letoan-vseo-introduction to sar - icise quy nhon · 1.introduction to sar remotesensing 2.the...
TRANSCRIPT
Introduction to SAR remote sensing
Thuy Le ToanCentre d’Etudes Spatiales de la Biosphère (CESBIO)
CNRS, CNES, IRD, Université Paul SabatierToulouse, France
1. Introduction to SAR remote sensing
2. The Synthetic Aperture Radar
3. Physical content of SAR data
4. Statistical properties of SAR images
5. SAR image pre-processing
Contents
Non-Imaging (ex. microwave radiometer, magnetic sensor)Imaging (ex: cameras, optical mechanical scanner, spectrometer, microwave radiometer)
PASSIVE SENSORSDetect the reflected or emitted electromagnetic radiation from natural sources.
ACTIVE SENSORSDetect reflected responses from objects irradiated by artificially-generated energy sources.Non-Imaging ( ex: microwave altimeter, laser)Imaging (Real Aperture Radar, Synthetic Aperture Radar)
RADAR: Radio Detection and Ranging
SLAR: Side Looking Airborne Radar, developed during the World War II, for all weather and day and night aircraft operations over land and sea,
SAR: Synthetic Aperture Radar, airborne systems developed in 1950’s
SAR: Active microwave imaging system
Synthetic Aperture Radar (SAR)
The electromagnetic spectrum
Frequency bandKaK
KuXCSLP
Wavelength (cm)0.8-1.11.1-1.71.7-2.42.4-3.83.8-7.57.5-1515 -30
30 -100
Frequency (GHz)40 - 26.526.5 - 1818 - 12.5
12.5 -88 - 44 - 22 - 11 - 0.3
Radar frequency bands
f (in Hertz)=C/l C=3.108 ml =wavelength in m
ESA Advanced Training Course in Land Remote Sensing 12-16 September 2011, Krakow, Poland Thuy Le Toan, CESBIO, France
Available spaceborne SAR data
+ IW mode
Thuy Le Toan, CESBIO, Toulouse, France
1B (2016)
Radar
1. Day and night operation (independence of sun illumination)
2. All weather capability (with clouds, rains)3. No effects of atmospheric constituents
(→multitemporal analysis)4. Sensitivity to dielectric constant
(water content, i.e. soil moisture, biomass)5, Sensitivity to surface roughness
(sea state, topography..)6. Sensitivity to target structure
(→ use of polarimetry) 7. Accurate measurement of distance
(→ interferometry)8. Subsurface penetration
Optical
1. Day operation (Vis bands)
2.Weather limitations (clouds, rain)3. Effects of atmospheric constituents
(→ correction needed)4. Sensitivity to chemical constituents
(chlorophyl, soil..)5. Low sensitivity to surface roughness
(e.g. through shadow)6. No use of polarimetry
7. No measurement of distance (but photogrammetry)
8. No subsurface penetration
Radar versus optical remote sensing
radar acquisition:range discrimination
optical acquisition:angular discrimination
(From Elachi, 1989)
Radar versus optical remote sensing
1. Complex wave target interaction mechanisms(Difficulties in signal understanding and image visual interpretation)
2. Complex statistical properties of SAR images , e.g. speckle effects (Difficulties in image analysis and processing)
3. Topographic effects on both radiometric and geometric properties
Radar remote sensing: main difficulties
All-weather system
ERS-1 SAR, 11.25 a.m. LANDSAT TM, 9.45 a.m.Ireland, 09/08/1991
A microwave sensor: cloud penetrating capabilities
Marginal atmospheric effects
Strong attenuation
Low attenuation
Very low attenuation by atmospheric constituents for l > ~5 cm
Rain cell
12
by superpixel
Effect of heavy rain on C-band backscatter
Vu Phan Viet Hoa, M2 2019
1. Introduction to SAR remote sensing
2. The Synthetic Aperture Radar
3. Physical content of SAR data
4. Statistical properties of SAR images
5. SAR image pre-processing
Contents
Active system: day and night operations
Principle of imaging radar
Side Looking Radar
Linear displacement of the antenna along the track
Azimuth direction
Range direction
Pulses
The range information comes from the time needed by the pulse to travel way and back
Why side looking ?
