Rocca.ppt
-
Upload
grssieee -
Category
Technology
-
view
1.485 -
download
0
Transcript of Rocca.ppt
Page 1Vancouver, IGARSS, July 28, 2011
F RoccaDipartimento di Elettronica e Informazione
Politecnico di Milano
SAR interferometry for sub millimeter land motion studies
Page 2Vancouver, IGARSS, July 28, 2011
ERS -1 - 1991 35 days revisit cycle
Page 3Vancouver, IGARSS, July 28, 2011
Terrain Deformation Monitoring:
The ERS – Envisat era
Page 4Vancouver, IGARSS, July 28, 2011
Decennial Etna motion: vertical
Page 5Vancouver, IGARSS, July 28, 2011
Decennial Etna motion: E-W
Page 6Vancouver, IGARSS, July 28, 2011
Piton de la Fournaise(Isle de la Reunion)
Mad
agas
car
LA REUNION
Piton de la Fournaise
Page 7Vancouver, IGARSS, July 28, 2011
Differential Interferograms: examples (1)S2 Ascending T84
π
-π15
-74
radia
ns
Master: 20031130; Slave: 20030921 Bt=70 [days] Bn=29 [m]
wrapped
unwrapped
wrapped
unwrapped
wrapped
unwrapped
Page 8Vancouver, IGARSS, July 28, 2011
S2 Ascending T84 Velocity Field
Page 9Vancouver, IGARSS, July 28, 2011
S4 Ascending T127 Velocity Field
Page 10Vancouver, IGARSS, July 28, 2011
S2 Ascending T84 Time Series: examples (1)
1
2 3
1
2
3
master
mastermaster
Page 11Vancouver, IGARSS, July 28, 2011
S4 Ascending T127 Time Series: examples (1)
1
2 3
1
2
3
master
mastermaster
Page 12Vancouver, IGARSS, July 28, 2011
Vulcano, Eolie Islands (Italy)
Descending track 49430 scenes Envisat S2
Ascending track 12937 scenes Envisat S2
Page 13Vancouver, IGARSS, July 28, 2011
Velocity field, along LOS
Page 14Vancouver, IGARSS, July 28, 2011
Velocity field, along LOS
Page 15Vancouver, IGARSS, July 28, 2011
Decomposition in East and Vertical velocities
Vertical velocity fieldEasting velocity field
eastwestupdown
Ascending and descending results both cover the crater area and other parts of the island: wherever the two data are simultaneously available, adecomposition from ascending and descending displacement to easting and vertical components is possible, on a grid of 100x100 meters resolution
Page 16Vancouver, IGARSS, July 28, 2011
Displacement Time Series, examples
Page 17Vancouver, IGARSS, July 28, 2011
Displacement Time Series, examples
Page 18Vancouver, IGARSS, July 28, 2011
Landslides detection and monitoringId
en
tify
ing
lan
dslid
es:
Pie
dm
on
t Piedmont landslides
Page 19Vancouver, IGARSS, July 28, 2011
Page 20Vancouver, IGARSS, July 28, 2011
Page 21Vancouver, IGARSS, July 28, 2011
Page 22Vancouver, IGARSS, July 28, 2011
Page 23Vancouver, IGARSS, July 28, 2011
Sentinel 1 A/B 22 years later
12 days revisit cycle
Page 24Vancouver, IGARSS, July 28, 2011
SAR interferometry 20 years later(1991 – 2011)
How about ground motion recovery?
• Target selection for coherence optimization• Absolute geometries• Shorter revisit times with constellations
Nowadays:The path delay measurement is reliable tothe mm, for a reasonable spatial resolution
Page 25Vancouver, IGARSS, July 28, 2011
Densifying the set of reference targets
(Persistent Scatterers)
Page 26Vancouver, IGARSS, July 28, 2011
Covariance matrices of multipass SAR
Rows and columns correspond to progressive times
Examples for:Persistent Scatterers (that terminate)Progressive decorrelationSeasonal effects
Page 27Vancouver, IGARSS, July 28, 2011
Page 28Vancouver, IGARSS, July 28, 2011
SqueeSAR uses nearly all interferograms, weighted with their coherence (see above). Phase linking is
then carried out, estimating the sequence of the N-1 phases using the N(N-1)/2 interferograms.
seasonalMarkov
Page 29Vancouver, IGARSS, July 28, 2011
Page 30Vancouver, IGARSS, July 28, 2011
PS
Page 31Vancouver, IGARSS, July 28, 2011
Squeezing all the information:Squeesar
Page 32Vancouver, IGARSS, July 28, 2011
An example: the InSalah Case (Algeria)
The InSalah Gas storage project is the first CO2 sequestration effort in an active reservoir.
