFTWARE WAS DEVELOPED AND TESTED USING MATLAB R2011b
%
% SAR DATA USED IN THIS SCRIPT ARE PART OF THE SAR DATA-SET ACQUIRED BY DLR
% IN 2008 IN THE FRAME OF THE ESA CAMPAIGN BIOSAR 2008
% DATA FOCUSING, COREGISTRATION, PHASE FLATTENING, AND GENERATION OF KZ
% MAPS WERE CARRIED OUT BY DLR.
% DATA PHASE CALIBRATION WAS CARRIED OUT BY THE AUTHOR
%
% TERRAIN ELEVATION DATA USED IN THIS SCRIPT ARE EXTRACTED FROM
% THE LIDAR DATA-SET ACQUIRED BY THE SWEDISH DEFENCE RESEARCH AGENCY (FOI)
% AND HILDUR AND SVEN WINQUIST'S FOUNDATION IN THE FRAME OF THE ESA
% CAMPAIGN BIOSAR 2008
% PROCESSING OF LIDAR DATA AND PROJECTION ONTO SAR GEOMETRY WAS CARRIED OUT
% BY THE AUTHOR
%
% YOU ARE WELCOME TO ADDRESS ME QUESTIONS/COMMENTS/CORRECTIONS AT
% [email protected]
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%
Ambiguities
TomoSAR principle
TomoSAR principle
TomoSAR principle
TomoSAR principle
TomoSAR principle
 • Why multiple baselines?
• Because: more equations!
• Increased robustness against disturbances (temporal
decorrelation…)
• and/or relaxation of hypotheses required in the single
baseline case
• more unknowns are available to characterize the vertical
structure of the scene

N=2 z N=3 z
N is large z

Top Height, Extinction

Mean, Std, Skewness
… Tomographic Reconstruction

Backscattered Power Backscattered Power Backscattered Power

MB allow to pass from model-based inversion to full tomographic reconstruction

Tebaldini & Rocca

 Rationale: Form a 2D synthetic aperture by collecting multiple SAR acquisitions
acquired along parallel flight lines
o Vertical resolution  total normal baseline span
o Vertical ambiguity  normal baseline spacing

Multi-baseline (MB) systems:

• Multiple pass systems:
Slant range
resolution cell
airborne and spaceborne SARs
• Multiple antenna systems:
ground based Radars
θ
MB campaigns involve:
Replicas
• Higher costs:
height

Height of ambiguity spaceborne: ≈ x 1

ground based: ≈ x N
• More sophisticated processing:
see single vs multi-baseline InSAR…

ground range
azimuth
Source:
Tebaldini & Rocca
 Tomographic SAR
Rationale: Form a 2D synthetic aperture by collecting multiple SAR acquisitions acquired along parallel flight
lines
o Vertical resolution  total normal baseline span
o Vertical ambiguity  normal baseline spacing

Multi-baseline (MB) systems:

Slant range
resolution cell
• Multiple pass systems:
airborne and spaceborne SARs
• Multiple antenna systems:
θ ground based Radars
Replicas
MB campaigns involve:
height

Height of • Higher costs:

ambiguity
spaceborne: ≈ x 1
ground based: ≈ x N
• More sophisticated processing:
see single vs multi-baseline InSAR…

ground range
azimuth

Source: Tebaldini & Rocca

 Basic concepts
Multiple baselines  Illumination from multiple points of view

Track n Track n
cross range
cross range baseline vector for the
Reference Track n-th track
(Master)
π/2 Master azimuth

Normal baseline slant range

for the n-th track
Track 1
θ
slant range Flight tracks orthogonal
to the blackboard

height By collecting several baselines it is possible to synthesize an

antenna along the cross range direction as well

3D focusing is possible in the coordinate

system: slant range, azimuth, cross range

Source: Tebaldini & Rocca

 Basic concepts
Resolution is determined by pulse bandwidth along the slant range direction, and by the lengths of the synthetic
apertures in the azimuth and cross range directions
The SAR resolution cell is split into multiple layers, according to baseline aperture

Track N
c r r B: pulse bandwidth
r  v  x  Av: baseline aperture
2B 2 Av 2 Ax Ax: azimuth aperture
Track n
SAR Resolution Cell λ: carrier wavelength
Track 1 For most systems:
Tomographic Res. Cell
Δv >> Δr, Δx

height
height

 z   v  sin  
Δv ground range

ground range
azimuth Δr
Source: Tebaldini & Rocca
 TomoSAR gives access to the 3D structure

