September 2008 E M A I L : I N F C R E A S O. C O M H T T P : / / W W W. C R E A S O. C O M

Transcription

1 Technical Description in a Nutshell September 2008 C R E A S O G M B H T A L H O F S T R A S S E 3 2 A D G I L C H I N G T E L E F O N ( ) F A X ( ) E M A I L : I N F C R E A S O. C O M H T T P : / / W W W. C R E A S O. C O M

2 Table of Contents Preface Overview Basic Focusing Gamma and Gaussian Filtering Interferometry ScanSAR Interferometry Polarimetry and Polarimetric Interferometry Persistent Scatterers Quality Assessment Tool Key Features sarmap, September

3 Preface SARscape is a product developed by sarmap sarmap is a Swiss company founded in 1998 as spin-off of the Remote Sensing Laboratories, University of Zurich. sarmap provides algorithms, software, and applications development including consultancies in the remote sensing field, particularly airborne and spaceborne SAR data processing. Further information is available at with contributions from Aresys Aresys is a Politecnico di Milano spin-off company, operating since 2003 in the field of digital signal processing applied to remote sensing systems. Aresys founders are a group of professors and engineers belonging to Politecnico di Milano, with a well-established experience in remote sensing area. Further information is available at and Privateers Privateers founded in 1996 is formed by a pool of leader scientists in the fields of optical and radar remote sensing, aerospace engineering, and agronomy. Key applications are in the marine environment, hydrogeological survey, forest inventory, agriculture monitoring, and space reconnaissance domain. Further information is available at sarmap, September

4 Synthetic Aperture Radar (SAR) systems can acquire data in different ways, such as - Single or dual channel mode (for instance HH or HH / HV or VV / VH); - Interferometric (single- or repeat-pass) mode; - Polarimetric mode (HH,HV,VH,VV); - by combining interferometric and polarimetric acquisition modes. Obviously, different acquisition modes are subject to different processing techniques. They are: - Processing of SAR Intensity The product generation is limited to the intensity processing. - Interferometric SAR (InSAR/DInSAR) Processing The product generation includes intensity, and interferometric phase processing. - Polarimetric SAR (PolSAR) Processing The product generation includes intensity, and polarimetric phase processing. - Polarimetric-Interferometric SAR (PolInSAR) Processing The product generation includes intensity, polarimetric, and interferometric phase processing. In order to support these processing techniques SARscape provides following modules: - Basic It includes a set of processing steps for the generation of SAR products based on intensity including a multi-purpose tool. This module is complemented by: - Focusing It supports the focusing of RADARSAT-1, ENVISAT ASAR, and ALOS PALSAR data. - Gamma and Gaussian Filtering It includes a whole family of SAR specific filters. They are particularly efficient to reduce speckle, while preserving the radar reflectivity, the textural properties and the spatial resolution, especially in strongly textured SAR images. - Interferometry It supports the processing of Interferometric SAR (2-pass interferometry, InSAR) for the generation of Digital Elevation Model and Coherence maps, as well as the processing of Differential Interferometric SAR data (n-pass interferometry, DInSAR) for the generation of Land Displacement/Deformation maps. This module is complemented by: - ScanSAR Interferometry It offers the capabilities to process InSAR and DInSAR data over large areas (400 by 400 km). - SAR Polarimetry and Polarimetric Interferometry It supports the processing of polarimetric and polarimetric interferometric SAR data. - Persistent Scatterers It enables to determine mm-scale displacements of individual features on the ground. - Quality Assessment Tool It includes a set of functions for the SAR data quality assessment in geometric, radiometric, and polarimetric terms. sarmap, September

