InSAR practical considerations

Transcription

1 InSAR practical considerations How to get data Understanding the Line of Sight Sources of error Resampling/downsampling data Multi-interferogram methods: time series PSInSAR Matthew Pritchard Cornell University

2 How to access data: Overview of available data sources & most up-to-date info is at roipac.org/accesstodata A tool to search for available ERS and Envisat data is freely available for download from ESA: called EOLISA If you would like information on baselines between potential datatakes, visit roipac.org No Cost data -- but need to apply for access ESA data to become available for no cost: details still being worked out Supersites available for small areas of interest (Mt. Etna, Los Angeles, etc.) see roipac.org for link GeoEarthscope: If at a research institution, write a Mini-Cat proposal (ERS, Envisat, Radarsat over North America) Alaska Satellite Facility ALOS datapool -- if a WInSAR member, sign a form and get access ERS/JERS within ASF satellite disk mask, Radarsat almost anywhere in the world -- write a 1-page proposal Western North America InSAR Consortium (ERS, Envisat over North America) Your institution must be a member (no cost to join!), need password For purchase Commercial companies (usually $100 s to $1000 s per scene) Access to software: InSAR: Overview of available packages focusing on the open source ROI_PAC Notes from UNAVCO short course on InSAR & ROI_PAC:

3 At roipac.org/modeling Visualizing the data & preparing for modeling Perl script: unw2png.pl will convert geocoded interferogram to Google Earth kml file and png (need free imagemagick software) Various GMT scripts Example Matlab script for loading data For GIS software, use.rsc file to create metadata file and perhaps rmg2mag_phs to make a single binary file Make los script: creates file with the satellite heading and incidence angle at each pixel

4 Access to topographic data Source I use most often: SRTM 1 degree tiles either at 3 arcsec (90m) or 1 arcsec for U.S. (30 m) Roipac.org/ContribSoftware get_srtm.pl and makedem.pl Also available from seamless USGS server. Can access available Lidar and NED data for U.S. ASTER GDEM -- posted at 1 arcsecond, but various analysis indicates it is closer to 3 arcsec Format needs to be I*2 (16 bit) binary file with no header Need to create.rsc file with upper left coordinates, # of rows and columns, and pixel spacing

5 Where in the world am I? Magnitude 6.6 earthquake: 26 December 2003 in Bam, Iran Arid and mountainous region with frequent earthquakes (collision between Arabian and Eurasian plates) From: Farsinet.com North Previously unmapped fault (right-lateral strike-slip) Bam Baravat 20 km Interferogram courtesy of Yuri Fialko 10 km Landsat satellite image from 1999, from Funning et al., 2005

6 What am I looking at? Each fringe: contour of ground deformation in direction of satellite radar beam North Each scene: 20 meters per pixel 100 s of km per image Resolve deformation ~mm/year 20 km This example: From European space Agency Envisat satellite (5.6 cm radar wavelength) Each fringe is 2.8 cm of deformation

7 Visualizing 3D deformation in a 1D interferogram Step 1: Fault motion produces 3D deformation field Step 2: Project 3D deformation onto satellite radar line-of-sight Trade-off between horizontal and vertical deformation creates asymmetric pattern Both images: Funning et al., 2005 Step 3: Create a fringe every /2 centimeters ( wrapped image )

8 Reconstructing the full 3D deformation field Use interferograms from different satellite look directions PLUS: use the amplitude images to track pixels that moved Observed interferograms Inferred vertical displacement Observed pixel tracking Inferred horizontal displacement Before After Fialko et al., 2005

9 SAR Track/Frame Geometry Iran Modified from Rowena Lohman

10 Hector Mine EQ Descending Ascending Modified from Rowena Lohman

11 What are the sources of error? How do we evaluate them? Unwrapping errors: assess by looking a image with different wrap rates Atmospheric/ionospheric errors: use multiple images and pairwise logic (e.g., Feigl & Massonnet, 1995) Orbital errors: understand their basic characteristics, try different orbital estimates, process tracks of different lengths, tandem pairs can be useful DEM errors: inspect the raw DEM, process interferograms with different baselines and timespans, tandem pairs can be useful