60
50
Forested areas Ice
40
height [m]

30
20
10
0
-10
200 600 1000 1400 1800 2200
slant range [m]

0.5
0 Snow
-0.5
z [m]

-1
-1.5
-2 0 1 2 3 4 5 6 7
y [m]

Man-made structures

Urban scenarios

Source: Tebaldini & Rocca

 Polarimetric TomoSAR over Vegetation

Foliage penetration Biomass estimation

 Urban TomoSAR

Finite number of scatterers

 TomoSAR principle

Orbit uncertainties &

tropospheric delay

Irregular sampling & = low resolution in s

Limited acquisitions
TomoSAR as spectral estimation problem

where

∥
so, gn is the Fourier transformation of γ′(s) at position

∥
 TomoSAR basics

The focused SAR image from the nth pass of a specific cell is nothing else but the
Fourier Transfer of the reflectivity function in the elevation direction at the position
fn.
 TomoSAR basics

The expected resolution in elevation depends on the slant-range r and the aperture
size in elevation

The extent of the illuminated objects and therefore the limits of the extent in
elevation s depends on r, , and the resolution in range . In the spaceborne
case, with large slant-range r, this is seldom a limitation (see Zhu & Bamler, 2010).
 PS-InSAR for pre-processing

• Using TomoSAR on a large area requires the removal of

atmospheric effects

• This can be done using PS-InSAR

• However, this requires a large number of PS to be found
• Therefore, this works best in urban areas

• PS-InSAR is used for pre-processing

• Afterwards, TomoSAR can be used

PS-InSAR
 PS-InSAR steps
• Import
• Selection of the master image
• Co-Registration
• Typically to a single-master
• Interferogram processing
• Not in every implementation. Several PS-InSAR implementations only
use the phases of the PS candidates
• PS candidates selection
• APS estimation
• Typically on a subset of the PS candidates
• PS point processing
• Post-processing
• Visualization
 Image co-registration

• Identical to the co-registration described in the InSAR section:

• Registration accuracy < 0.2 pixel is required

• This requirement is even much higher during TOPS processing

• Different methods available

• Based on the amplitude
• Based on complex data – searching maximal coherence
•
Slave images are resampled to the master image

• Results need to be checked and the parameters may need to be

adjusted
 PS candidate selection

• Classical way: amplitude dispersion index

• Other possible ways, e.g. based on coherence

• Be aware: amplitude dispersion index not ideal for

TomoSAR, because the existence of several scatterers in a
resolution cell can increase the index
APS estimation

Example: SARProZ
 Spectral estimators

• Beamforming:
inverse Fourier Transform; coarse spatial resolution; radiometrically consistent
T
  4   4   4 
Sˆ v   a H v Rˆ a v  a v    exp  j b1v  exp  j b2 v   exp  j bN v  
  r   r   r 

• Capon Spectral Estimator: 1

Sˆ v  
spatial resolution is greatly enhanced, at the expense of radiometric accuracy; a v Rˆ 1a v 
H

• Methods based on the analysis of the Eigenstructure of R (MUSIC, ESPRIT…):

determination of the dominant scatterering centers; mostly suited for urban scenarios

• Methods based on sectorial information (Truncated SVD, PCT…):

optimal basis choice (e.g.: Legendre), depending on a-priori info about the scene vertical extent

• Model based methods (NLS, COMET…):

model based; high radiometric accuracy; high computational burden; possible model mismatches
• Compressive sensing:
localization of few scattering centers via L1 norm minimization; mostly suited for urban scenarios
 Multiple scatterers

where
gn = complex value observed for the nth pass
𝛾 = complex amplitude of kth scatterer
𝑠 = elevation of kth scatterer
𝑛 + 1= number of available images
𝑛 =number of scatterers inside a resolution cell
𝑣 =noise
𝑓 =frequency of sampled FT which depends on the baseline
Non-linear least-square estimation

( )×
SVD on real data
 Model selection

• Selecting a statistical model for given data

too complex –
too simple –
appropriate model order too high
order too low

• Selecting the correct model is the model selection problem

 Model selection methods

• Bayesian Information Criterion (BIC)