1 million CO2 tons are reinjected into the subsurface each year.
PSInSAR estimated volume and pressure changes, and finally the permeability within the reservoir.
Page 33Vancouver, IGARSS, July 28, 2011
Page 34Vancouver, IGARSS, July 28, 2011
Elastic Model Definition
Apertures (dislocations) and volumetric variations along a fault plane
Inversion using the Poisson ratio in the overburden
Inverted parameters:Fault plane position Center of fault Reject
Page 35Vancouver, IGARSS, July 28, 2011
Inversion results
UD component
Measures Model
Page 36Vancouver, IGARSS, July 28, 2011
Inversion results
EW component
Measures Model
Page 37Vancouver, IGARSS, July 28, 2011
Page 38Vancouver, IGARSS, July 28, 2011
Arrival time and Trajectories
Pressure changes → propagating fronts We use the time derivative of the pressureFrom the time arrivals → permeability
A. Rucci, D. W. Vasco, Fluid pressure arrival time tomography, SEG 2009, Houston
Arrival time Trajectories
Page 39Vancouver, IGARSS, July 28, 2011
The reject across the fault
Page 40Vancouver, IGARSS, July 28, 2011
Cosmo Skymed data:Total change: 1.8mm
28kt/mm
Page 41Vancouver, IGARSS, July 28, 2011
We can map surface deformations into permeability changes; the InSalah story
is condensed in 1 cm surface motion
1 mm sensitivity is achieved today with reduced resolution, but can improve with numerical atmospheric models.
The revisit time is paramount
Page 42Vancouver, IGARSS, July 28, 2011
Effects of motion direction
• The UD component is practically LOS• The SNR of the EW component is just a few dB
down • The NS component can only be retrieved with
speckle correlation. The dispersion σd, referred to the azimuthal resolution is, for N looks:
NN
39.01
2
3 2
Page 43Vancouver, IGARSS, July 28, 2011
MAI: Multiple Aperture InterferometryHowever, if the wide Doppler range of Sentinel - 1 (±0.70) is fully used say for SPOT instead than TOPS scan, then the equivalent azimut resolution is δeq:
mL eq
eqeq
6.057
4.1
42*2
With say 1500 looks (4500m2), the dispersion of the ground motion estimates along the NS direction could be of the order of 1 cm. Multiple Aperture Interferometry proposes to use wide Doppler ranges to achieve a very high equivalent azimuth resolution, even if not completely filling the entire Doppler band (pass band speckle correlation [1]). [1] De Zan F., 2011, Coherent Shift Estimation for Stacks of SAR Images, GRSL to appear
Page 44Vancouver, IGARSS, July 28, 2011
We still have to solve efficiently the problem of the Atmospheric Phase Screen
that biases the outcomes
Page 45Vancouver, IGARSS, July 28, 2011
APS estimation procedure
Model Parameters
DEM
Page 46Vancouver, IGARSS, July 28, 2011
Mathematical modeling of the APS
CByAxyxzKyxyxAPS turbulence ),(),(),(
Characterized by a variogram
K,A,B,C LMS from the data Z from a DEM
)(1 L
d
eSN 4-parameter model
Page 47Vancouver, IGARSS, July 28, 2011
Typical example of atmospheric data
1cmpath traveladditional way two
12
r
r rad
Page 48Vancouver, IGARSS, July 28, 2011
APS from Ground Based RADAR
10-1
mm
2
APS Variogram (stable targets)
100
10-3
10-2
10-1 102101100 10310-2
10 ms 16.6 min
Urban
Mountain
days
APS Variogram (mountain)
mm
2
104
103
102
101
100
10-1
10-3 10-2 10-1 100 101
GBSAR measures APS fluctuations from ms to months (range = 4 km).
