Lecture slides on Synthetic Aperture Radar at ASTI, Quezon City, Philippines

1. Theory and applications of
Synthetic Aperture Radar
Yosuke Aoki
Earthquake Research Institute, The University of Tokyo
Email: yaoki@eri.u-tokyo.ac.jp
12 February 2019
DOST-ASTI,
Diliman, Quezon City, Philippines

2. Schedule
1. Theory and application of Synthetic Aperture Radar
2. Processing Stripmap SAR images
2-a Deformation of the 2017 Leyte earthquake
2-b Deformation of the 2019 Luzon earthquake
2-c Deformation of the 2020 Taal eruption
3. Damage detection from SAR images
2-a The 2017 Leyte earthquake
2-b The 2020 Taal eruption
4. Processing ScanSAR images
4-a. The 2019 Cotabato and Davao del Sur (Mindanao) earthquakes
4-b. The 2020 Taal eruption

3. Country of disasters
Normal faulting
World’s riskiest cities from natural hazards
(Lloyd’s City Risk Index)
1. Taipei, 2. Tokyo, 3. Manila, 4. Seoul, 5. Shanghai,
6. Osaka, 7. Hong Kong, 8. Istanbul, 9. Mexico City, 10. Lima,
11. Los Angeles, …… 17. New York ……
 The main sources of risk in Tokyo are earthquakes and typhoons.
 The main sources of risk in Manila are earthquakes, volcanoes, and
typhoons.
 Earthquakes and volcanic eruptions are less often than typhoon but
more damage once they occur.

4. Image of an earthquake
Nature Cover page of the 8 July 1993 issue (Vol. 364, No. 6433)
Massonnet et al. (Nature, 1993)
 Coseismic deformation of the 1992
Landers (California, USA; Mw=7.3)
earthquake measured by Synthetic
Aperture Radar (SAR).
 Amazing spatial resolution (~3-5 meters)
 No need for a ground-based instruments
 Available day and night. All weather.
Compare with optical measurements.

5. The 1995 Kobe earthquake (Mw=6.9)
Ozawa et al.
(GRL, 1997)
 A L-band SAR satellite JERS-1 was available between 1992 and 1998.
 No L-band SAR satellite available between 1998 and 2006, stagnating
research in Japan.
 What is L-band? Why L-band satellites are so important in Japan (and
Indonesia)?

6. Frequencies of electromagnetic waves
 Electromagnetic
waves of longer
wavelength are better
at transmitting
vegetation. Big
advantage for
vegetated areas such
as Indonesia and
Japan!
 L-band satellites are
better than C- or X-
band satellites in
vegetated areas.

7. Previous and current SAR satellites
 ALOS-4 (L-band) is to be launched in late 2021.
 NISAR (L-band) is to be launched in 2021 (http://nisar.jpl.nasa.gov).

8. ALOS (2006-2011)
2007 Chuetsu-oki earthquake (Mw=6.8)
Aoki et al. (EPS, 2008)
2011 Tohoku-oki earthquake (Mw=9.0)
Feng & Jónsson (GRL, 2012)
LUSI Aoki & Sidiq (JVGR, 2014)

9. ALOS-2 (2014-)
2018 Palu earthquake (Mw=7.5)
Ho, Satake, Watada, Hsieh, Chuang, Aoki et al.
(JGR Solid Earth, submitted)
2017 Erta’Ale volcano eruption
Xu, Xie, Aoki, Rivalta, and Jónsson
(JGR Solid Earth, submitted)7
from the southern caldera (Fig. 3).145
146
147

10. Combining ALOS with other SAR satellites
Asama Volcano, Japan
Wang, Aoki, and Chen (EPS, 2019)
Usu Volcano, Japan
Wang and Aoki (JGR Solid Earth, 2019)

11. SAR amplitude image
 Taal volcano
 SAR image is complex with
phase and amplitude.
 Larger amplitude is
represented by white.
 Higher amplitude in the cities.
 Lower amplitude on the lake
and ocean.