𝑘
𝑘 = arg max ln 𝑝 𝑦 𝜃 𝑘 , 𝑘 − ln 𝑛
2

• Akaike Information Criterion (AIC)

• Minimum Description Length (MDL)

TomoSAR processing steps
3D scatterer reconstruction
 Urban TomoSAR different methods

• SVD based methods

• SVD
• Truncated SVD (T-SVD)
• Wiener SVD
• Butterworth-SVD
• Compressive Sensing based methods
• Basis Pursuit (BP)
• TWIST
 TWIST

Another solution to SAR tomography is by L1 norm minimization, which is also the core of
compressive sensing:
g  K   

The argument listed above can be stated by

k
where is a weighted value adjusted according to the noise level.
 TWIST

• SAR tomography with compressive sensing

• (see Zhu, Xiaoxiang’s work in the references)

• Often, TomoSAR with compressive sensing is based on Basis

Pursuit (BP)
• Very high super-resolution
• However, time-consuming

• Alternatively: Two-Step Iterative Shrinkage Thresholding

(TWIST) for TomoSAR
• Very efficient
• Less super-resolution capability
 TWIST

When using TWIST, the least squares fitting is needed to calculate  0

Calculate  1 according to

where  is the normalization equation.

If t 1

in which, ,  are the weighting coefficients

TomoSAR: different methods

(a) s_Sc1=-20m, (b) s_Sc1=8m, (c) s_Sc1=23m;

s_Sc2=30m; s_Sc2=30m; s_Sc2=30m
 TSVD, Twist & BP

TSVD TWIST BP
 TSVD, Twist & BP
柏林实验区
（a）TSVD
85 seconds

red: single;
（b）TWIST blue: double;
89 seconds
black: multi

（c）BP
1378
seconds
 Differential TomoSAR
• Extend the model to the time-domain and include
an estimation of the deformation
• 4D TomoSAR

• Basically just an extension of the previously

described method
• Including an additional dimension for the focusing

• Can be further extended by including seasonal

motion
• Sometimes called 5D TomoSAR
 D-TomoSAR

Simulation results:

BSVD TWIST

= for D-TomoSAR, the noise suppression gets even more important

 D-TomoSAR

Example: TSX stack from Las Vegas:

BSVD TWIST
= clear result in TWIST
 Practical implementations

Due to the high computational demand for

compressive sensing TomoSAR, the processing can
be divided:

1. PS-InSAR for pre-processing – APS estimation

2. Basic TomoSAR processing for model estimation
3. Depending on the number of scatterers:
• One scatterer per resolution cell: use PS-InSAR for
processing
• Two or more scatterer: use TomoSAR
 Geodetic SAR tomography

• Fusion of SAR imaging geodesy and TomoSAR

• SAR imaging geodesy:

• Very high absolute geo-positioning capability of SAR
• Especially with TerraSAR-X due to the very precise orbit
• see Eineder et al, Cong et al, Balss et al….