5 1. Overview SARscape is a modular set of functions dedicated to the generation of products based on the following spaceborne Synthetic Aperture Radar (SAR) sensors - ERS-1 and 2 SAR - JERS-1 SAR - RADARSAT-1 and 2 - ENVISAT ASAR - ALOS PALSAR - TerraSAR X-1 - COSMO SkyMed-1-2 (3-4 after launch) - RISAT-1 (after launch) and following airborne SAR systems: - OrbiSAR-1 (X- and P-band) - E-SAR It consists of eight complementary modules: Basic The Basic module includes a set of processing steps (e.g. focusing, multi-looking, co-registration, despeckling, feature extraction including coherence, geocoding, radiometric calibration and normalization, segmentation, and mosaicing) for the generation of airborne and spaceborne SAR products based on intensity and coherence. This is complemented by a multi-purpose tool, which includes a wide range of functions - from image visualisation, to Digital Elevation Model import and interpolation, to cartographic and geodetic transforms. Focusing The Focusing module - developed by Aresys - extends focusing capabilities to RADARSAT-1 (Fine Beam and Standard Beam), ENVISAT ASAR (Alternating Polarization, Image, and Wide Swath), and ALOS PALSAR (Fine, Dual Pol, Full Pol, ScanSAR) data. In all cases an optimised phase preserving ω-κ processor is adopted. Gamma and Gaussian Filter The Filter module, which is an extension of the Basic module, includes a whole family of SAR specific filters. The algorithms developed by Privateers are based on Gamma/Gaussian-distributed scene models. They are particularly efficient to reduce speckle, while preserving the radar reflectivity, the textural properties and the spatial resolution, especially in strongly textured SAR images. Interferometry (InSAR and DInSAR) This module supports the processing of Interferometric SAR (2-pass interferometry, InSAR) for the generation of Digital Elevation Model and Coherence maps, as well as the processing of Differential Interferometric SAR data (n-pass interferometry, DInSAR) for the generation of Land Displacement/Deformation maps. sarmap, September

6 ScanSAR Interferometry (InSAR and DInSAR) The ScanSAR Interferometry module - developed by Aresys based on an original algorithm of Politecnico di Bari (POLIBA) - extends the Interferometry module to the ScanSAR mode, offering the capabilities to generate interferograms over large areas (400 x 400 km). This module enables also the generation of Digital Elevation Model, Coherence, and Land Displacement maps. Polarimetry & Polarimetric Interferometry (PolSAR and PolInSAR) The PolSAR and PolInSAR module supports the processing of polarimetric and polarimetric interferometric SAR data. Persistent Scatterers (PS) The Persistent Scatterers module developed in collaboration with Aresys enables to determine mm-scale displacements of individual features on the ground. Quality Assessment Tool (QAT) The Quality Assessment Tool module provides qualitative and quantitative information about the quality of SAR data in geometric, radiometric, and polarimetric terms. Note that Basic, Interferometry, and Quality Assessment Tool modules are standalone. The other modules require the Basic or the Interferometry, as shown in the table below: SARscape modules Basic Interferometry Focusing * Gamma and Gaussian Filtering ScanSAR Interferometry PolSAR and PolInSAR Persistent Scatterers *) Focusing requires Basic or, alternatively, the Interferometry module. sarmap, September

7 SARscape runs on Personal Computers (with a minimum configuration of 512MB RAM, 18GB HD and 500MHz processor) and it is interfaced to ENVI (ITT VIS.) environment (ENVI 4.1 onwards), hence enabling to simply launch the processing functions via standard dialog panels. This, in turn, calls the data processing routines which are written in C++. The figure shows how SARscape appears in ENVI software environment. SARscape supports Windows 2000/XP/Vista and Linux 2.6 operating systems. sarmap, September

