DEM GENERATION

Michele Crosetto

[email protected]

SAR TECHNIQUES
1. Geometric applications, based on the phase:

 Digital Elevation Model (DEM) generation

- Technique: interferometric SAR (InSAR)
 Deformation monitoring
- Technique: differential interferometric SAR (DInSAR)

2. Thematic applications, mainly based on amplitude.

1
DEM generation

Source: J. Moreira Modulo SAR del Master del IG en “Airborne SAR Remote Sensing”
3

DEM vs. DTM

2
Methods for DEM generation
- Examples of methods:

1. Stereoscopy with optical images: photogrammetry

2. Airborne laser scanner

3. Stereoscopy with SAR amplitude images: radargrammetry

4. Interferometric SAR

5. Etc.

Interferometric SAR
(InSAR)

3
Complex SAR images

- From a complex SAR image we can extract:

Q
  tan 1   a phase image
I

A I2  Q2 an amplitude image

SAR phase
Two components:

1. The phase shift that occurs in the interaction with the target: o
2. The geometric phase geom:

4   R
 geom  2    f 0  t 

where:
- f0 is the main frequency (e.g. 5.3 GHz in C-band)
-  is the wavelength (e.g. 5.66 cm in C-band)
- R is the sensor-target distance

4
InSAR principle
S 4 
 M  MP   o

M
4 
 S  SP   o

4 
SP  Int   S   M  SP  MP  
MP 
 Int  
SP  MP 
4 

- For each pixel:

- M, S known (from orbits)
- MP known (SAR slant range)
P
- SP measured using int 9

InSAR procedure
1. Acquisition of a pair of complex SAR images

2. Image co-registration 

3. Interferogram generation and filtering 

4. Coherence calculation 

5. Phase unwrapping 

6. InSAR model calibration

7. Generation of an irregular 3D point grid 

8. Generation of a regular 3D point grid (interpolation)

10

5
Acquisition geometry

 r sin 
z 2  Height ambiguity: the height difference that
2 B causes a phase jump of 2

- An interferogram has a good sensitivity to heights if the height

ambiguity is small

- Key parameter: B

11

BASE _|_ ALT AMBIG

int
Acquisition geometry [m] [m]
042 2.33 4313.44
033 27.52 365.69
072 66.74 150.64
012 -92.77 -108.29
066 -112.86 -89.18
079 -138.87 -72.50
081 -219.04 -45.96
070 250.87 40.13
068 -258.50 -38.88
085 301.61 33.35
014 325.84 30.85
031 -343.42 -29.29
069 -351.60 -28.58
053 -368.64 -27.29
041 -375.44 -26.77
057 377.79 26.60
023 -418.64 -24.01
027 -420.99 -23.87
061 452.82 22.22
063 485.81 20.71
12
010 527.83 19.08

6
 InSAR procedure
1. Acquisition of a pair of complex SAR images

2. Image co-registration 

3. Interferogram generation and filtering 

4. Coherence calculation 

5. Phase unwrapping 

6. InSAR model calibration

7. Generation of an irregular 3D point grid 

8. Generation of a regular 3D point grid (interpolation)

13

Co-registration

XM,YM XSHIFT
XS, YS

YSHIFT

Master image Slave image

14

7
Co-registration

15

Co-registration

16

8
Co-registration

17

Co-registration

18

9
Co-registration procedure
 Amplitude-based image matching:
 Input data: master and slave image
 Key: maximizing cross-correlation
 Main output: azimuth shift, range shift, correlation
 Estimation of the slave to master image transformation:
 Input data: table of shifts and correlations
 Key: robust least square estimation
 Main outputs: transformation parameters in az/rg
 Slave image resampling:
 Input data: slave image
 Key: image interpolation
 Main output: slave in master geometry