The ta power law (Kolmogorov) has been verified on variograms from t > 1 s
APS has considerably less power during night time
APS in short time APS in long time
σφ=1 rad
2hours ~ 40km
mm
2
mm
2
Page 49Vancouver, IGARSS, July 28, 2011
K values: synoptic view
Ascending Descending
Page 50Vancouver, IGARSS, July 28, 2011
6am
Page 51Vancouver, IGARSS, July 28, 2011
6pm
Page 52Vancouver, IGARSS, July 28, 2011
What we can do today with good control of the
Atmospheric Phase Screen
Page 53Vancouver, IGARSS, July 28, 2011
The CESI experimentEstimated displacement (Radarsat 1 data)
-4
-2
0
2
4
6
8
10
12
14
Tempo
Sp
os
tam
en
to [
mm
]
SPOSTAMENTO ORIZZONTALE EFFETTIVO
SPOSTAMENTO ORIZZONTALE STIMATO
-10
-8
-6
-4
-2
0
2
4
6
8
10
12
Tempo
Sp
os
tam
en
to [
mm
]
SPOSTAMENTO VERTICALE EFFETTIVO
SPOSTAMENTO VERTICALE STIMATO
Vertical displacement (ground truth)Vertical displacement (estimated)
Horizontal displacement (ground truth)Horizontal displacement (estimated)
Rmse = 0.58mm (h!); 0.75mm (v)
Page 54Vancouver, IGARSS, July 28, 2011
A Ground Based SAR at the XX dam
The future: UAV, geosynchronous, both?
IBIS-L
Page 55Vancouver, IGARSS, July 28, 2011
Corner 2
Corner 1
Corner 4
Corner 3
Page 56Vancouver, IGARSS, July 28, 2011
Permanent Scatterers analysis : displacement series
16:48 19:12 21:36 00:00 02:24 04:48 07:12 09:36 12:00 14:24-2
-1.5
-1
-0.5
0
0.5
1
1.5
2Multi-target displacement
Acquisition time [hh:mm]
[mm
]
17-Sep
PSgr-21596
Page 57Vancouver, IGARSS, July 28, 2011
Permanent Scatterers analysis : displacement series
16:48 19:12 21:36 00:00 02:24 04:48 07:12 09:36 12:00 14:24-2
-1.5
-1
-0.5
0
0.5
1
1.5
2Multi-target displacement
Acquisition time [hh:mm]
[mm
]
17-Sep
PSgr-15597
Page 58Vancouver, IGARSS, July 28, 2011
Conclusions for the Ground Based SAR (15 h. observation)
The data show coherence > 0.8
The atmospheric effect is very low for the good meteorological conditions
The rms noise is of the order of 0.1mm
Page 59Vancouver, IGARSS, July 28, 2011
What happens if we are far away from PS? we have to estimate the APS from other
sources:
Numerical Weather Predictions
Meris
GPS
Page 60Vancouver, IGARSS, July 28, 2011
InSAR vs NWP
T351, desce, 10UTC
30 images
10 std IWV maps
T172, asce, 21UTC
41 images
20 std IWV maps
The Rome dataset
Page 61Vancouver, IGARSS, July 28, 2011
InSAR vs NWP
Spectral decorrelation
Long spatial wavelengths have a few hours correlation,
short spatial wavelengths decorrelate in less than 1 hour
Page 62Vancouver, IGARSS, July 28, 2011
InSAR vs NWP
Analysis and separation of the stationary (layered) delay
Layered delay Differential delay
Page 63Vancouver, IGARSS, July 28, 2011
MM5 vs InSAR, Rome MM5 vs InSAR, Rome T172 T172 9pm asce
Prediction of the change with height
Page 64Vancouver, IGARSS, July 28, 2011
InSAR vs NWP
MM5:Positive in retrieving the change with heightVery low correlation of turbulent termsStrong dependence on the starting time
WRF performs better than MM5
Page 65Vancouver, IGARSS, July 28, 2011
Meris vs InSARMeris vs InSAR
Cloud coverage can be a problem
Rome T351, morning passes