8 2. Basic The Basic module includes a set of processing steps for the generation of spaceborne and airborne SAR products based on intensity and coherence. This is complemented by a multi-purpose tool, which includes a wide range of functions - from image visualisation, to Digital Elevation Model import and interpolation, to cartographic and geodetic transforms. Following processing capabilities are supported: Focussing and Multilooking The SAR focusing based on ω-k frequency domain algorithm generates for ERS-1/2 and JERS-1 SAR data a complex image from amplitude and phase information. A multi-look image can be generated from an SLC image by averaging the power across a number of lines in both the azimuth and range directions. Co-registration In cases where multiple image data sets cover the same region, it is necessary that pixels in different images correspond so that pixel-by-pixel comparisons can be carried out. Spatial registration may be necessary, and also resampling, in cases where pixel sizes vary. Coregistration is carried out automatically, based on maximising correlation in a number of windows. Filtering It supports a number of conventional single-image filters based on the classical minimum mean square error and on the anisotropic non-linear diffusion theory. This set of filters is complemented by two advanced multi-temporal multi-sensor despeckling filters. Feature Extraction Different feature parameters, which are generally used as inputs for classification or quantitative analysis, can be extracted from intensity and coherence. These are based on first order and time-series statistics. Geocoding, Radiometric Calibration and Normalisation of SAR data Ellipsoidal or terrain geocoding, using nominal parameters or Ground Control Points (GCP), enables the transformation from radar co-ordinates (either ground or slant range) into a given cartographic reference system. It is worth mentioning that in case those GCPs are requested, their identification in the SAR image is supported by an archive. With respect to the geometric correction (or geocoding), it is based on a rigorous range-doppler approach. The radiometric calibration, which is performed during the geocoding, is based on the exploitation of the radar equation. In addition, the radiometric normalisation enables to empirically correct the effects of the incidence angle on sigma nought by applying a modified cosine correction. Mosaicing It often occurs that the area of interest is covered by several images. In this case SAR images (amplitude and/or coherence) or products (classification results) can be combined in order to obtain an entire coverage of the area. When mapping large areas using SAR data, individual image frames have to be seamlessly joined together to create a consistent mosaic across the region. This method refines the radiometric variations by comparing the images to be mosaiced in the overlapping areas. sarmap, September

9 Segmentation Segmentation is performed on SAR/InSAR and/or optical multi-spectral data in four steps, i.e. i) by filtering the data by means of single- or multi-temporal anisotropic diffusion, ii) by applying on single- or multitemporal data an extended Canny edge detector, iii) by performing a single- or multi-temporal region growing, and finally iv) by applying a time-varying segmentation aiming at detecting temporal changes. Tools A wide range of tools is implemented in SARscape. They consist of the following functions: - GTOPO30, ACE and SRTM Digital Elevation Model data reading and automatic mosaicing - Multi-source Digital Elevation Model combination - Shape, image and Digital Elevation Model cartographic and geodetic transformation - Slope image generation - Point target analysis (SAR data) - Image statistics calculation - Multi-image operations - Specific indexes (e.g. NDVI) generation - Image resampling - Images and DEMs interpolation - Sample selection - Pan sharpening - Raster to shape data conversion - Raster data slicing - Complex interferogram to phase and module conversion and backward transformation - Cartographic to slant geometry conversion - Filter classification - File transformations sarmap, September

10 A note on SAR (Intensity) Data Processing There is no standard processing chain. In primis, the processing depends upon how SAR data have been acquired (acquisition modes and available SAR systems). The type of product that is envisaged determines, additionally, how intermediate SAR products (for instance terrain geocoded backscattering coefficient data) will be further processed. With respect to the first point assuming the availability of a multi-temporal SAR raw data set three processing procedures can be applied: 1. Single-sensor, Single-mode, Multi-temporal Approach This is the classical one. Multi-temporal SAR data are acquired in the same mode (for instance ENVISAT ASAR Image Mode 4). Since the geometry remains the same, following steps should be considered: Focusing Multi-looking Co-registration Multi-temporal speckle filtering (for instance De Grandi) Terrain geocoding, radiometric calibration and normalization 2. Single-sensor, Multi-mode, Multi-temporal Approach This is the most appropriate one in case of single-sensor data availability. SAR data are acquired in different geometries and thereby data acquisitions are not linked to standard repeat-pass cycle geometry (ERS-1/2 like). Since the acquisition geometry is different, following steps should be at the best considered: Focusing Generation of 1-look Intensity Terrain geocoding, radiometric calibration and normalization Multi-temporal speckle filtering (for instance Anisotropic Non-Linear Diffusion) 3. Multi-sensor, Multi-mode, Multi-temporal Approach This is the most advanced one, since based on the principle of satellite s constellation. SAR data are acquired in different geometries by different sensors. Therefore, data acquisitions can be optimized in terms of temporal, spatial (and spectral) resolution. In this scenario, the processing chain corresponds to the previous one. Note that particularly in this case, it is imperative that the data are accurately absolutely radiometrically calibrated. sarmap, September

