interferometric and geodetic validation of sentinel-1
TRANSCRIPT
CEOS Calibration and Validation Workshop 2015
Interferometric and GeodeticValidation of Sentinel-1
experiences from the 1st year of operation
Yngvar Larsen
Tom Rune LauknesNorut
Zbigniew PerskiPGI
October 27-29, 2015, at ESA-ESTEC
John DehlsNGU
Petar MarinkovicPPO.labs
2 / 56
Motivation: Sentinel-1 Quality Control
Recurrent research question in field of SAR/InSAR:
What is the quality of Sentinel-1 sensor for different monitoring applications?
Sub-question:
“How good is what we measure/estimate?”
Context:
“Full scale error propagation for SAR/InSAR not straightforward...”
3 / 56
Objective
We need empirical error bars
“Error bars are a representation of the variability of data, or uncertainty in a reported measurement. They give a general idea of
how precise a measurement is, or conversely, how far from the reported value the true (error free) value might be. [Wikipedia]”
4 / 56
Objective
This is a workshop we are presenting →
work in progress!
5 / 56
Presentation Structure
● Context
● Gentle introduction to Test Sites
● Validation studies:
Part #1: Geo-Localization (amplitude)
Part #2: InSAR validation (phase)
● Summary, recommendations & future plans
6 / 56
Validation in context of...
MethodologiesSAR and InSAR applications
SAR SystemsSentinel-1, TerraSAR-X, Radarsat-2, etc.
Ground equipmentlarge number of CRs deployed
In-situ measurementsGNSS, leveling, inclinometers, weather stations, ...
Test sites
Poland, Norway, 'open sourced ones' (eg Australia), ...
7 / 56
Validation in context of...
MethodologiesSAR and InSAR applications
SAR SystemsSentinel-1, TerraSAR-X, Radarsat-2, etc.
Ground equipmentlarge number of CRs deployed
In-situ measurementsGNSS, leveling, inclinometers, weather stations, ...
Test sites
Poland, Norway, 'open sourced ones' (eg Australia), ...
8 / 56
Sentinel-1 Validation: General Project Setup
● A number of CR validation sites 4 in total→● Three sites in Poland● One site in Norway
● CR's monitored by in-situ (GPS and leveling) and other SAR● Different CR design/size● Different design of control networks
● Specifically, here we will focus on:● Two sites in Poland● One site in Norway● Sentinel-1 validation● As a reference:
– SAR system TerraSAR-X→– GNSS campaigns over test-sites
9 / 56
Sentinel-1 Validation: General Project Setup
● A number of CR validation sites 4 in total→● Three sites in Poland● One site in Norway
● CR's monitored by in-situ (GPS and leveling) and other SAR● Different CR design/size● Different design of control networks
● Specifically, here we will focus on:● Two sites in Poland● One site in Norway● Sentinel-1 validation● As a reference:
– SAR system TerraSAR-X→– GNSS campaigns over test-sites
10 / 56
Poland Test Sites: Overview
● ~111 concessions for shale gas exploitation granted in PL
● Possible subsidence expected
● Monitoring required by law
● 3 sites defined for validation and methodology development
● Projects in numbers
– Duration: Jan 2013 – Dec 2016
– Number of CRs deployed: 60
– TSX, S-1, R-2
– In-situ: GNSS, leveling, weather
11 / 56
Poland Test Sites: Overview
12 / 56
CR Design: Babiak, PL site
13 / 56
CR Design: Babiak, PL site
0.7m – 1.7m
0.6m – 1.3m
12 deg
14 / 56
Norwegian Test Site: Summary
● Long term landslides monitoring (maximum 5 cm/yr LOS)
● 4 ‘Big’ CRs & 10 'Small' TSX optimized
● ‘Big’ CRs collocated with cnts GPS
● Up to 500 meters of height difference between CRs
● InSAR ongoing since ~2008
● Equipped with the snow protection● Winter scenes due to snow could be
quite problematic
15 / 56
Norwegian Test Site: Summary
16 / 56
Norwegian Test Site: Summary
2.5 m
“symmetric trihedral”
Validation Part #1:
Absolute Geo-Localization (Amplitude)
18 / 56
Methodology and Remarks
Semantics & Definitions
Offset := predicted - estimated
Assumptions
● Methodologies for peak extractions, residual bistatic correction, atmospheric contributions, datum transformations are all considered known.