12 Wrapped vs. Unwrapped Hector Mine EQ Color Cycle ~ 3 cm Color Cycle = 300 cm Modified from Rowena Lohman

13 Modified from Rowena Lohman

14 Modified from Rowena Lohman

15 Modified from Rowena Lohman

16 Unwrapped after masking Unwrapped before masking Modified from Rowena Lohman

17 Orbital Errors ( Ramps ) ~0.1-1 m uncertainty in satellite positions Orbital fringes not always 100% removed In particular, not sensitive to long wavelength deformation How to overcome? Simultaneously solve for position and geophysics Reported vs. Actual Modified from Rowena Lohman

18 Atmospheric contamination: Two types Turbulence Vertical stratification Both From: Pritchard & Simons, 2004

19 Can we remove the atmospheric signal from interferograms? 0) Use interferograms themselves to estimate linear or exponential phase with elevation: constant for image or spatially variable 1) Direct water vapor and dry delay observations: From satellite (e.g., Li et al., 2005) From GPS & other ground sensors (e.g., Webley et al., 2002) 2) Data stacks or APS: Assume atmosphere random in time or low-pass time domain filtering (e.g., Ferretti et al., 2001; Simons and Rosen, 2007) 3) Global and Regional Models computed by data center (~100 km horizontal resolution by ECMWF, NCEP; North American RR ~ 32 km) (e.g., Doin et al., 2007; Elliott et al., 2007) 4) Regional or Local Model computed by user (<3 km horizontal resolution) (e.g., Foster et al., 2006) Based on several studies, we can t remove everything. Will likely always need to account for atmosphere via covariance matrix

20 Ionosphere: C- and L- bands, spanning 4000 km Azimuthal streaking seen in polar regions (e.g., Joughin et al., 1996; Gray et al., 2000) ALOS: Dec Dec B-perp 1.1 km ALOS: Feb Aug m B-perp ERS: From: Pritchard, 2003 ALOS: Mar Mar m B-perp

21 Ionosphere Types of effects: 1) Broad phase delay (Ecuador example?) 2) Turbulent effects (scintillation or spread F: origin of Chile and Wenchuan examples?) 3) Faraday rotation From: NOAA From: Xu et al.,, 2004; originally from Aarons, 1982 Ionospheric corrective measures: 1) Throw out bad scenes 2) Split spectrum 3) Optimium time of day

22 Unexpected deformation can cause errors Vertical component of deformation from Southern California GPS station (in mm) Annual and sub-annual cycles Not a perfect sinusoid: Amplitude varies from year to year Presumably related to natural and human-induced groundwater changes From: Dong, JPL webpage

23 How we can be fooled by space geodesy? GPS: InSAR: Both: Reference frames Antenna environment: multipathing, antenna changes, cutting down trees atmospheric effects: (can mitigate with multiple interferograms, dense GPS, etc.) Ionospheric effects (mostly a problem near magnetic poles/equator; near dawn/dusk) Digital Elevation Model errors Orbital errors (hard to measure long-wavelength signals like post-glacial rebound) Contamination by signals that were not expected GPS stations measure contraction in LA Basin But interferograms reveal other deformation that can contaminate tectonic signal Both from: Bawden et al., 2001

24 Should we believe GPS/InSAR?: Part 1 How well do coincident measurements agree? Compare large earthquakes in South America: RMS different few cm 90 InSAR and GPS points for Mw 8.1 Antofagasta, Chile earthquake. GPS stations first occupied in 1992, so GPS was immature (Pritchard et al., 2002) 10 InSAR and GPS points for Mw 8.4 Arequipa, Peru earthquake. Only 4 different GPS stations included (Pritchard et al., 2007) For other earthquakes also agree to few cm: Landers, Northridge, Hector Mine (Massonnet et al., 1993, 1998; Zebker et al., 1994; Fialko et al., 2001; Jonsson et al., 2002)

25 Compare resampling methods The problem: reduce points to Compare: 1. Data resolution matrix method (Lohman & Simons, 2005) 2. Curvature based (Simons et al., 2002) 3. Quadtree (Jonsson et al., 2002) 4. Uniform Potential downsides: 1. Need to specify a model 2 & 3: Sensitive to noisy data 4. Need too many points to get near model detail From: Lohman & Simons, 2005