11 Examples Single date mosaic based on 1250 JERS-1 SAR images of Central Africa Multi-temporal mosaic based on 180 ENVISAT ASAR images (Senegal) sarmap, September

12 Map of rice emergence during the crop season 2003/4 derived from ENVISAT ASAR and Radarsat-1 combined acquisitions (Luzon - Philippines) Map of burnt areas based on multi-temporal ERS-2 SAR data (Canada). Note the agreement between the EO product and the burnt areas detected by the Canadian Forest Service. Different colours correspond to the plant re-growth. sarmap, September

13 3. Focusing The Focusing Module - developed by Aresys - extends SARscape focusing capabilities to RADARSAT-1, ENVISAT ASAR, and ALOS PALSAR data. It is based on a ω-k focusing kernel and it handles mono as well as bistatic SAR, ScanSAR, Spotlight and any other modes. The processor core is based on the well known ω-k kernel. As a major result of the crossbreeding between geophysics and SAR: the ω-k is suited to provide fast and accurate focusing for any squint angle (0 o to 89 o ) and wide apertures. In particular, it has been designed with emphasis on geometric accuracy (localisation) for SAR Interferometry single and multi mode. The Focusing module is profiled and optimised by using Intel Math Kernel libraries (TM), that takes full advantage of the multithreading and streaming SIMD extensions of the latest INTEL family processors, BLAS and LAPACK linear algebra libraries, and AMD and Intel architecture optimised FFT routines. This module is able to provide Single Look Complex (SLC) and detected products starting from the following raw data: - ENVISAT Image Mode data - ENVISAT Alternating Polarization data - ENVISAT Wide Swath 1 data - RADARSAT-1 Fine Beam mode - RADARSAT-1 Standard Beam mode - ALOS PALSAR Single Polarization - ALOS PALSAR Dual Polarization - ALOS PALSAR Full Polarization Note that Single Look Complex data generated with this module are not appropriate to derive absolutely radiometric calibrated values. 1 WSM processor is based on the original algorithm developed by Politecnico di Bari (POLIBA) sarmap, September

14 4. Gamma and Gaussian Filtering This module, which is dedicated to the speckle removal in SAR images, includes a whole family of SAR specific filters. The algorithms developed by Privateers are based on Gamma/Gaussian-distributed scene models. They are particularly efficient to reduce speckle noise, while preserving the radar reflectivity, the textural properties and the spatial resolution, especially in strongly textured SAR images. The Filter module provides different specific algorithms, which can be applied to: - Single- or multi-look, single channel detected products - Single- or multi-multichannel detected products - Single Look Complex (SLC) SAR images - Separate SLC channels - Full polarimetric SAR data In presence of scene texture, to preserve the useful spatial resolution, e.g. to restore the spatial fluctuations of the radar reflectivity (texture), an A Priori first order statistical model is needed. With regard to SAR clutter, it is now well accepted that the Gamma-distributed scene model is the most appropriate. The Gamma-distributed scene model, modulated by, either an independent complex-gaussian speckle model (for SAR SLC images), or by a Gamma speckle model (for multilook detected SAR images), gives rise to a K- distributed clutter. Nevertheless, the Gaussian-distributed scene model remains still popular, mainly for mathematical tractability of the inverse problem in case of multi-channel SAR images (multivariate A Priori scene distributions). The supported SAR filtering methods are: A. Detected SAR data Single channel detected SAR images - Gamma-Gamma MAP (Maximum A Posteriori) filter - Gamma-DE (Distribution-Entropy) MAP filter - Gamma APM (A Posteriori Mean) filter Multichannel detected SAR images: - Gamma-Gaussian MAP filter for uncorrelated speckle - Gaussian-Gaussian MAP filter for correlated speckle - Gaussian-DE MAP filter for correlated speckle B. Complex SAR data - Single Look Complex (SLC): Complex Gaussian-DE MAP filter - Separate complex looks or interferometric series: Complex Gaussian-Gamma MAP filter C. Polarimetric SAR data - Complex Wishart-Gamma MAP for polarimetric SAR data - Complex Wishart-DE MAP for polarimetric SAR data sarmap, September