L
q
The larger the antenna, the narrower the aperture (finer resolution)
'L
Antenna scattering
Resolution (r=
q.R)
R
Numerical example:L » 10m, R » 1000 km (spaceborne radar), l » 5 cm (C band) è resolution » 5 km
Angular aperture(horizontal plane)
Llq =
Antenna length (horizontal direction)
Wavelength
r’=q’.R
Ra=l R
L
Azimuth direction
Range direction
TransmittedpulsesH
=800
km
R a=5
km
L=10 m
f0 =5.3 GHz
Azimuth resolution= 5 km
18
Synthetic aperture technique An array of antennas is equivalent to a single antenna moving along the flight line LS if the received signals are coherently recorded and added, and the target assumed to be static during the period
XnX2X1
P
LS
The echoes from X1, X2, ..Xnare recorded coherently (amplitudeAnd phase as a function of time)
Ra=l RR
Azimuth resolution
Finest resolution: Ra=L
2LS
LS
v1/PRF
1 pulse è 1 echo = 1 image linePixel size : v/PRFPRF : Pulse Repetition Frequency
Image formation : Azimuth
Improvement of the range resolutionbased on frequency modulation of the transmitted pulse
5 km à 10 m (for B=15 MHz)
t Modulation bandwidth : B compt compt = 1/B
ct2
c2B
The SAR image
Satellite orbit
Ground range
Azimuth
Look angleOff nadir Slant range
Azi
mut
h
Raw data Detected imageAfter range compression
1. Introduction to SAR remote sensing
2. The Synthetic Aperture Radar
3. Physical content of SAR data
4. Statistical properties of SAR images
5. SAR image pre-processing
Contents
The radar scattering
the amplitude, phase and polarisation of Es are modifiedwith respect to Ei
Incident electric field Ei
Backscattered electric field Es
Basic measurement
The basic measurement made by a SAR is the amplitude and phase of the scattering matrix S. This is the complex image.
Main types of images:
A is the amplitude image.I = A2 is the intensity image.(the phase of a single image is not exploitable)
26
u polarimetric images derived from S the complex image if the SAR system is polarimetric
u interferometric images(coherence, phase) derived from S
u multi polarisation intensity images
u multitemporal intensity images
Sentinel-1 dual pol
Sentinel-1
Measurements from multiple images
27
The radar cross-sectionThe radar cross-section (RCS) is defined as
R is the radar-target distancePi is the incident power, Ps is the power scattered by the target.
[ ]222m44
i
spqpq P
PRS pps ==
28
The backscattering coefficientFor distributed targets each resolution cell contains many scatterers and the phase varies rapidly with position.
The differential backscattering coefficient, so, is
[m2/m2]
where is the area of the illuminated surface over which the phase can be considered constant.
i
so
PP
ARD
=24ps
AD
29
What is a SAR image?ASAR VV image After speckle filtering Combining filtered
HH and VV images
§ The image represents physical processes. § Pixels are measurements.§ Image is interpretable based on understanding of the physical processes
What is a SAR image?
The image is seen as a picture.
Pixels are numbers.