12. How SAR works
 Radar = Radio detection and
ranging
 The satellite trasmits
electromagnetic wave
obliquely to the ground and
observes reflected waves.
 The transmission needs to be
oblique to distinguish different
points by different line-of-
sight distance.
 Flat surface does not generate
much reflected waves.

13. Shadow and layover
 Layover: Different points cannot be separated because they are at the
same distance from the satellite.
 Shadow: Rugged topography does not allow the electromagnetic wave
to reach.

14. Resolution in azimuth direction
 The resolution in azimuth
direction is a function of
antenna size.
 Moving antenna enhance the
resolution as if the target
were viewed by a big
antenna.

15. SAR interferometry and
Young’s interference experiment
Optical path difference
Width of fringes
which is inversely propotional
To the separation of two
satellites (baseline) and
proportional to the wavelength
of the electromagnetic wave.

16. Effect of orbital separation
Critical baseline
Satellites with longer
wavelength has larger critical
baseline.
L-band satellites require less
strict orbital controls than C-
band and X-band satellites.
Range resolution

17. Effect of topography
Phase difference

18. Effect of topography
Height difference over which the phase
difference is one cycle
Sensitivity to topography is higher with
higher frequencies.

19. Effect of topography
Topography of Etna volcano
Massonet & Feigl (Rev. Geophys., 1998)
Longer baseline is better to
measure topography in higher
sensitivity, but the baseline
should not exceed the critical
baseline.

20. Effect of Digital Elevation Model (DEM)
The 2007 Chuetsu-oki earthquake
Furuya, Takada, and Aoki (2010)
Interferogram with
higher-resolution DEM
gives more detailed
deformation field.
(top) GSI 50 m
(bottom) ASTER 15 m

21. Correcting interferograms
Interferogram = Orbit separation + Topography + Deformation
Pritchard (Phys. Today, 2006)

22. Phase unwrapping
phase change
integer
ambiguity
What we get by InSAR is the phase fraction (wrapped phase). We need
to estimate integer ambiguity to extract the real line-or-sight changes
(unwrapped phase).

23. Limitation of InSAR: Line of sight (LOS)
 InSAR measures the line-of-
sight component of the
displacement, not the 3D
displacement as in GNSS.
 Insensitive to north-south
displacements.
 More sensitive to vertical
displacements, but impossible
to separate vertical and
horizontal displacements.
LOS change from ascending and
descending orbits

24. 2.5D analysis
(Quasi) East-West and vertical displacements
or velocities can be separated if both
ascending and descending images are
available.
Wang & Aoki (JGR Solid Earth, 2019)

25. Limitation of InSAR: Swath width
2008 Wenchuan (Mw=8.0) earthquake
Hao et al. (GRL, 2009)
 The swath width is 50-70 km for stripmap mode. It takes some time to
observe the whole deformation field if the deformation is extended in
east-west direction.
 ALOS-2 has ScanSAR mode with a width of 350 km.

26. Limitation of InSAR: Decorrelation
2017 Leyte earthquake

27. Limitation of InSAR: Decorrelation
2017 Leyte earthquake

28. Limitation of InSAR: Decorrelation
2017 Leyte earthquake
 Change in surface feature
caused by landslide, surface
faulting, volcanic ash, etc,
decrease the coherence to
degrade the observation.
 Temporal decorrelation is
severe in vegetated regions
such as in Philippines and
Japan. Images with
temporal separation of 1
year can be incoherent in
Philippines.

29. Limitations of InSAR:
Atmospheric disturbance
 Electromagnetic waves refract in the
presence of water vapor, making the
apparent line-of-sight distance longer.
 Interferograms contain long-wavelength
patterns even if no real deformation is
present.
 Precise correction of the atmospheric effect
requires a precise knowledge of the spatial
variations of water vapor.
 Removal of altitude-correlated signals does
a reasonably good job. Lohman & Simons
(Geochem. Geophys. Geosyst., 2005)

30. Limitations of InSAR:
Ionospheric disturbance
Massonnet & Feigl (GRL, 1995)
 Electromagnetic waves propagate slower than
speed of light in the presence of ionized
atmosphere, making apparent line-of-sight
distance longer.
 The ionospheric effect is frequency-dependent,
so multiple frequencies of electromagnetic
waves allow us to extract the ionospheric effect.
 GNSS has two frequencies, but current SAR
satellites use only single frequency.
 L-band waves are more susceptible to the
ionospheric effect than C- and X-band waves.