15 The figure shows an example of: - ERS-1 PRI (upper left) and its filtered version (upper right) - Radarsat-1 SGF (lower left) and its filtered version (lower right) The filter applied here is a Gaussian-Gaussian Map filter for multi-channel detected multi-look SAR images sarmap, September

16 5. Interferometry This module supports the processing of Interferometric SAR (2-pass interferometry, InSAR) and Differential Interferometric SAR (n-pass interferometry, DInSAR) data for the generation of Digital Elevation Model, Coherence, and Land Displacement maps. The processing includes the following steps: - Phase-preserving SAR focussing - A ω-k frequency domain algorithm has been implemented to guarantee the lowest phase distortion and to preserve the original interferometric coherence. - Coarse and fine coregistration using Digital Elevation Model provides the proper parameters in order to align the same points in the two images. - Common Doppler bandwidth filtering - If there is a difference in the Doppler Centroids, this step allows minimisation of the decorrelation. - Interferogram generation - Spectral shift filtering is performed on the image pair, and the Hermitian product is calculated. - DEM flattening - Synthetic fringes are generated from a coarser resolution DEM or ellipsoidal height, using a backward geocoding approach, and then cross-multiplied by the SAR interferogram. This step allows removal of most of the low frequency components of the wrapped phase, to ease the following phase unwrapping. - Adaptive filtering - The flattened complex interferogram is filtered to improve the phase SNR, improving the height accuracy of the final DEM. - Phase unwrapping - The filtered interferogram is unwrapped by using a region growing approach. - Phase edit Unwrapping errors can be corrected in a semi-automatic or completely manual way. - Geometry optimisation, based on Ground Control Points - The orbits are refined in order to obtain an accurate conversion of the phase information to height and a proper geo-location. Both manual and automatic procedures for the identification of the most suitable image pixels to be used as GCPs are implemented. - The synthetic phase is added back to the unwrapped phase, so that the final DEM is derived only from the SAR data. - Phase to map conversion - A rigorous approach is implemented, that takes into account the range- Doppler nature of the SAR acquisition. The final product is obtained in the desired cartographic reference system by taking into account all the necessary geodetic and cartographic parameters and transformations. - Mosaicing - The area not covered is minimised by combining DEM obtained from different acquisition orbits. - Phase to displacement The phase values are converted to displacement and geocoded onto a map projection. The user can enter a specific direction (azimuth angle) and inclination angle along which he assumes that the main deformation occurred, or even enter a a-priori known deformation field as input. For the generation of displacement maps it is mandatory to run a second iteration of the processing, from the interferogram flattening to the phase unwrapping, after the execution of the baseline fit step (which is aimed at correcting orbital inaccuracies). sarmap, September

17 2-pass SAR Interferometry Example Figure 1 show the Digital Elevation Model of the whole Switzerland, where height values are ranging from 195 m (darkest grey tones) to 4634 m (brightest grey tones). The horizontal resolution of this product is 25 metres, while the height accuracy varies from 7 (moderate topography) to 15 metres (steep topography). Digital Elevaton Model of Switzerland sarmap, September

