RADARSAT-2 Image Quality and Calibration Update

Transcription

1 RADARSAT-2 Image Quality and Calibration Update by Dan Williams, Yiman Wang, Marielle Chabot, Pierre Le Dantec, Ron Caves, Yan Wu, Kenny James, Alan Thompson, Cathy Vigneron

2 Image Credits and Disclaimer Language RESTRICTION ON USE, PUBLICATION OR DISCLOSURE OF PROPRIETARY INFORMATION AND IMAGES This document contains information and images that are proprietary to MacDonald, Dettwiler and Associates Ltd. ( MDA ), to its subsidiaries, and/or to third parties to which MDA may have legal obligations to protect such information or images from unauthorized disclosure, use or duplication. Any disclosure, use or duplication of this document or of any of the information or images contained herein is expressly prohibited. The statements contained herein are based on good faith assumptions and provided for general information purposes only. These statements do not constitute an offer, promise, warranty or guarantee of performance. The products depicted are subject to change, and are not necessarily production representative. Actual results may vary depending on certain events or conditions. This document should not be used or relied upon for any purpose other than that intended by MDA. COPYRIGHT 2015 MacDonald, Dettwiler and Associates Ltd., subject to General Acknowledgements for the third parties whose images have been used in permissible forms. All rights reserved.

3 Outline Overview of commercial SAR modes Newly-released maritime ScanSAR modes Image Quality Monitoring and Calibration Updates Overview Geolocation Accuracy Resolution Noise Level Radiometric Accuracy Beam Pointing / Inter-Wing Phase Polarimetric Corrections Other Conclusion 3

4 RADARSAT-2 Commercial SAR Modes

5 RADARSAT-2 Commercial SAR Modes 20 Beam Modes 15 Stripmap (Single Beam) 4 ScanSAR (2 to 8 Beams) 1 Spotlight (Steered Beam) Swath Widths km Nominal resolutions Range: ~3-100 m Azimuth: ~ m 5

6 Designed for improved ship detection performance Some visible noise banding Estimated relative radiometric accuracy ~1.5 db DVWF available only in HH polarization 6

7 Overview of Image Quality and Calibration RADARSAT-2 Image Quality and Calibration remains high priority for the mission IQ performance remains excellent Ongoing work to refine calibration and minimize artifacts Key aspect of introduction of a new mode is the calibration (e.g. Extra-Fine, MSSR modes, Wide Quad Pol Modes) Point Target Measurements (Corner Reflectors, Antenna Dishes) Distributed Target Measurements (Amazon) 7

8 Point Target Monitoring Corner reflectors MDA owned, located in Vancouver and Quebec City, Canada Others used CONAE (Argentina)* JAXA (Tomakomai, Japan) JPL (California) U of Zurich (Dubendorf, Switzerland)* DTSO (Adelaide Australia)* SFU (Bolivia) * temporary deployments X-Band and S-Band antenna dishes in Canada Gatineau, Prince Albert, Saskatoon, St-Hubert, Aldergrove, Masstown Higher radar cross-section than corner reflectors, but less stable and accurate Results need to be filtered to eliminate bad measurements Dish not tracking sensor Contamination from surrounding clutter Snow in reflectors in winter Ground truth accuracy limitations 8

9 Point Target Monitoring Results Geolocation accuracy has improved New enhanced definitive (post-processed) orbits provide orbit knowledge accuracy better than 1 m with 48-hour latency Small refinements made to geolocation calibration and SAR processing in 2015 Measured geolocation accuracy of point target monitoring results with Downlinked orbit data: <= 6 m RMS error in most Single-Beam and Spotlight modes <=10 m RMS in Extended Low mode <=30 m RMS in ScanSAR modes Excluding uncertainties in terrain elevation knowledge Improves with increasing incidence angle Processing with Definitive orbits improves accuracies further Resolution and sidelobe ratios are stable and consistent with expectations 9

10 Ground Range Geolocation Error (m) Ground Range Geolocation Accuracy (with Downlinked Orbit data, over Corner Reflectors) Stable since 2010 Standard Quad (Far) Standard Quad (Near) Fine Quad Extra-Fine Ultra-Fine Spotlight -20 1/1/2010 1/1/2011 1/1/2012 1/1/2013 1/1/2014 1/1/2015 1/1/

11 Azimuth Geolocation Error (m) Azimuth Geolocation Accuracy (with Downlinked Orbit data, over Corner Reflectors) Small bias corrected thanks to calibration and processing refinements during summer 2015 Standard Quad (Far) Standard Quad (Near) Fine Quad Extra-Fine Ultra-Fine Spotlight -20 1/1/2010 1/1/2011 1/1/2012 1/1/2013 1/1/2014 1/1/2015 1/1/