Image is affected by specklenoise
Example of an intensity imageAPP HH image 400 x 400 pixels (of 12.5m)Gaoyou, Jiangsu province, China, 2004 05 24
31
Same image, after speckle filtering
32
Initial HH and VV images HH and VV image after filtering
33
§ The image represents physical processes. § Pixels are measurements.§ Image is interpretable based on understanding of the physical processes
PolarisationMagnetic field Electric field Trajectory of the
electric field
Propagation direction
Transverse plane
Plane orthogonal to the propagation direction
Propagation direction
Electric field
Horizontal polarisation Vertical polarisation
Example of linear polarisations
Transmission H
HHHHReceive H
SHH SHH SHH SHH
SVH SVH SVH SVH
Receive V
HH VVHVALOS-PALSAR , Adamawa region, Cameroon, 11 -12-2009
38
HH VV
HV HHVVHV
Province of Quebec, Canada. RADARSAT2, 14 March 2009
ice
Scattering mechanisms
Surface scattering
Volume scattering if penetration
Surface scatteringwater
Volume scattering
Surface scattering
snow
soil, rock
Surface scattering
Volume scattering
Volume-surfacescattering
vegetation
Scattering mechanismsnThe backscattered signal results from:
- surface scattering- volume scattering- multiple volume-surface scattering
nThe relative importance of these contributions depend on- surface roughness- dielectric properties of the medium
n All of these factors depend on- the radar frequency- the polarisation- the incidence angle
Surface scattering
The roughness of the surface (wrt to the wavelength) governs the scattering pattern
er2
Smooth surface Rough surface
The dielectric constant (moisture content) of the medium governs the strength of the backscatter
er1er1
er2
er2 > er1 medium 2 is wetter than medium 1
Wetter media
Surface scattering: effect of surface roughness
Quaternary lithology:Bathurst Island, Canada
RADARSAT (C band, HH, 45°)
Limestone Higher backscatterbecause of rougher surface
From : RADARSAT Geology Handbook
Mud fragments (smooth surface)low radar backscatter
Mud
Lime stone
Seedbed Harrowed Ploughed
Tillage Rms
(mean)
Rms
(std)
Corr. Length
(mean)
Corr. Length
(std.)
Seedbed 0.6 0.3 3.7 2.6 Harrowed 1.6 0.7 3.8 2.9 Ploughed 2.7 1.0 6.9 2.7
Tillage roughness statistics
RMS heights s and correlation length l (mean and std in cm)
( ) ( ) ( )[ ] 23224220 sin21cos4-
+= qaqs klklks pppp
qqqqa
2
2
sincos
sincos
-Î+
-Î-=hh ( ) ( )
( )qqqqa2
22
sincos
sin1sin1
-Î+Î
+Î--Î=vv
Surface scattering
Herek = 2p /l (Wave number)
pp = HH or VVq = incidence angles = surface RMS heightl = surface autocorrelation length
.
Radar backscatter depends on the dielectric and geometry of the target, and on the frequency, polarisation and incidence angle of the wave e.g. Small Perturbation Method for surface scattering
( ) 00 =qs hv
σ0 is the notation used for the backscatter from a distributed scatterer, and is normalised by pixel area.
Scattering from soil surface
e = dielectric constant
Dielectric constant as a function of soil volumetric water content
Soil dielectric constant
( Le Toan, T. , 1982, "Active microwave signatures of soil and crops: Significant results of three years of experiments"
Experimental results using a ground-based radar scatterometer
Our first ERS experiment in 1992on irrigated area in Gharb, Morocco
We mapped irrigated fields, but to retrieve soil moisture was found hard with a single ERS data!