18 Differential SAR Interferometry (n-pass Interferometry) Example An example of differential interferometry showing differential interferogram, coherence and displacement map illustrated in Figure 2. The images are relevant to the analysis of the earthquake event on December 26 th 2003 in the city of Bam (Iran). In particular: - Figure A shows the differential interferogram generated from the ENVISAT ASAR pair acquired on 3 December 2003 (pre-earthquake) and 11 February 2004 (post-earthquake); a reference InSAR DEM was used for subtracting the topography induced fringes. - Figure B shows the coherence image relevant to the 3 December 2003 / 11 February 2004 ENVISAT ASAR pair. It is important to note that lowest coherence values (darkest tones) correspond both to steep slopes or vegetated areas (especially visible in the lower part of the image) and to the Bam built zone (image centre), which was completely destroyed by the earthquake. - Figure C shows the deformation map obtained considering a strike slip fault (horizontal movement) with a N_S oriented fault plane. Red and green tones represent the areas of largest deformation (in opposite direction), while yellowish areas correspond to smaller deformation zones. Figure A Differential Interferogram (Bam, Iran) sarmap, September

19 Figure B Coherence (Bam, Iran) Figure C Displacement map (Bam, Iran) sarmap, September

20 6. ScanSAR Interferometry This module - developed by Aresys based on an original algorithm from Politecnico di Bari (POLIBA) - extends the SARscape Interferometry module to the ScanSAR mode, offering the capabilities to generate interferograms over large areas (400 x 400 km). This module enables also the generation of Digital Elevation Model, Coherence and Land Displacement maps based on ENVISAT ASAR Wide Swath data. In addition this module provides also hybrid interferometric products, by combining StripMode (ASAR Image Mode) data with ScanSAR (ASAR Wide Swath) ones. The main processing capabilities are: ScanSAR Single Look Complex (SLC) Pair Overlap Estimation ScanSAR data alignment, and burst overlapping, is estimated using sensor orbit descriptions. Doppler Common Band Filtering Common azimuth bandwidth is selected on the master and slave bursts, minimising the signal decorrelation. Burst Coregistration Coregistration coefficients are computed using sensor orbit descriptions. The slave burst is then resampled over the master burst grid. Interferogram Generation Coregistered bursts are filtered along the range direction in order to minimise the geometrical decorrelation; either Synthetic fringes or a flat Earth interferogram can be used for this purpose. The interferogram is subsequently generated as an Hermitian product between master and coregistered slave bursts. Single Swath Interferogram and Single Swath Detected Image Generation Single burst interferograms are collected and coherently summed providing a multilooked interferogram image for each ScanSAR swath. In the same step the single bursts absolute value data are collected, weighted with a de-scalloping window, and incoherently summed, providing two (master and slave) single swath, coregistered detected images Swaths Recomposing Single swath data (interferogram and detected data) are composed into multi swath images. sarmap, September

21 As example, the Figure below shows a differential interferogram (left) and corresponding coherence (right) generated from an ENVISAT ASAR Wide Swath pair (400 km by 400 km) acquired in the area of Bam (Iran). Respectively on September 2nd 2003 (pre-earthquake) and on June 8th 2004 (post-earthquake). Compare this differential interferogram with that one illustrated in Figure 2a. Differential interferogram (left) and corresponding coherence (right) generated from an ENVISAT ASAR Wide Swath pair (400 x 400 km). sarmap, September

22 7. Polarimetry and Polarimetric Interferometry This module supports the processing of polarimetric and polarimetric interferometric SAR data. The processing capabilities for SAR Polarimetry are: Polarimetric Calibration Matrix Polarimetric calibration allows minimizing the impact of non ideal behaviours of a full polarimetric SAR acquisition system, to obtain an estimate of the scattering matrix of the imaged objects as accurate as possible from their available measurement. Polarimetric calibration is performed based on defined calibration matrixes. Polarimetric Signature This function provides an estimate of the polarimetric signature of a point-target-like object, whose location is specified either in terms of its range and azimuth indexes within a slant-range acquisition, or through its known cartographic coordinates. Polarimetric Synthesis The availability of a full polarimetric acquisition, for example based on a linearly polarized base, allows synthesizing any desired elliptical polarization. Starting from a set of full polarimetric linearly polarized Single Look Complex (SLC) data, this module allows to synthesize a new set of SLC data in a desired orthogonal basis, either circular, linear rotated of 45 degrees or generic elliptical of user defined orientation and ellipticity. Polarimetric Features This function performs the computation of conventional polarimetric features - among others the Linear Depolarization Ratio, the Polarimetric Phase Difference, and the Polarization HH-VV Ratio - based on the scattering matrix. Pauli Decomposition The Pauli coherent decomposition provides an interpretation of a full polarimetric SLC data set in terms of elementary scattering mechanisms: sphere/plate/trihedral (single- or odd-bounce scattering), dihedral oriented at 0 o (double- or even-bounce) and diplane oriented at 45 o (qualitatively related also to volume scattering). Pauli Decomposition based on ALOS PALSAR PLR data (Data copyrighted by JAXA). sarmap, September