12 Geolocation Circular Accuracy The figures below summarize measured circular location errors as a function of incidence angle, with downlinked orbit data, Jan 1st 2012-Present <= 6m RMS in Single-Beam and Spotlight modes, for incidence angles >= 20 deg <= 20m RMS in ScanSAR modes (using special test products with refined pixel spacings that satisfy the Nyquist condition) Better than the conservative geolocation accuracy estimates in RADARSAT-2 Product Description Much better than original performance goals 12

13 Enhanced Definitive Orbit Validation of Geolocation Accuracy Example geolocation test results for RADARSAT-2 Corner Reflectors, after atmospheric correction in Spotlight and Ultra-Fine modes Old New Enhanced Definitive * * after reprocessing with latest bias corrections in SAR processor 13

14 Enhanced Definitive Orbit Validation of InSAR Performance Example of interferograms processed with the old definitive orbit (left) and the new enhanced definitive orbit (right) Achieved goal of <1 fringe per 100 km ground range 14

15 Ground Range Resolution Normalized to 35 o Incidence (m) Ground Range Resolution (SGX Products) Very stable (when normalized by incidence angle) /1/2010 1/1/2011 1/1/ /31/2012 1/1/2014 1/1/2015 1/1/2016 Wide Standard (Far) Extended High Standard Quad (Far) Standard (Near) Extended Low Standard Quad (Near) Fine Multi-Fine Wide Fine Fine Quad Extra-Fine Ultra-Fine Spotlight 15

16 Azimuth Resolution (m) Azimuth Resolution (SGX Products) Very stable 32 Wide /1/2010 1/1/2011 1/1/2012 1/1/2013 1/1/2014 1/1/2015 1/1/2016 Standard Extended High Extended Low Standard Quad Fine Multi-Fine Wide Fine Fine Quad Extra-Fine Ultra-Fine Spotlight 16

17 Elevation Pattern Gain (db) Gamma0 (db) Radiometric Accuracy Monitoring (1) The main method for monitoring absolute radiometry is through antenna elevation pattern analysis in Single Beam modes: Measure backscatter profile as a function of range over a homogeneous area of the Amazon rainforest Convert into a measured elevation pattern by subtracting noise, scaling by the assumed mean backscatter function of the Amazon, and backing out the elevation pattern correction applied during processing Align the measured pattern with the reference pattern from the calibration parameters, take the power ratio between the 2 patterns, and track it over time Amazon Reference Backscatter vs Incidence Angle Co-Pol Cross-Pol Incidence Angle (deg) Measured Pattern Derived from Amazon Scene Reference Pattern from Calibration Parameters Elevation Angle (deg) 17

18 Mean Radiometric Offset Overall results (for all beams) are gradually improving (ongoing calibration refinements since initial operations in April 2008) Mean difference from Amazon reference is nearly zero (~ -0.1 db on Ascending passes, db on Descending passes) Mean per-scene differences are typically within +/- 1 db (s = 0.3 db) 18

19 Noise Level Monitoring Data collected in receive-only modes Antenna receives but does not transmit Processed image contains only noise Acquisitions cover all pulse types used in commercial modes Processed image levels are analyzed and converted to estimates of noise levels in the raw data These estimates are compared to the calibrated noise level associated with each pulse 19

20 Noise Levels Results are stable since initial operations in April 2008 Measurements in all polarizations remain within ~1 db of calibrated levels 20

21 Median NESZ (db) NESZ Variation from Median (db) NESZ Variation from Median (db) Noise Equivalent Sigma0 (NESZ) The NESZ depends on the pulse noise level, beam-related noise factors, and processing gains For the new maritime modes, it varies as a function of azimuth angle as well as elevation angle within each ScanSAR image block: Lowest near the centre of a block Highest near the corners of a block Overall NESZ range: OSVN: -34 db to -24 db DVWF: -31 db to -20 db DVWF OSVN 21

22 Beam Pointing Monitoring Elevation beam pointing is monitored using elevation pattern analysis over the Amazon The shifts needed to align the measured and reference elevation patterns are recorded and trended over time Nominal beam centre Elevation Pointing Error Actual beam centre Azimuth Pointing Error Azimuth beam pointing is monitored by comparing pitch and yaw measured on-board with Doppler centroid frequency estimated adaptively during SAR processing For image quality monitoring products over the Amazon This shows trends in how the beam pointing varies with respect to nominal 22

