Transcript Experiments with Time-series InSAR algorithms

Time-series InSAR with DESDynI:
Lessons from ALOS PALSAR
Piyush Agrama, Mark Simonsa
and Howard Zebkerb
aSeismological
Laboratory, California Institute of Technology
bDepts of EE and Geophysics, Stanford University
Motivation
• InSAR time-series techniques crucial for many of
DESDynI’s stated objectives - Geohazards,
Hydrology and subsurface reservoirs
• Why ALOS PALSAR?
– L-band mission similar to DESDynI.
– Lifetime similar to DESDynI.
• ALOS PALSAR products – a good proxy for
DESDynI products.
Overview
• Noise levels at L-band vs C-band.
• Topography related artifacts
• PS-InSAR at L-band
• Novel time-series techniques: MInTS.
Comparison of Noise Levels
• Typical resolution of interest – 100m x 100m.
• Analysis of filtered interferograms with
shortest time span.
• ERS vs ALOS PALSAR.
• Experiments conducted over the San
Francisco Bay Area.
L-band 46 day correlation similar to C-band at 1 day
ALOS PALSAR
ERS Tandem
• ERS Looks = 80.
ALOS Looks = 336.
Bperp
(in m)
Phase
noise
( mm)
Bperp
(in m)
Phase
noise
(mm)
– Factor of 2 gain.
410
8.5
275
6.4
98
6.1
113
7.8
159
16.5
94
4.0
932
18.1
-
-
102
7.1
-
-
39
17.2
-
-
Average
12.3
Average
6.1
• Factor of 2 observed in
InSAR data.
• L-band Decorrelation
at 45 days ~ C-band
decorrelation at 1 day
Areal coverage similar.
ALOS coherence threshold = 0.7 .
ERS coherence threshold = 0.7.
L-band 46 day correlation 2x C-band at 35 days
ALOS PALSAR
ERS (35 day)
• ERS Looks = 80.
ALOS Looks = 336.
Bperp
(in m)
Phase
noise
( mm)
Bperp
(in m)
Phase
noise
( mm)
– Factor of 2 gain.
410
8.5
670
15.2
98
6.1
325
11.0
159
16.5
498
15.7
932
18.1
376
11.5
102
7.1
1010
N/A
39
17.2
200
10.5
Average
12.3
Average
12.8
• Factor of 2 gain in
phase noise due to
coherence threshold.
• Temporal decorrelation
at L-band is significantly Areal coverage similar.
lower.
ALOS coherence threshold = 0.7 .
ERS coherence threshold = 0.4.
L-band vs C-band
Temporal correlation
Phase noise (mm)
C-band
L-band
L-band
C-band
• Decorrelation noise higher at C-band for longer temporal
baselines.
• Other noise sources - atmosphere etc. are assumed to
be on the same order at both C and L bands
Implications for DESDynI
• Lower temporal decorrelation for many interferograms
favors L-band.
– More redundant IFG networks for time-series.
– More coherent IFGs with longer time spans than C-band
• Reduced temporal decorrelation improves the spatial
coverage significantly (for same coherence threshold).
• Improved coherence => Better unwrapping.
• Overall: Comparable sensitivity to C-band time-series
InSAR products but with greater spatial coverage for
rapid interferograms, much better for longer time spans.
ALOS and topo-related errors
• Due to orbit drift, correlation between Bperp
and temporal baseline is 0.7.
• DEM error cannot be distinguished easily from
deformation features (SBAS).
Parkfield, CA
PS-InSAR at L-band with ALOS
• Not as straight-forward as at C-band due to sensor
management.
• ALOS PALSAR – Need to combine different imaging
modes.
• Different noise characteristics of FBD and FBS modes.
• Need appropriate weighting of the modes when
selecting PS.
• Does work: example over Long Valley Caldera, CA.
Long Valley Caldera
PS pixel mask for ALOS PALSAR
•
•
•
•
C-band image from Hooper et al (2004)
23 ALOS PALSAR images with baselines < 4 Km.
PS density is similar to C-band.
Fine tuning needed for handling different modes.
Velocities heavily contaminated by topo-related errors.
Implications for DESDynI
• Plan no systematic relationship between temporal and
spatial baselines.
• L-band allows us to implement simple SBAS/ linear
inversion approach more reliably due to better
coverage.
• Other topo-related errors - like tropospheric delay - at
same level as ALOS PALSAR.
• Traditional time-series approaches like SBAS and PSInSAR should work better for DESDynI than ALOS.
Novel time-series techniques will improve
over current methods
• In many situations, deformation estimates at 500m x
500m suffices to model geophysical phenomenon.
• Can exploit the spatially correlated nature of
deformation at these spatial scales.
• Can decompose the data into independent
components at various spatial scales- e.g, wavelets.
• Multiscale InSAR time-series (MInTS) developed by
Hetland and Simons.
Multiscale InSAR Time Series (MInTS)
Create Interferograms
Unwrapped phase
Coherence
Time-series products
Create data mask for each
IFG and interpolate holes
Reconstruct data using
inverted coefficients
Compute wavelet
coefficients and Weights
for each IFG
Invert wavelet coefficients
using temporal model
(similar to GPS)
MInTS results at Parkfield
Parkfield, CA
• Resolution of 200m x 200m.
• Same stack of 84 IFGs used for SBAS and MInTS.
• Linear velocity and sinusoidal seasonal terms estimated.
Conclusions
• Shorter repeat period and acquisitions in a consistent
imaging mode over targets make DESDynI superior to ALOS
PALSAR.
• Better orbital control and plan significantly decreases
uncertainties in deformation estimates due to topo-related
errors.
• Uncertainty in deformation time-series will match current Cband products but yield much greater spatial coverage.
• Novel time-series techniques like MInTS can significantly
improve deformation estimates over regions where traditional
techniques like SBAS and PS fail.