• The fusion allows getting precise absolute 3D

positions
 Geodetic SAR tomography

from Zhu et al, 2016 – 3D absolute positioned

TomoSAR point cloud from Berlin

from Zhu et al, 2016 – Amplitude of the seasonal

motion derived from one stack
 References
Reigber, A.; Moreira, A.; , "First demonstration of airborne SAR tomography using multibaseline L-band data," Geoscience and Remote
Sensing, IEEE Transactions on , vol.38, no.5, pp.2142-2152, Sep 2000
G. Fornaro, F. Serafino, and F. Soldovieri, “Three-dimensional focusing with multipass SAR data,” IEEE Trans. Geosci. Remote Sens., vol.
41, no. 3, pp. 507–517, Mar. 2003.
F. Lombardini, “Differential tomography: a new framework for SAR interferometry,” IEEE Trans. Geosci. Remote Sens., vol. 43, no. 1, pp.
37–44, Jan. 2005.
G. Fornaro and F. Serafino, “Imaging of Single and Double Scatterers in Urban Areas via SAR Tomography,” IEEE Trans. Geosci. Remote
Sens., vol. 44, no. 12, pp. 3497–3505, Dec. 2006.
F. Lombardini and M. Pardini, “3-D SAR Tomography: The Multibaseline Sector Interpolation Approach,” IEEE Geosci. Remote Sens. Lett.,
vol. 5, no. 4, pp. 630–634, Oct. 2008.
G. Fornaro, D. Reale, and F. Serafino, “Four-Dimensional SAR Imaging for Height Estimation and Monitoring of Single and Double
Scatterers,” IEEE Trans. Geosci. Remote Sens., vol. 47, no. 1, pp. 224–237, Jan. 2009.
 X. X. Zhu and R. Bamler, “Tomographic SAR Inversion by L1-Norm Regularization—The Compressive Sensing Approach,” IEEE Trans.
Geosci. Remote Sens., vol. 48, no. 10, pp. 3839–3846, Oct. 2010.
X. X. Zhu and R. Bamler, “Let’s Do the Time Warp: Multicomponent Nonlinear Motion Estimation in Differential SAR Tomography,” IEEE
Geosci. Remote Sens. Lett., vol. 8, no. 4, pp. 735–739, 2011.
X. X. Zhu and R. Bamler, “Very High Resolution Spaceborne SAR Tomography in Urban Environment,” IEEE Trans. Geosci. Remote Sens.,
vol. 48, no. 12, pp. 4296–4308, 2010.
X. X. Zhu and R. Bamler, “Super-Resolution Power and Robustness of Compressive Sensing for Spectral Estimation With Application to
Spaceborne Tomographic SAR,” IEEE Trans. Geosci. Remote Sens., vol. 50, no. 1, pp. 247–258, 2011.
 X. X. Zhu and R. Bamler, “Demonstration of Super-Resolution for Tomographic SAR Imaging in Urban Environment,” IEEE Trans. Geosci.
Remote Sens., vol. 50, no. 8, pp. 3150–3157, Aug. 2012.
F. Lombardini and M. Pardini, “Superresolution Differential Tomography: Experiments on Identification of Multiple Scatterers in
Spaceborne SAR Data,” IEEE Trans. Geosci. Remote Sens., vol. 50, no. 4, pp. 1117–1129, 2012.
D. Reale, G. Fornaro, and A. Pauciullo, “Extension of 4-D SAR Imaging to the Monitoring of Thermally Dilating Scatterers,” IEEE Trans.
Geosci. Remote Sens., vol. 51, no. 12, pp. 5296–5306, 2013.
G. Fornaro, A. Pauciullo, D. Reale, and S. Verde, “Multilook SAR Tomography for 3-D Reconstruction and Monitoring of Single Structures
Applied to COSMO-SKYMED Data,” IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens., vol. 7, no. 7, pp. 2776–2785, Jul. 2014.
X. X. Zhu, S. Montazeri, C. Gisinger, R. F. Hanssen, and R. Bamler, “Geodetic SAR Tomography,” IEEE Trans. Geosci. Remote Sens., vol. 54,
no. 1, pp. 18–35, Jan. 2016.
Tomographic scene reconstruction

Assuming typical airborne or spaceborne MB geometries, SAR Tomography can be

formulated according to one simple principle:
Each focused SLC SAR image is obtained as the Fourier Transform of the scene
complex reflectivity along the cross-range coordinate

 4 
y n r , x    s r , x , v  exp   j bn v dv
 r 
yn(r,x) : SLC pixel in the n-th image
s(r,x,v): average complex reflectivity of the scene within
the SAR 2D resolution cell at (r,x)
bn : normal baseline for the n-th image
λ : carrier wavelength

 The cross-range distribution of the complex reflectivity can be retrieved through

Fourier-based techniques

Source: Tebaldini & Rocca

 Tomographic scene reconstruction

Example: Tomographic reconstruction of a forest scenario

Simulated Beamforming
Simulated Backscattered Power Distribution (BPD) 40 40
N=6
N = 12
30 30

height [m]

height [m]
20 20

10 10

height, z
0 0

-10 -10
0 0.5 1 0 0.1 0.2
BPD Estimated BPD
Capon Spectrum COMET - 2 Layers
40 40
N=6 N=6
N = 12 N = 12
30 30

height [m]

height [m]
20 20

|s(r,x,z)|2 10 10

Contributions from volume backscattering 0 0 Source:

Contributions from ground backscattering
-10 -10
Tebaldini &
Contributions from ground-trunk interactions 0 0.2
Estimated BPD
0.4 0
Estimated BPD
0.5
Rocca
BIOMASS tomographic phase