Page 66Vancouver, IGARSS, July 28, 2011
Meris vs InSARMeris vs InSAR
T351, Meris vs APS power spectra
Similar frequency content after removing the stationary component
Page 67Vancouver, IGARSS, July 28, 2011
Meris vs InSARMeris vs InSAR
And the accuracy is not enough…
Meris IWV [mm]
InSAR IWV [mm]
Page 68Vancouver, IGARSS, July 28, 2011
Meris vs InSARMeris vs InSAR
In desert areas Meris shows good performances (Li et al)
Meris has higher resolution and spectral content than MM5
But in Rome the correlation with InSAR APS is too low
Page 69Vancouver, IGARSS, July 28, 2011
GPS vs InSARGPS vs InSAR
The Como test-site
480 descending, 10am (28 images) 487 ascending, 9pm (38 images)
Page 70Vancouver, IGARSS, July 28, 2011
GPS vs InSARGPS vs InSAR
Delay,Como experiment,descending track
Different ways for estimating the stationary term, for different stations (color)
Page 71Vancouver, IGARSS, July 28, 2011
Ascending Descending
GPS std
InSAR std
diff std
corr coeff
GPS std
InSAR std
diff std
corr coeff
3.69 3.37 3.43 0.53 3.43 1.72 2.85 0.56
2.34 4.27 4.27 0.28 2.73 5.26 3.52 0.79
3.61 5.00 1.98 0.95 0.63 3.35 3.03 0.59
2.94 3.36 1.06 0.95 5.36 6.07 2.79 0.89
2.18 1.15 1.27 0.89 1.67 1.83 2.05 0.32
GPS vs InSARGPS vs InSAR
Best performances, ~50% success
Page 72Vancouver, IGARSS, July 28, 2011
Closer to any PS, the estimate of the APS improves: in a circle of PS with diameter D (m), the mean square APS reduces q times, q is lower than
1000036.0max
Dq
D
qmax
q
Page 73Vancouver, IGARSS, July 28, 2011
In the case of interferogram chaining, if there is a limited coherence γ from one take to the next, the measured phase
is:
2
2
1Ttarget, 1
2
22
4w
L
Nw
NTvatm
eq
L = number of looks (> 4); N = number of takes in the observation timeσatm= dispersion of the APS;w1, w2 noises with unit varianceSignal and noise grow with the number of takes
Even if γ =0.3, just 6 looks are equivalent to a PS, as the coherence is limited only by APS and not by decorrelation.
Page 74Vancouver, IGARSS, July 28, 2011
MM5 has a strong “random” component and a more stable stratification estimation. WRF performs better, further work is needed.
Meris has shown positive results in flat deserted areas, but not in our case studies.
GPS has a success rate of 50%.
The connection between InSAR and the other methodologies lies in the stationary term estimation.
Permanent Scatterers (or interferogram chaining) are still the best way to accurately estimate the APS.
Page 75Vancouver, IGARSS, July 28, 2011
The competition:Optical and GPS levelling:
Approximate results
Page 76Vancouver, IGARSS, July 28, 2011
A recent survey comparison for an accelerator design in Japan
Page 77Vancouver, IGARSS, July 28, 2011
Page 78Vancouver, IGARSS, July 28, 2011
Page 79Vancouver, IGARSS, July 28, 2011
Page 80Vancouver, IGARSS, July 28, 2011
Page 81Vancouver, IGARSS, July 28, 2011
Photon-counting detector with an accuracy of 20 ps (3.3 mm two way) Max point rate ~ 1000 pts/sec. Low atmospheric effects
Photon counting devices
Page 82Vancouver, IGARSS, July 28, 2011
The future?