SAR and auxiliary data
● Standard Level 1 products and metadata● Precise orbits● One year of SAR data (Oct'14 – Oct'15)
Focus on results!
Babiak
Lewino
150km
PL Sites in S-1 slice
Babiak Test Site
Babiak Test Site
Babiak Test Site
4km
4km
Quick stats of observation setup:
● 2 x (15 +5) CRs observed over period of 1 year
● Observation period 1+ years (mid 2014 - today)
● TSX SM: 2 x 35
● S-1 IW: 1 x 25
26 / 56
TSX Results: internal validation
● To establish the reference, and validate CR coordinates
● Out of box geolocalization - only annotated parameters
● Impact of datum correction
● NO DYNAMIC (EARTH TIDES) CORRECTIONS APPLIED
27 / 56
TSX GeoLocalization: Out-of-Box
Rng mean/stddev [m]: -0.3056 0.0924
Azi mean/stddev [m]: -0.5359 0.0919
Rng mean/stddev [m]: -0.1893 0.0982
Azi mean/stddev [m]: -0.5981 0.1520
28 / 56
TSX GeoLocalization: Datum Correction
Rng mean/stddev [m]: 0.06433 0.0924
Azi mean/stddev [m]: 0.00401 0.0919
Rng mean/stddev [m]: 0.1807 0.0982
Azi mean/stddev [m]: -0.0581 0.1520
Datum correction: ETRS89 ITRF2008→
(No dynamic correction larger stddev)→
29 / 56
S-1 GeoLocalization: Out-of-box
30 / 56
S-1 GeoLocalization: Out-of-box
● Two clusters because of Center-of-Gravity correction
● Before/after May'15
31 / 56
S-1 GeoLocalization: Out-of-box
Bistatic difference:● correction applied in processor wrt center of the swath
32 / 56
S-1 GeoLocalization: With corrections
Corrections:● APS model by A.Schubert / M.Jehle et al (UZH team)● Bistatic refined● CoG for old scenes (before 2015-05-01)
33 / 56
S-1 GeoLocalization: With corrections
Rng mean/stddev [m]: -0.2801 0.1376Rng mean/stddev [m]: -0.2801 0.1376
Azi mean/stddev [m]: 3.0671 0.6749Azi mean/stddev [m]: 3.0671 0.6749
Rng mean/stddev [m]: -0.2501 0.1475
Azi mean/stddev [m]: 2.2986 0.6826
34 / 56
S-1 GeoLocalization: With corrections
Rng mean/stddev [m]: -0.2801 0.1376
Azi mean/stddev [m]: 3.0671 0.6749
Rng mean/stddev [m]: -0.2501 0.1475
Azi mean/stddev [m]: 2.2986 0.6826
● axes now 1:5 for rg:az
● the same as for the resolution difference
● a circular point cloud is expected (and observed!)
35 / 56
S-1 GeoLocalization: With corrections
Rng mean/stddev [m]: -0.2801 0.1376
Azi mean/stddev [m]: 3.0671 0.6749
Rng mean/stddev [m]: -0.2501 0.1475
Azi mean/stddev [m]: 2.2986 0.6826
We observed ~0.7 meters offset between clouds in azimuth direction!?
36 / 56
S-1 GeoLocalization: CR in Swath Overlap● In Norway we have CR in IW1/IW2 swath overlap
NB: We were very lucky!
37 / 56
S-1 GeoLocalization: CR in Swath Overlap
Difference between apparent CR position in IW1/IW2
38 / 56
S-1 GeoLocalization: Issues Discussion
● The same CR observed in two swaths
● Discrepancy in measured peak positions in IW1/IW2
● Offset difference
● Range:● Delta time := 0.8 sec almost the same atmosphere→● Definitely there's no 20 cm APS difference as observed
● Unexplained cause?