19

InSAR procedure
1. Acquisition of a pair of complex SAR images

2. Image co-registration 

3. Interferogram generation and filtering 

4. Coherence calculation 

5. Phase unwrapping 

6. InSAR model calibration

7. Generation of an irregular 3D point grid 

8. Generation of a regular 3D point grid (interpolation)

20

10
Interferogram computation

4 
 Int   S   M  SP  MP  

21

Filtering

4 
 Int   S   M  SP  MP  

22

11
InSAR procedure
1. Acquisition of a pair of complex SAR images

2. Image co-registration 

3. Interferogram generation and filtering 

4. Coherence calculation 

5. Phase unwrapping 

6. InSAR model calibration

7. Generation of an irregular 3D point grid 

8. Generation of a regular 3D point grid (interpolation)

23

Coherence

24

12
Coherence

25

Pixel selection
 The interferometric phase can only be exploited over those pixels
characterized by low noise.
 We need to perform a pixel selection. This is typically carried out using a
coherence-based selection, which is based on the coherence of the
interferograms.

26

13
Pixel selection - coherence
 The coherence  measures the degree of correlation between a pair of
SAR images.
 Given two complex SAR images s1 and s2 the coherence is given by:

 

E s1  s *2 
   E s 
E s1
2
2
2

 This is estimated using windows (n by m pixels – the reliability depends

on the number of pixels) :

 s 1  s*2
  n m

  s   s
2 2
1 2
n m n m

27

Pixel selection - coherence

 The standard deviation of the interferometric phase noise can be
estimated from the coherence.

where L is the number of multilooks (Bamler y Hartl, 1998).

28

14
Coherence

Coherence of two
ERS SAR
images over
Barcelona

T=1 day

29

Coherence

Coherence of two
ERS SAR
images over
Barcelona

T=1085 days

30
30

15
InSAR procedure
1. Acquisition of a pair of complex SAR images

2. Image co-registration 

3. Interferogram generation and filtering 

4. Coherence calculation 

5. Phase unwrapping 

6. InSAR model calibration

7. Generation of an irregular 3D point grid 

8. Generation of a regular 3D point grid (interpolation)

31

Phase unwrapping & aliasing

32

16
Phase unwrapping & aliasing

33

Phase unwrapping & aliasing

 Condition:

34

17
Phase unwrapping
Wrapped phase:
Q
Due to the fact that the interferometric phase comes from:   tan  
1

we obtain values contained in the interval: I

–     (i.e. we only know the interferometric phase modulus 2).
This phase is named wrapped phase.
Unwrapped phase:
As a consequence, there is an ambiguity in the phase value determination. We
have to determine k(i,j) in a way that:

(i,j)= (i,j) + 2k(i,j)

where: k(i,j) is an integer function,

(i,j) is the unwrapped phase,
(i,j) is the wrapped phase, and (i,j) are the pixel coordinates
This operation is called phase unwrapping.

35

Phase unwrapping
There are different phase unwrapping methods, e.g.:
 Ghost-lines: it is also known as branch-cut, residue-cut or
minimum spanning tree. Algorithms based on the direct integration of the
phase differences, avoiding some determined paths (initially developed by
Goldstein et al., 1988).
 Least squares (LS): Algorithms based on least squares (Ghiglia and
Romero, 1989). From LS is derived another method based on the Green
function formulation, which is theoretically equivalent to the LS method
(Fornaro et al., 1996).
 Minimum cost flow (Costantini, 1996). Another minimization criterion
using the norm Lp, with 0  p  2 was suggested by Ghiglia and Romero
(1996).
 There are other (many) algorithms, as for example those based on Kalman
filter, the region growing approach, etc.

36

18
 Phase unwrapping: direct integration
1. We assume that the absolute value of the
unwrapped phases differences is always less than in adjacent pixels, i.e.:

where  is the unwrapped phase. Otherwise, we cannot unambiguously

determine k: an aliasing occurs.