31. Other limitations of InSAR
 Oblique incidence of electromagnetic waves makes layover and shadow
in areas of steep topography.
 Temporal resolution is limited by the recurrence time of the satellite.
ALOS-2: >14 days
ALOS: 46 days
Sentinel-1: 6 days (originally 12 days)

32. Earthquake deformation
Zagros Mountains, Iran
Lohman & Simons (Geochem. Geophys. Geosyst., 2005)
 InSAR is capable of detecting
displacements by
earthquakes down to M5.0-
5.5.
 InSAR is capable of precisely
locating M5.0-5.5
earthquakes whose location
error is up to a few tens of
kilometers with teleseismic
waveforms only.

33. Earthquake deformation
2017 Leyte earthquake (M 6.5)2019 Luzon earthquake (M 6.1)

34. Postseismic deformation
Gourmelen & Amelung
(Science, 2005)
 Three possible mechanisms of
postseismic deformation
- Deep afterslip (a few years)
- Viscoelastic relaxation (long)
- Poroelastic (< a few months)
 Gourmelen & Amelung (2005)
detected postseismic deformation
of earthquakes that occurred a
few tens of years before in
Nevada, USA. The observed
deformation is due to viscoelastic
relaxation of the lower crust and
upper mantle.

35. Poroelastic
deformation
Jónsson et al. (Nature, 2003)
 InSAR is capable of discriminating postseismic deformation of different
origin.
 Jónsson et al. (2003) succeeded in constraining the mechanism of early
postseismic deformation of two M6.5 earthquakes in South Iceland
Seismic Zone.

36. Interseismic
deformation
Fialko (Nature, 2006)
 Stacking large numbers of
interferograms allows us to infer
interseismic displacements as low as
~3-5 mm/yr by reducing noise.
 Fialko (2006) succeeded in separating
contribution from San Jucinto and San
Andreas faults in Southern California.
 Applicable to the Philippine fault
and major faults elsewhere?
Spatially variable fault slip rates on
an active fault is an important
information for hazard assessment!

37. Creeping the Philippine fault in the
Leyte Island
Fukushima et al. (EPS, 2019)
 InSAR is capable of
measuring spatial and
temporal variations of slip
rates and along a creeping
fault.
 Only SAR can measure
creeping rates in such high
spatial resolution.
 Important contribution of
assessing earthquake
hazards.

38. Volcano deformation
Amelung et al. (Nature, 2000)
 Volcanoes are inaccessible in
some areas because of steep
topography and danger from
volcanic activity.
 InSAR is capable of measuring
deformation of where ground-
based measurements are not
available.

39. Heuristic volcano
deformation study
Pritchard & Simons (Nature, 2002)
 Pritchard & Simons (2002) found
that some volcanoes in South
America which were recognized
inactive are actually accumulating
magma.
 A discovery only done by remote
sensing technique!

40. Application to Indoneisan volcanoes
Chaussard, Amelung & Aoki
(JGR Solid Earth, 2013)
Many volcanoes have sub-
optimal ground-based
network, so a systematic
monitoring by remote
sensing techniques is
required!
Applicable to Philippine
volcanoes?