23 Elevation Beam Pointing Results Mean pointing difference overall ~= 0 Typically within +/- 0.1 o, s ~= 0.02 o Overall results are stable since initial operations in April 2008 The measurements contain a few outliers, due to variations in scene content (Amazon scenes not perfectly uniform) 23

24 Azimuth Beam Pointing Results Mean azimuth beam pointing from antenna boresight follows seasonal trends (+/- ~50 Hz ~= 0.01 o ), likely due to thermal effects Consistent since solar array tracking has been used year-round (since August 2013) Compensated by adaptive Doppler centroid estimation in the SAR processor and by inter-wing phase calibration adjustments 24

25 Inter-wing Phase Monitoring A phase difference between fore and aft receive apertures (wings) causes an azimuth beam pointing shift (Doppler shift) from boresight The main impacts of uncompensated phase differences are subtle radiometric residual ripples in Spotlight images Not typically visible except in scenes of uniform ground cover Not a significant issue in other commercial modes range We use Spotlight images over the Amazon to monitor and correct for this: Observed radiometric errors are used to estimate inter-wing phase imbalances, which drive regular seasonal calibration adjustments 25

26 Inter-wing Phase Balance and Seasonal Calibration Adjustments Seasonal trends resemble those of the mean Doppler Centroid frequency delta from boresight Calibration adjustments track them at discrete intervals in both H and V receive polarizations Changes are compensated through adaptive processing and calibration adjustments to maintain optimal image quality 26

27 Polarimetric Calibration Monitoring For RADARSAT-2, polarization distortion is characterized over homogeneous regions of the Amazon rain forest: Supports calibration over the entire swath Known, stable, polarimetric signature: Reciprocity Azimuth symmetry High signal to noise ratio No ground infrastructure cost Input: SLC products with HH absolute radiometric calibration applied Exclude Exclude Procedures: The homogoneous area of each product is selected manually (minor imperfections in excluded regions are OK) This area is partitioned into range sections, each spanning a fixed elevation angle (~0.2 ) An average 4 x 4 covariance matrix is calculated over each section, representing the observed polarimetric signature Imbalance and cross-talk on TX and RX are estimated from each matrix These are tracked and trended over time FQ4W Covariance matrix for Amazon rain forest a 0 0 d Co-pol g 0 backscatter: a ~ 6.5 db 0 b b 0 C= X-pol g 0 backscatter: b ~ 12.5 db 0 b b 0 Co-pol g 0 product: d ~ 9.5 db d 0 0 a 27

28 Polarimetric Balance Results Following calibration, accuracy is well within performance goals: TX imbalance: annual mean ~= 0, s = ±0.1 db intensity, ±1 phase RX imbalance: annual mean ~= 0, s = ±0.1 db intensity, ±2 phase 28

29 Mean Absolute Value (db) Polarimetric Cross-Talk Results Following calibration, cross-talk levels remain excellent Less than -30 db, well within performance goals C12 C13 C42 C Year C12 = spatial averaged Hermitian product of the HH and HV complex scattering amplitudes C13 = spatial averaged Hermitian product of the HH and VH complex scattering amplitudes C42 = spatial averaged Hermitian product of the VV and HV complex scattering amplitudes C43 = spatial averaged Hermitian product of the VV and VH complex scattering amplitudes 29

30 Other Aspects of Image Quality Monitoring Type of Monitoring Point target sidelobe ratios Azimuth pattern shape Ambiguity levels Chirp replica coefficients BAQ table usage Payload local oscillator Antenna Diagnostics Product Quality Control at Gatineau HQ Description Stable (within specifications where clean measurements are possible) Stable since initial operations (correlations of measured patterns with reference patterns > 98%) Stable (results of occasional spot checks in selected modes are consistent with models) Stable Stable (good use of available dynamic range) Stable (based on timing analysis of raw echo data) Nominal (antenna is operating well and elements are stable year to year, with seasonal magnitude and phase variations) Each product is checked for artifacts, geolocation discrepancies, coverage errors any problems are reported and tracked 30

31 Conclusion RADARSAT-2 image quality remains stable and well within operating objectives Geolocation accuracy has improved New, enhanced definitive (post-processed) orbit data now available Refinements made in calibration and processing to remove small biases Two newly-released maritime ScanSAR modes (Ocean Surveillance, Ship Detection) Trade-off between visual appeal and ship detection performance Calibration adjustments will continue to be applied as needed to maintain quality As per seasonal fluctuations in inter-wing phase As per any long-term gradual changes in other monitoring measures (e.g. radiometric, noise, and polarimetric) 31

32 THANK YOU! 32