26 Multiple interferograms: Time series of interferograms Data available in southern California From: Yuri Fialko

27 New techniques: Time series of interferograms Possible pairs with Perpendicular baseline < 200 m From: Yuri Fialko

28 Strategies for combining multiple interferograms Prerequisite: Need to co-register interferograms either in radar or geographic coordinate. You can do this in ROI_PAC using process_2pass_master.pl by setting the Do_sim flag in the *.proc file A time-invariant view is stacking: Just take co-registered interferograms and add them together -- divide by the total time interval to get a rate A time-variable view is called time-series: including methods called SBAS, PSInSAR, etc. Advantages: Deformation is time-dependent separate signal & error (atmosphere, unwrapping, DEM)

29 New techniques: Time series of interferograms The Basic Idea Date

30 New techniques: Time series of interferograms The Basic Idea Date A stack of interferograms provides multiple constraints on a given time interval

31 New techniques: Time series of interferograms The Basic Idea Date Goal: Solve for the deformation history that, in a least-squared sense, fits the set of observations (i.e., interferograms), Many different methods (e.g., Lundgren et al. (2001), Schmidt & Burgmann, 2003), but SBAS (Berardino et al. (2002)) is perhaps most common one

32 Persistent scatterers (PS or PSInSAR) Select pixels with stable scattering behavior over time Long Valley Caldera, Hooper et al Only focus on good pixels InSAR Spatial 1 time Need neighborhoods of good pts PS 1 point Need > scenes Added bonus: DEM errors! Works with large baselines From: Rowena Lohman

33 StaMPS method (Hooper et al., 2004) Example: Imperial Valley, CA From: Rowena Lohman

34 What is the local rate of deformation? Review: Will InSAR work for you? Sensitivity of single igram ~1cm How many years to get signal this big and will it be overcome by noise? Can you stack several igrams together? What is the scale of deformation? Pixel size ~10m, but generally need to average many together Image size is ~100 km, but if too broad worry about precision of orbits What is the local noise? How much vegetation/precipitation/water vapor/human cultivation? Can you only make igrams with data from the same seasons? Can you get L-band data and find persistent scatterers? What data is available? Is there data from multiple satellites and/or imaging geometries? Is a digital elevation model available? Do you need rapid response for hazard assessment?

35 Review: How to set up InSAR capability? 1) Establish access to data Main sources: see next slide How? Can be purchased commercially. Lower cost/no-cost data available with restrictions. In Europe, through ESA. In U.S., through ASF and UNAVCO. Some foreign access is allowed to UNAVCO Can useful interferograms be made with available data? Worry about ground conditions, radar wavelength, frequency of observations, perpendicular baseline, availability of advanced processing techniques 2) Purchase/Install software to process and visualize data Open source: ROI_PAC, DORIS, NEST, RAT and IDIOT (TU Berlin) Commercial: Gamma, TR Europa, Vexcel/Atlantis, DIAPASON, SARscape 3) Download/create DEM (SRTM is only +/- 60 degrees latitude, but ASTER G-DEM to 89) 4) Download precise orbital information & instrument files (Only ERS & Envisat) 5) Interpret results, create stacks, time series, persistent scatterers. May need to buy/downoad/create new software 6) Publish new discoveries and software tools!

36 For More Information: Good overview of classical & space based geodesy (but no InSAR): John Wahr s online textbook Introductions to InSAR: 2 page overview from Physics Today Overviews of applications: Massonnet & Feigl, Rev. Geophys., 1998; Burgmann et al., AREPS, More advanced InSAR: The definitive SAR book: Curlander & Mcdonough, 1990 More technical reviews: Rosen et al., IEEE 2000; Hanssen s Radar Interferometry book, 2001; Simons & Rosen, Treatise on Geophysics, 2007; Time series analysis: Berardino et al., IEEE, 2002; Schmidt & Burgmann, JGR, 2003 Persistent scatterers: Ferretti IEEE, 2001; Hooper et al., GRL, 2004; Kampes Persistent Scatterers book, 2006