Capon spectrum - HH channel

60
LiDAR height
Cross range 40

Height [m]
20

π/2 0

200 400 600 800 1000 1200 1400 1600 1800 2000

θ Slant range Capon spectrum - HV channel

60
LiDAR height
40

Height [m]
Elevation 20

200 400 600 800 1000 1200 1400 1600 1800 2000

Capon spectrum - VV channel

60
LiDAR height
40
Height [m]

20

200 400 600 800 1000 1200 1400 1600 1800 2000
Slant range [m]

Source:
Tebaldini & Rocca
 Tomographic analysis: TropiSAR

A closer look…

Source: Rocca
 Tomographic analysis: TropiSAR

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
Source: Rocca
 Tomographic analysis: TropiSAR

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
Source: Rocca
 Tomographic analysis: TropiSAR

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
Source: Rocca
 Tomographic analysis: TropiSAR

Co-polar signature at the ground

layer reveals ground-trunk double
bounce interactions dominate the
signal from flat areas despite the
presence of a 40 m dense tropical
forest

Source: Rocca
 Towards BIOMASS

 The scattering mechanisms at P-band in a very dense tropical forest:

Ground scattering is strongly visible and double bounces in flat terrain topography are
visible everywhere.

Volume scattering is significantly related to the high range biomass

 It was found that scattering contributions from about 30 m above ground

exhibit high sensitivity to forest biomass value ranging from 250 t/ha to
450 t/ha.

 SAR tomography allows to map not only vertical forest

structure but also biomass.
 Forest temporal decorrelation

P-band SAR tomography

key tool to SEE through the forest
suitable long wavelength to penetrate the dense forest
layer
key indicator to tropical forest biomass
Orbit constraint: temporal decorrelation
Revisit time ≥ 1 day in a sun synchronous satellite configuration
Forest scattering changes with time

GOAL: Study the temporal decorrelation of scattering mechanisms of the radar

signal in a tropical forest as a function of height and polarization.
 References
Polarimetric and tomographic phenomenology of forests
 Reigber, A.; Moreira, A.; , "First demonstration of airborne SAR tomography using multibaseline L-band data," Geoscience and
Remote Sensing, IEEE Transactions on , vol.38, no.5, pp.2142-2152, Sep 2000
 Mariotti d'Alessandro, M.; Tebaldini, S.; , "Phenomenology of P-Band Scattering From a Tropical Forest Through Three-
Dimensional SAR Tomography," Geoscience and Remote Sensing Letters, IEEE , vol.9, no.3, pp.442-446, May 2012
 Frey, O.; Meier, E.; , "3-D Time-Domain SAR Imaging of a Forest Using Airborne Multibaseline Data at L- and P-Bands,"
Geoscience and Remote Sensing, IEEE Transactions on , vol.49, no.10, pp.3660-3664, Oct. 2011

Multi-layer optimization
 S. Sauer, L. Ferro-Famil, A. Reigber, E. Pottier, “Multi-aspect POL-InSAR 3D Urban Scene Reconstruction at L-Band”, Eusar
2008
 Y. Huang, L. Ferro-Famil, A. Reigber, “Under Foliage Object Imaging Using SAR Tomography
and Polarimetric Spectral Estimators”, Eusar 2010

Coherence optimization
 M. Neumann, L. Ferro-Famil, A. Reigber, “Multibaseline Polarimetric SAR Interferometry Coherence Optimization”, IEEE
Geoscience and Remote Sensing Letters, 2008
 E. Colin, C. Titin-Schnaider, W. Tabbara, “An Interferometric Coherence Optimization Method in Radar Polarimetry for High-
Resolution Imagery”, IEEE Transactions on Geoscience and Remote Sensing, 2006

SKP decomposition: theory, algorithms and physical implications

 S. Tebaldini, “Algebraic Synthesis of Forest Scenarios from Multi-Baseline PolInSAR Data”, IEEE Transactions on Geoscience and
Remote Sensing, 2009
 Tebaldini, S.; Rocca, F.; , "Multibaseline Polarimetric SAR Tomography of a Boreal Forest at P- and L-Bands," Geoscience and
Remote Sensing, IEEE Transactions on , vol.50, no.1, pp.232-246, Jan. 2012
 Questions?
[email protected]