Page 83Vancouver, IGARSS, July 28, 2011
Recent, forthcoming, and proposed satellites:
- X band: 4xCSK, 2xTSX, Kompsat, TSX, 2xCSK2
- C band: Rsat2, Sentinel - 1 A/B; 3xRCM
- L band: 2xSAOCOM, Palsar 2, Desdyni, TSL
- The proposal for a geosynchronous SAR (EOPUS)
Page 84Vancouver, IGARSS, July 28, 2011
C band data have shown their effectiveness
X band data are good for urban applications
L band data are useful for forest studies
P band data, only, reach ground (tomography)
Remarks
Page 85Vancouver, IGARSS, July 28, 2011
GEOsynchronous SAR for Atmosphere and Terrain observation
Prof. A. Broquetas – UPC Univesitat Politècnica de Catalunya, Electromagnetic and Photonic Eng. Group
Dr. D. D’Aria - ARESYS
Prof. N. Casagli, G. Righini – Università degli studi di Firenze, Dept. of Earth Sciences
Prof. S. Hobbs – Cranfield University, Cranfield Space Research Centre
Prof. A. Monti Guarnieri, Prof. F. Rocca – Politecnico di Milano, Dept. of Electronic and Inf. Science
Prof.sa R. Ferretti – University of L’Aquila, CETEMPS
Prof. M. Nazzareno Pierdicca – Sapienza University of Roma, Dept. of Electronic Engineering
Prof. G. Wadge – University of Reading, Environmental Systems Science Centre
Prof. Dr. H Rott – University of Innsbruck
Dr. C. Svara, Dr. A. Torre – Thales Alenia Space Italia
Earth Explorer EE8 proposal: COM3/EE8/32 GEOSAT
Page 86Vancouver, IGARSS, July 28, 2011
Geosynchronous satellite
The geosynchronous concept is as old as SAR [K. Tomiyasu, 1978]; however, it has never been realized.
To make it feasible we might use:
The 2.4 kW - TWT developed for CoreH20 Long integration times (up to 12 hours), stable targets, PS Quick looks every 20’ to measure and compensate the APS
Page 87Vancouver, IGARSS, July 28, 2011
The problem: water vapor fluctuations
Water vapor changes defocus the scene, unless controlled and compensated. Examples and statistical analysis made by:
Terrasar-X (PSINSAR), GPS, Ground based radar, MM5
Page 88Vancouver, IGARSS, July 28, 2011
GEOSAR for Atmosphere
Mediterranean Hurricane ECMWF medium-scale maps
Yielding high spatial (200 x 10 m) and temporal (20’) resolution water-vapor maps for:
- Weather forecast
- Correction of tropospheric delays for GPS and SAR
Page 89Vancouver, IGARSS, July 28, 2011
GEOSAT: The Wide Beam
GEOSAT may be a guest payload on Italian Space Agency (ASI) SIGMA missions, placed appox 9o longitude.
Look direction would then be close to SN for Europe.
Backgound mission
wide beam over central Europe (2000 km).
NESZ = -19 dB, 0.5x 0.5km resolution,
20’ revisit.
Fine resolution 10 x 10 m (twice daily revisit).
Applications: WV maps, glaciers, urban.
Page 90Vancouver, IGARSS, July 28, 2011
GEOSAT coverage: the SPOT beamGeoseismic risk in Europe
Number of Lanslides in Europe
A 700 km SPOT beam centered on Naples, would be used for monitoring:
- MEDIterranean hurriCANE and Water vapor (thus extending southward the wide beam),
- Many many landlsides
- Active volcanoes
Page 91Vancouver, IGARSS, July 28, 2011
Any TV antenna becomes a good reflector
80 cm antenna SNR = 21 dB (12 hours); SNR=2 dB (10 mins)
Number of home satellite
antennas
1999
millions
2002
millions
1999-2002
Increase
% total
population
% millions % millions
and Pacific 19.5 17.5 -11 -2.1 1 1
and US 13.7 20.1 47 6.4 4 6
EUROPE 33.8 43.6 26 8.7 4 5
Latin America and the 1.6 2.7 62 1.0 0 1
North Africa and the 8.9 11.9 29 2.6 3 4
0.0 0.0 177 0.0 0.0 0
Sub Saharan 0.4 1.2 83 0.3 0 0
World 77.9 96.8 22 16.8 1 2
47 Millions of users parabolas in Europe (2002)
Page 92Vancouver, IGARSS, July 28, 2011
Page 93Vancouver, IGARSS, July 28, 2011
GEOSAT could provide short time (20 min to hours) monitoring for:
atmospheric and hydrogeological effects volcanoes and glaciers
It would have:
wide-swath and SPOT beams
complementarity to LEO for revisit, look angle
compatibility with Telecom, as a guest payload.
Page 94Vancouver, IGARSS, July 28, 2011
Conclusions For ground motion applications, spatial resolution is not essential, but short
revisit time and good vertical precision are paramount
With a dense and long set of PS (desert areas), we are already much below the millimeter error. In the future, NWP and GPS may help.
The geosynchronous satellite could help to obtain continuous observation.
Optical and GPS systems, at the moment, do not offer either spatial continuity or sufficient precision.