● Azimuth:
● Unexplained average of 1.6 meters
● Misalignment between swaths?
● Timing jitter reported & fixed, also values too small to explain this
IPF issue ?
39 / 56
S-1 GeoLocalization: IPF issues?
IPF issue ?
40 / 56
S-1 GeoLocalization: IPF issues?
Wild theory: what if bistatic correction in the processor NOT wrt mid-IW2, but wrt mid-swath for each swath individually?
Validation Part #2:
InSAR Quality Control (Phase)
Methodology: InSAR validation
● Closed loop validation of S-1 vs TerraSAR-X● Cross validation of double-differences● 'Raw' phases: no-unwrapping, no corrections,
preliminary qualified products● Reduced network size: working on single-arcs● On-going work
Time period: ● Oct'14 – Oct'15
Data:● S-1: 25 x IW @33 deg● TSX: 35 x SM @36deg
Network Design: Babiak, PL test site
Network Design: Babiak, PL test site
~ 200 m
Results: InSAR validation
TSX @36
S-1 @33
S-r09
T-r15T-r16
Results: InSAR validation
TSX @36
S-1 @33
Outlier: Snow in CR ?
Results: InSAR validation arc #1
TSX @36
S-1 @33
TSX stddev [mm]: 0.4873
S1 stddev [mm]: 0.6829
● 'raw' InSAR double differences● No in-processing outlier removal● Results still being refined● Dealing with APS● STDDEV wrt a smoothed version
to get rid of signal and outlier
S-r09
T-r15T-r16
Results: InSAR validation arc #2
TSX @36
S-1 @33
TSX stddev [mm]: 0.5622
S1 stddev [mm]: 0.6805
S-r09
T-r15T-r16
Results: InSAR validation arc #3
TSX @36
S-1 @33
TSX stddev [mm]: 0.6025
S1 stddev [mm]: 0.7674
S-r09
T-r15T-r16
Norway: InSAR vs Continuous GPS @ Landslides
Norway: InSAR vs Continuous GPS @ Landslides
Unfiltered S-1 InSAR time-series
Norway: InSAR vs Continuous GPS @ Landslides
Filtered / smoothed S-1 InSAR time-series
Norway: InSAR vs Continuous GPS @ Landslides
Possible InSAR processing issues?
Ionosphere?
Filtered / smoothed S-1 InSAR time-series
Summary
● Different quality aspects of S-1 independently validated
● GeoLocalization accuracy of S-1 (ballpark)● Azimuth: ~2.5 m (with stddev of 0.7m)● Range: -0.3 m (with stddev of 0.14m)
Potential issues of swath misalignment observedHopefully(?) our findings are in an agreement with other teams?
● InSAR validation of S-1● Phase STDDEV in the ballpark of 0.7 mm
Should be ~1mm We got it in the controlled environment!→
Anyone else validating S-1 InSAR and willing to share findings?
Plans and future work
● This is an on-going validation project:
Updates at upcoming conferences and workshops
● Making work public and open source:
Plan is that parts of this work will eventually be open-sourced (algos and code) and/or results made available in coordination with funding agencies.
● Current focus on:● Assembling and interpreting of results● InSAR time series processing and validation
Acknowledgments:● Copernicus Programme, all Sentinel-1 results: “Contain modified Copernicus Sentinel
data (2014/2015)”● ESA SEOM Programme – InSARap Study● Polish Ministry of Environment● Norwegian collaborators: NVE (Norwegian Water Resources and Energy Directorate)● Nuno Miranda, ESA, for helpful pointers @Fringe15 on the initial validation results● Eelco Doornbos and Wim Simons of TU Delft (Astrodynamics and Space Missions
Department) for their clarifications of Sentinel-1 orbital dynamics
Plans and future work
● This is an on-going validation project:
Updates at upcoming conferences and workshops
● Making work public and open source:
Plan is that parts of this work will eventually be open-sourced (algos and code) and/or results made available in coordination with funding agencies.
● Current focus on:● Assembling and interpreting of results● InSAR time series processing and validation
Public version of this talk available on SlideShare
Easier to find it via my twitter stream: twitter.com/pmar