41. Application to Philippine
volcanoes, 2007-2011
Morales Rivera et al.
(FRINGE, 2015; JGR Solid Earth, 2019)
various regions within the area (Fig. 2).
0 105
Kilometers
¯
100 m
300 m
200 m
0 105
Kilometers
¯
100 m
300 m
200 m
¯
100 m
300 m
200 m
0 0.2 0.4 0.6 0.8 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
−10
−5
0
5
10
cm/yr
Taal
Lake
LOS Laguna
de Bay
Kilometers
500 m
1000 m
1500 m
2000 m
0 31.5
Kilometers
¯
500 m
1000 m
1500 m
2000 m
¯
100 m
300 m
200 m
0 0.2 0.4 0.6 0.8 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
−3
−2
−1
0
1
2
3
cm/yr
LOS
t could possibly serve as evidence of the
ctonic interactions within the region. The
ould suggest that pressurization of the
system led to inflation of the volcano,
in stress transfers within the region and
while the volcano continued to inflate. But
ata and modeling is necessary to have an
ding of the interactions occurring within the
hich is out of the scope of this study.
on
olcano erupted lava flows during 2006 that
placed in the SE flank of the volcano
ian Institute, Global volcanism report,
at http://www.volcano.si.edu). A LOS
decrease is observed at Mayon volcano over
osits, with a maximum rate of 3.5 cm/yr, likely
with cooling and compaction (Fig. 3).
3.3. Bulusan
LOS velocity decrease is observed on the Western Flank
of Bulusan volcano (Fig. 4), at a maximum rate of 3.5
cm/yr. The signal coincides with the end of the 2007
eruption phase, indicating possible depressurization of
the volcanic system.
0 21
Kilometers
¯
1500 m
1000 m
500 m
0 21
Kilometers
¯
1500 m
1000 m
500 m
0 105
Kilometers
¯
100 m
300 m
200 m
cm/yr
LOS
0 0.2 0.4 0.6 0.8 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
−3
−2
−1
0
1
2
3
Taal
Mayon
Bulusan
Kanlaon
deforming area is unknown due to the loss of coherence
towards the W-SW flank of the volcano. The signal
appears to be morphostructurally confined within a
valley, and several scarp features can be observed with
Google Earth imagery, suggesting that mass movements
are the likely cause of the signal.
0 21
Kilometers
¯
1500 m
2000 m
1000 m 0 21
Kilometers
¯
1500 m
2000
1000 m
0 105
Kilometers
¯
100 m
300 m
200 m
cm/yr
LOS
0 0.2 0.4 0.6 0.8 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
−4
−3
−2
−1
0
1
2
Ongoing deformation should be
monitored in a systematic way!

42. The 2020 Taal eruption (coeruptive)

43. The 2020 Taal eruption (posteruptive)

44. Post-eruptive thermal contraction of
Usu volcano, Japan
 The 2000 vent: LOS velocity of about
38 mm/yr in the ALOS-1 period
(2006-2011). Negligible deformation
in the ALOS2 period (2014-2017).
 The 1977 vent: Maximum LOS
velocities of about 66, 45 and 43
mm/yr in the JERS (1992-1998),
ALOS-1 and ALOS-2 periods.
 The 1943 vent: Steady deformation
with a maximum LOS velocity of
about 20 mm/yr in 1992-2017.
Ascending Descending
2000
1977
1943
Wang & Aoki (JGR Solid Earth, 2019)

45. Time-varying volcano deformation: LUSI
Aoki & Sidiq (JVGR, 2014)

46. Land subsidence
Figure 7. Identification of subsided areas in Jakarta based on DInSAR technique. (a) Interferogram
from 2007-2011 based on ALOS-PALSAR data, (b) subsided area in Pluit region from 2007-2011 or
1472 days based on ALOS-PALSAR data, (c) subsided area in Pluit area from 2014 -2016 or 658 days
based on ALOS-2 data, image not rectified.
4. Conclusions
This research shows the ability of SAR data to identify ground deformation in Jakarta area by
analysing the amplitude and phase components. Sentinel-1A data and S1TBX software are useful to
obtain the land surface changes based on amplitude analysis. This research also found that
atmospheric phase affects much to C-band SAR data as already identify by previous studies [11] and
c
b
a
2nd International Conference of Indonesian Society for Remote Sensing (ICOIRS) 2016 IOP Publishing
IOP Conf. Series: Earth and Environmental Science 47 (2016) 012022 doi:10.1088/1755-1315/47/1/012022
Surabaya, ~30 mm/yr of subsidence
Aditiya, Takeuchi & Aoki (2017) North Jakarta,
Up to 260 mm/yr (!) of subsidence
Agustan et al. (2016)
“If we look at our models, by 2050 about 95% of North Jakarta will
be submerged.” – Heri Andreas (ITB), BBC Indonesian, Aug. 2018

