The STUN algorithm for Persistent Scatterer Interferometry

Transcription

1 [1/27] The STUN algorithm for Persistent Scatterer Interferometry Bert Kampes, Nico Adam 1. Theory 2. PSIC4 Processing 3. Conclusions

2 [2/27] STUN Algorithm Spatio-Temporal Unwrapping Network (STUN) 4 1D temporal + 2D spatial phase unwrapping Goal: 4 Unwrap the phase in a single-master stack 4 Optimal estimation of displacement parameters Key Features: 4 Integer Least-Squares (ILS): estimator 4 Variance Component Estimation (VCE): weights 4 Alternative Hypothesis Tests: robust

3 [3/27] STUN Processing Steps Point selection: 4 ~4 PS/km 2 in reference network 4 ~200 PS/km 2 for estimation 4 Discard ~99% Reference Network Computation 4 Optional interferogram trend correction Tie more points to network Explicit phase unwrapping 4 Sparse grid Minimal Cost Flow (MCF) Optional Atmospheric Correction 4 Kriging Interpolation Final Estimation

4 [4/27] Integer-Least Squares (ex. 1) 16 samples, 10 years Noise σ = 69.2 [deg] Signal = [rad/y] ESTIMATED: 67.7 [deg] (γ=0.55) [rad/y] 2nd BEST FIT: (γ=0.41) [rad/y]

5 [5/27] Integer Least-Squares (ex. 2) 14 samples σ = 32.3 b1=5.79 b2=1.35 ESTIMATED: 27.1 [deg] (γ=0.91) nd BEST FIT: (γ=0.77)

6 [6/27] Integer Least-Squares Readily extendible for more parameters: 4 Search of ambiguities solution space 4 Efficient search strategy exist (GPS application) 4 No increase in computation time Weighted least-squares: 4 Stochastic model for double-difference phase observations 4 Variance Component Estimation (VCE) Software available at Delft University of Technology 4

7 [7/27] Variance Component Estimation (ex. 1) 31 SLC images 30 interferograms 400 PS points 200 arcs (doubledifferences)

8 [8/27] Variance Component Estimation Weights of the SLC scenes 4 Improves quality of estimated parameters 4 Reduces number of incorrectly estimated ambiguities 4 Automatically detect incorrectly processed interferograms 4 Realistic quality description of estimates Iterative estimation procedure See paper for equations

9 [9/27] Real data application

10 [10/27] PSIC4 Study Initiated by ESA at FRINGE 2003 Cross-Comparison of Persistent Scattering Processing Techniques

11 [11/27] Processed Area 25 x 40 km 2 Rural area Mountainous: m Subsidence due to mining

12 [12/27] Baseline Distribution SLC selected time m 1000 m Perpendicular Baseline No extreme Doppler/ large Baseline

13 [13/27] Processed Interferograms Differential Interferograms 4 SRTM DEM Single Master ERS-2 4 March 1999 Coregistration 4 Geometry 4 Point Targets Sorted according to perpendicular baseline

14 [14/27] Selected Points Area: rg: 2400 az: ~50 million pixels Points: SCR > 1.5 ~200,000 PS Phase data extracted at sub-pixel peak positions

15 [15/27] Variance Component Estimation Average of estimated components at ~600 independent arcs σ φ [deg] temporal baseline [years] SLC sigma ~ [deg] Accounts for random noise and atmospheric difference signal at arcs of typical length (1250 m)

16 [16/27] Reference Network Points in reference network selected based on amplitude dispersion index: 4 Expected to be temporally coherent Network constructed 4 ~10 arcs per point At all arcs, estimate: 4 DEM error differences 4 Displacement rate differences Integer least-squares Estimator 4 Weighted

17 [17/27] Parameter Integration Least-squares adjustment of estimates between PS points Yields DEM errors and Displacement rates at the PS points Alternative Hypothesis Tests Red: rejected arcs

18 [18/27] Parameters at Reference Network Reference network ~1600 PS DEM error Displacement Rate

19 [19/27] Estimated Parameters at PS 60,000 PS accepted Subsidenc e DEM update Displacement Rate -13 mm/y Uplift +5 mm/y

20 [20/27] Estimated Quality A posteriori variance factor Unwrapped data 4 Not yet corrected for atmospheric signal Precision decreases the further away from reference point (asterisk) Subsidence area: this factor is locally larger: 4 Functional model not correct?

21 [21/27] Residual Phase Residual phase in interferogram 4 DEM error corrected 4 Displacement rate This is interpreted as 4 Random noise + 4 Atmospheric signal Kriging Interpolation

22 [22/27] Structure Functions Each panel shows the structure function of the residual phase in an interferogram. Atmospheric signal: 4 power-law 4 slope in loglog plot Red: estimated slope 4 input for Kriging

23 [23/27] Kriging Interpolation Residual Phase Kriging

24 [24/27] Final Estimation mm/year Data corrected for estimated atmospheric signal

25 [25/27] GIS Interface (geotiff)

26 [26/27] Conclusions

27 [27/27] Conclusions STUN = Spatio-Temporal Unwrapping Network 4 Integer Least-Squares 4 Variance Component Estimation 4 Alternative Hypothesis Tests PSIC4 Processing Report 4 Point Selection 4 Reference Network 4 Unwrapping Our paper gives more details on theory and displacement models Visit our Poster: 4 DLR s Results of the PSIC4 Study

28 [28/27] Thank you!