2. The key point of the method is the estimation of the partial derivatives of ,
x y y, which can be obtained considering the wrapped phase differences
between adjacent pixels :  x ( x, y)   ( x  1, y)   ( x , y) , using this
algorithm:
if  x   then  x   x
if  x   then  x   x  2  
if  x   then  x   x  2  

37

Phase unwrapping: direct integration

 Phase unwrapping is obtained through an
integration of phase differences.
 During the integration we find problems if the key hypothesis is not valid
in some pixels of the interferogram:

 Example of integration along an “horizontal path”:

Note: in this example phase values are rescaled from [-,] to [0,1]

0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3 This is not
0.0 0.0 0.4 0.3 0.0 0.0 0.4 0.3 respecting the
above hipothesis:
0.9 0.8 0.6 0.5 -0.1 -0.2 -0.4 -0.5
 < 
0.8 0.8 0.7 0.6 -0.2 -0.2 -0.3 -0.4

wrapped phase  unwrapped phase

38

19
 Phase unwrapping: direct integration
 Let’s choose a “vertical integration path”:
This is not
0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3 respecting the
above hipothesis:
0.0 0.0 0.4 0.3 0.0 0.0 0.4 0.3  < 

0.9 0.8 0.6 0.5 -0.1 -0.2 0.6 0.5 Note the jump is
0.8 0.8 0.7 0.6 -0.2 -0.2 0.7 0.6 located in another
place
wrapped phase  unwrapped phase

 With the above set of phases, if we change the integration path we get a
different estimation, i.e. different unwrapped phases.
 the unwrapped phase cannot depend on the integration path!

 There are different strategies to avoid the “critical points” during integration.

39

BCN
Coherence

22 x 22 km
Multilook 5x1

Tandem
ERS-1 3rd Nov 95
ERS-2 4th Nov 95
T=1 day

(mean= 0.65)

40

20
BCN

41

BCN - Montjuïc

• Montjuïc (180 m)

• One fringe = 45 m

42

21
BCN

43

BCN

44

22
InSAR procedure
1. Acquisition of a pair of complex SAR images

2. Image co-registration 

3. Interferogram generation and filtering 

4. Coherence calculation 

5. Phase unwrapping 

6. InSAR model calibration

7. Generation of an irregular 3D point grid 

8. Generation of a regular 3D point grid (interpolation)

45

Generation of the 3D points

- Inputs:
- M: orbit master
- S: orbit slave
- for each pixel: col, lin, Unw

- Compute: P(X,Y,Z)
- Three equations for each pixel:

MP  R o  R  col
fD
MP  VM    MP 
2

SP  MP  Unw   o  
4 

46

23
InSAR model calibration
- The model parameters (R0, R, T0, T, 0 y fD) can be affected by errors.
This in turn affects the positioning of P  :

MP  R o  R  col
fD
MP  VM    MP 
2

SP  MP  Unw   o  
4 
- Calibration of the model using ground control points:
- Inputs: - Orbits M y S
- Ground control points: col, lin, Unw, X, Y, Z

- Estimation by least squares of these paramenters:

R0, R, T0, T, 0, fD

47

Characteristics of
InSAR DEMs

48

24
Characteristics of InSAR DEMs
+ Relatively high spatial resolution:
- 20 by 20 m with medium SAR resolution (and good coherence)
- 5 by 5 m with high SAR resolution (and good coherence)

+ Precision:
- 5-8 m st. dev. with medium SAR resolution
- 2-3 m st. dev. with high SAR resolution
With two conditions: - good coherence
- moderate topography
- Problems with:
- low coherence areas
- mountain areas  phase unwrapping problems
- atmospheric effects

49

InSAR DEM
validation

50

25
 Second test case: heterogeneous terrain

Coherence of the
processed
sub-images
[5000 pixels]

(compressed 4
times in azimuth)
Azimuth

Area = 35 x 25 km
Base = 168 m 1.

Mean coherence
equals 0.57 0.
Slant Range [1500 pixels]

51

Ground Control Points

Geocoded
coherence
image and
distribution of
the GCPs

Area size is
approximately
25 x 35 km

52

26
GCP identification

GCP identification: homologous points on the orthophoto (left) and the

amplitude image (right)

53

InSAR validation

Reference DEM of the analysed area

54

27
 InSAR validation

Map of the height

differences
(InSAR vs
reference DEM)

Mean error: 1.2 m

St. Dev.: 18.1 m

55

Global InSAR missions

56

28
 A 11-day survey

https://www2.jpl.nasa.gov/srtm/
57

DEM spacing = 30 m 58

29
 59

http://www.intelligence-airbusds.com/worlddem/
https://directory.eoportal.org/web/eoportal/satellite-missions/content/-
/article/tandem-x

60

30