47. Subsidence
in Manila
1993-1998
ENVISAT
2003-2010
ENVISAT
Raucoules et al. (RSE, 2013)
Subsidence of up
to ~70 mm/yr

48. Damage detection
mum case the temporal baseline between each acquisition is only one repeat
so Section 4).
(or more) SAR images, fulfilling the requirements mentioned above, are co-registered
master image and resampled to its reference grid (Figure 4). Additionally, a common
sures that only the overlapping parts of the spectrums are used. Thereby, the spatial
fect (see Section 2.1.) is reduced [74]. In the next step, interferograms between
he two slave images are generated: One pre-disaster InSAR pair (t1 and t2) and one
R pair (t2 and t3). Then, for both InSAR pairs the coherence is computed according to
e Section 2.1). Moreover, as described by Equation (9), also two SAR intensity
computed using again the co-registered pre- (t1 and t2) and co-disaster (t2 and t3)
airs [8]. The damage caused by the natural disaster is then assessed by detecting the
he corresponding image pairs (see Section 2.3 for more details).
pre co
Two possible metrics for damage
detection as a function of coherence C
e.g., Arciniegas et al.
(IEEE TGRS, 2007)
Hoffmann et al.
(Int. J. Remote Sens., 2007)
Here we only look at co-
disaster coherence.
Pre-disaster coherence should
also be used to for damage
detection to evaluate temporal
decorrelation.
Source of decorrelation
 Surface rupture
 Too much deformation
 Too much vegetation
 Landslide, ashfall
 Snow

49. Example: The 2017 Leyte earthquake

50. Example: The 2017 Leyte earthquake

51. Example: The 2020 Taal eruption

52. Example: The 2020 Taal eruption

53. New (unconventional) technique:
pixel offset
InSAR Pixel offset
Measurement Phase Amplitude
Error 20-30 mm ~300 mm
Large deformation Incoherent Capable
The 2016 Kumamoto earthquake

54. Pixel offset analysis:
Kumamoto earthquake
 Pixel offset analysis can give two-component displacements.

55. New (unconventional) technique:
Amplitude changes
Tobita et al. (EPS, 2006)
 Tobita et al. (2006) compared amplitude of SAR
images before and after the 2004 Sumatra-
Andaman earthquake to detect uplifted and
subsided regions on the coast.
 Subsidence decreases
amplitude of the SAR
image on the coast.
 Applicable to the 2020
Taal eruption?

56. What to do: Tectonics
 Estimating spatial variation of slip rates along the creeping Philippine
fault.
 Imaging deformation field associated with major earthquakes like the
2017 Leyte, 2019 Luzon, and 2019 Minadao earthquakes.
 Damage detection by earthquakes (and typhoon) from the coherence of
interferograms. Establishing a system to systematically process the
incoming data may be beneficial.
 Postseismic deformation?
 Combining SAR measurements with GNSS.

57. What to do: Volcano deformation
 Monitoring all Philippine (and beyond?) volcanoes
systematically with SAR images by processing incoming
images.
 Monitoring Taal and Mayon volcanoes in particular!
 Detecting ashfall, pyroclastic flow, and lava flow from InSAR
coherence. Establishing a system to systematically process
the incoming data may be beneficial.
 Combine InSAR with GNSS measurements (if any) to correct
InSAR measurements.

58. Summary
 SAR can measure Earth’s surface day and night, regardless
of the weather.
 SAR can measure surface displacements associated with
various phenomena with a spatial resolution, but with
some limitations.
 SAR can be a powerful tool in detecting damages caused
by landslide, flooding, tsunami, ….
 Philippine people will benefit a lot from SAR images!