Le Toan, T, J.C. Souyris and P. Macchia, 1993
The relationship between radar backscatter at C band 23° VV and soil moisture is modulated by surface roughness
Volumetric soil moiture content (%)
ERS
back
scat
terin
g co
effic
ient
(dB)
Surface scattering: Effect of roughness and moiture
Mattia et al., 2000
Low backscatterè dry soil
High backscatterè Wet soil
Low backscatterè Smooth surface High backscatter
è rough surface
BUT :
Simultaneous effect of roughness and moisture on the radar signal
For constantroughness
For constantmoisture
Sensitivity to soil moisture and surface roughness
Polarisation in surface scattering
VV
u no depolarisationno HV or VH backscatter
u Fresnel Reflectivity RH > RV
V H
u some depolarisationHV or VH backscatter > 0
u Fresnel Reflectivity RH = RV
Smooth surface Rough surface
50Thuy Le Toan, CESBIO, Toulouse, France
Volume scattering
Single and multiple scattering
51Thuy Le Toan, CESBIO, Toulouse, France
Polarisation in volume scattering
V
V
V
H
Point scatterer
-> no depolarisation
Anisotropic scatterer
-> depolarisation
Multiple scattering
-> depolarisation
V
H
H,V
52Thuy Le Toan, CESBIO, Toulouse, France
Austrian pine VHFl > 3 m
P bandl= 70 cm
L bandl= 27 cm
X bandl= 3 cm
The main scatterers in a canopy are the elements havingdimension of the order of the wavelength
What are the scatterers in the volume scattering?
* Order 1: simple scattering
soil soilsoil(negligible)
* Order 0
soil
0.
0.
00vegsoilvegsoil -++= ssss
Scattering from vegetation
54Thuy Le Toan, CESBIO, Toulouse, France
hv
k
scatterers size,shape scatterers orientation
scatterers density
scatterers water content
soil roughness
soil moisture
Dielectric properties
Geometric, structural properties
Vegetation scattering
55Thuy Le Toan, CESBIO, Toulouse, France
1) Direct Crown scattering
3) Trunk scattering
4) Multiple trunk-ground5) Attenuated ground6) Direct ground scattering
2)
2
3
54
1 6
Trunk
Primary Branches
Secondarybranches
Higher orderbranches
Leaves, Needles
Scatterers contribution
Direct trunk-ground
Effet de la pol. PALSAR
Polarisation HH
ALOS PALSARAmazon forest
J.M. Martinez, IRD
ESA Advanced Training Course in Land Remote Sensing 12-16 September 2011, Krakow, Poland Thuy Le Toan, CESBIO, France
Polarisation HV
Polarisation HH and HV
Effet de la pol. PALSAR
J.M. Martinez, IRD
HH
HV
ALOS PALSAR Amazon forest
Varzea Dry Season Varzea Wet Season
SAR image SAR image
Sub-canopy observations
Document S.Saatchi, JPL
Scattering on leaves, ears
Attenuated ground scattering
Stem-ground interaction
Scattering from a cereal canopy
HH VV
Hongze (Jiangsu) 2004 09 06
Hongze(Jiangsu)
2004 09 06
Rice mapping using HH/VV
Hybrid rice
(Alexandre Bouvet)
Japonica rice
Mapping of rice varieties
EO : mapping of rice varieties
63
Differences in information content with optical data
Optical image Sentinel-2
Radar image Sentinel-1
Phase in SAR images (1/2)iji
ijij eSS f=The phase difference between scatterers of the incident wavestravelling from the radar to a scatterer and back to the radar changes as:
Dr is the difference in the travel distance
Since the SAR resolution cell contains a large number of scatterers, the phaseof pixels seems randomly distributed
Df =2pDrl
_____
pp-The phase of a single SAR image is of no utility
The SAR measurement contains an amplitude and a phase
Phase in SAR images (2/2)If the scene is observed in 2 images, in which the scatterers remain unchanged in the resolution cell, the phase difference between pixels of the 2 images can be exploited
Polarimetry: the radar measures at the same time HH, VV, HV, VH and their phase difference
Interferometry: 2 radars observe the scene with a small shift in the look angle; or the same radar at different dates from lightly shifted orbit
Relief Terrain displacement
iso-displacement curvesiso-altitude curvesEtna Landers
i
B
Digital elevation models Cartography of terrain displacements
Accurate range measurementRadar Interferometry
• Impact of a geothermal plant on the environment. Interferogram processed from two ERS images, acquired at two years interval. The fringes characterize the ground subsidence around the plant. One observe a subsidence of about 6 cm (2 fringes) which covers 17km x 8km.
ERS intensity image Interferogramme
Mesa, USA/ Mexico border
Radar InterferometryERS Interferometry
Sụt lúnTrượt lỡ đất
Đô thị Đập thủy điện
Kỹ thuật SAR giao thoa xác định biến dạng
-40 100-20-30 -10mm/yr
1996-2002 2006-2010
Kỹ thuật SqueeSAR, 18 ảnh ALOSKỹ thuật SqueeSAR, 14 ảnh ERS
Vận tốc lún trung bình (mm/nam)
Sụt lún ở khu vực Tp. Hồ Chí Minh
So sánh trị đo thủy chuẩnTuyến đo thủy chuẩn
Và đường đồng mức của tốc độ hạ thấp của mực nước ngầm
-40 -30 -20 -10 0 10-40
-30
-20
-10
0
10
43
29
48
15
7
35
33
28
RMSE = 4.6 (mm/yr)R2 = 0.86
Vận tốc lún - thủy chuẩn (mm/yr)
Vận
tốc
lún
-Squ
eeSA
R (m
m/y
r)
Sụt lún ở khu vực Hà Nội
-40 100-20-30 -10mm/yr
Vận tốc lún trung bình mm/năm
2007-2011
SqueeSAR technique, using 22 ALOS images.
The BIOMASS mission: Quantifying biomass for carbon assessment
Polarimetric Interferometry (PolinSAR)
Phas
e C
entre
hei
ght(
m)
Tree height
Ground level
HV
VVHH
Phas
e C
entre
hei
ght(
m)
Tree height
Ground level
HV
VVHH
Garestier, 2006
Tree height inversion
Pol-InSAR provides the canopy height estimates
Key requirement:
High temporal correlation between two images, hence P-band is essential
HV
HH
VV
§ Interferometry provides height information § The measured height depends on polarisation§ PolInSAR retrieves canopy height using models
50
40
30
20
10
0mPeat Swamp forest, Mawas, Kalimantan
Hajnsek et al., IEEE GRS 2009
Height (m)
PolInSAR provides the canopy height estimates
Tomography generates images of different forest layers from multi-orbit SAR acquisitions
10.90.80.70.60.50.40.30.20.10
Normalised backscatter
intensity
45
30
15
0
Height (m)
SAR resolution cell SAR Tomography
resolution cellEnables Isolation
of:• Canopy• Ground• Topography
x
y
z
o
TomoSAR provides 3 D images of the forests
x
y
z
o
Boreal forestKryclan, Sweden
Tropical forestParacou, French Guiana
TomoSAR provides 3 D images of the forests
Azimuth [pixel]
Gro
und
rang
e [p
ixel
]
Biomass map obtained by inversion power layer 30m (t.ha-1)
1000 2000 3000 4000 5000 6000 7000 8000
2000
2500
3000
3500
4000
4500
5000
5500
6000
6500
7000
50
100
150
200
250
300
350
400
450
500
550
600
BIOMASS map at 50 m, by tomography
SAR Tomography provides high accuracy biomass maps
P-band SAR image
Ho Tong et al., 2015, 2016
Secondary productsBiomass will allow DEM production under dense tropical canopies
TropiSAR data
Image by courtesy of P. Dubois-Fernandez
Paracou
1. Introduction to SAR remote sensing
2. The Synthetic Aperture Radar
3. Physical content of SAR data
4. Statistical properties of SAR images
5. SAR image pre-processing
Contents
Statistical propertiesof SAR images
Thuy Le Toan, CESBIO, Toulouse, France
The image is seen as a picture.
Pixels are numbers.
Image is affected by speckle noise
Example of an intensity imageAPP HH image 400 x 400 pixels (of 12.5m)Gaoyou, Jiangsu province, China, 2004 05 24
Thuy Le Toan, CESBIO, Toulouse, France
1.What is speckle?DefinitionPhysical originStatistics
2. How to filter it?Spatial filters (single channel)Multi-channel filter
Speckle in SAR images
Thuy Le Toan, CESBIO, Toulouse, France
Measuring backscatter
Speckle = multiplicative noise
RSAT standard image
Speckle noise
fD
eÂ
mÁ
1 2
3 4
1 2
3 4
pixel n°1 pixel n°2
pixel n°3 pixel n°4
pixel n°1 pixel n°4mÁ
eÂ
Imageradiometry
Origin of speckle noise
After speckle filtering
Speckle noise must be reduced…
Before speckle filtering
…to start understanding SAR image content
Statistics of speckle
Thuy Le Toan, CESBIO, Toulouse, France
The Gamma distribution
Multilooking reduces the effect of speckle.The distribution tends to normality as L increases.
Thuy Le Toan, CESBIO, Toulouse, France
Speckle filtering
Thuy Le Toan, CESBIO, Toulouse, France
Spatial filtering
Thuy Le Toan, CESBIO, Toulouse, France
Purpose of filter:(1) Preserve radiometry Þ unbiased
(2) Minimise the variance of subject to the Mconstraints in (1)
Multi-image Intensity Filtering
1I 2I
Filter
2J1J
MI
MJ
. . . . . . .
. . . . . . .
( ) ( ) MkyxJyxI kk ££= 1,,
kJ
Original Images
Filtered images
Thuy Le Toan, CESBIO, Toulouse, France
ENL=20
Mean = -14.70 dB
SD = 1.58 dB
ENL=4
Mean = -14.70 dB
SD = 2.47 dB
ENL=20
Mean = -11.89 dB
SD = 1.21 dB
ENL=4
Mean = -11.89 dB
SD = 2.31 dB
-8 dB
-20 dB
γoHV
Non filtered – 4 looks
Filtered – 20 looks
1
2
1f
2f 1
2
1f
2f
Multi-image filtering reduces variance and preserves radiometry
Mermoz et al., 2016
HH
HHVV
VV
IntensityimagesHongze (Jiangsu)2004 09 06
Multi-channel speckle filtering
Filtered images using HH, VV at multidates (10)
HH
Thuy Le Toan, CESBIO, Toulouse, France
1. Introduction to SAR remote sensing
2. The Synthetic Aperture Radar
3. Physical content of SAR data
4. Statistical properties of SAR images
5. SAR image pre-processing
Contents
Image preprocessing
1. Calibration, to convert the data to standardgeophysical measurement units.
2. Geocoding, to allow the data to be referenced to a map and to allow geolocation.
3. Registration, to make sure multiple images can be matched point to point.
There are well-developed methods for each of these operations.
Thuy Le Toan, CESBIO, Toulouse, France
Calibration
Calibration is the process converting voltages measured at the sensor to geophysical units; (usually s0).
Relative calibration ensures that the values everywhere in a single image are proportional to s0.
Calibration is essential to allow comparison of data from
¨ different sensors¨ different times¨ different positions
Thuy Le Toan, CESBIO, Toulouse, France
Generation of stacked time series from a set of downloaded pre-processed images
PEPS: CNES platform that provides free access to data from the Sentinel satellites
OTB Orfeo ToolBox: CNES open source for state-of-the-art remote sensing
A CNES-CESBIO processing chain to handlethe large amount of S1 data
Thuy Le Toan, CESBIO, Toulouse, France
o Fast parallel processing of 10m resolution imageso Automatically downloads, calibrates, orthorectifies and filters speckle noiseo Multi-image filtering particularly adapted to dense time serieso Images subset directly superposed to Sentinel-2 100x100 km2 tiles
Thuy Le Toan, CESBIO, Toulouse, France
Generation of stacked time series from a set of downloaded pre-processed images
A CNES-CESBIO processing chain to handlethe large amount of S1 data
106