23 Entropy-Anisotropy-Alfa Decomposition The Entropy-Anisotropy-Alpha decomposition performs an eigen-decomposition of the coherency matrix of a full polarimetric SLC data set. In order to facilitate the analysis of the physical information provided by the eigen decomposition of the coherency matrix, three secondary parameters are defined as a function of the eigenvalues and eigenvectors, namely - Entropy - it is related to degree of randomness of the scattering process - Anisotropy - it measures the relative importance of the second and third eigenvalue of the eigendecomposition - Alpha - it relates the type of scattering mechanism Entropy-Anisotropy-Alfa Decomposition based on ALOS PALSAR PLR data (Data copyrighted by JAXA). Entropy-Anisotropy-Alfa Classification It performs an unsupervised classification of the results of a previous Entropy-Anisotropy-Alpha decomposition step, identifying the main scattering type of every pixel in the target area. The main processing capabilities for Polarimetric SAR Interferometry are: Single Look Complex (SLC) Coregistration This function performs an unsupervised classification of the results of a previous Entropy-Anisotropy- Alpha decomposition step, identifying the main scattering type of every pixel in the target area. Polarimetric Phase Difference (PPD) / Interferogram Generation This function performs i) the Polarimetric Phase Difference using the HH and VV polarization of the same acquisition, and ii) it generates interferograms using the same polarization of the PolInSAR pair. Polarimetric / Interferometric Coherence This function generates correlation/coherence images based on Polarimetric Phase Difference / Interferogram products. Coherence Optimization This function estimates the main scattering mechanisms of a full polarimetric pair of linear acquisitions, identifying those mechanisms that correspond to the highest value of interferometric coherence. The corresponding interferograms and coherence maps are provided as result. sarmap, September

24 8. Persistent Scatterers This module jointly developed by sarmap and Aresys enables to measure very small ground displacements and to trace the deformation rate related to natural or man-induced phenomena (e.g. volcanic or seismic activity, landslides, subsidence, building collapses, etc.), with excellent accuracy and spatial and temporal detail. This technique, extending SAR interferometry to the analysis of large multi-temporal observations, enables to improve the measurement accuracy from few centimetres (classical interferometry approach) to few millimetres (Persistent Scatterers approach). In addition to that, limitations typical of SAR interferometry (i.e. atmospheric distortions or temporal de-correlation) are significantly reduced. The method relies on the identification of a certain number of coherent radar signal reflectors (Persistent Scatterers, PS). The processing is then focused on the analysis of the phase history of these very reliable single points (pixels), as opposed to the conventional interferometric approach. The product consists of the displacement history of every single PS together with its average displacement rate. Examples of PS candidates are typically represented by urban features (e.g. house roofs, bridges, antennas, etc.), other man-made structures (e.g. green-houses, dams, metallic and concrete features such as well fields surrounding structures, pipelines and dwells etc.), well exposed outcropping rock formations. Two are the minimum pre-requisites: i) minimum density of the scatterers (at least 5 to 10 per sqkm) must be detectable in the SAR viewing geometry; ii) a large number of SAR observations (at least 15 to 20 images) obtained from the same sensor with the same acquisition geometry. Furthermore, the temporal distribution of the acquisitions shall also be adequate compared with the expected dynamics of the displacements under analysis. The Persistent Scatterers module which supports SAR data acquired by ERS and ENVISAT satellites is a stand alone module. It is anyway important to mention that, in order to exploit at the best the potential of this module, its combined use with the Basic and Interferometry modules is of advantage. The SARscape integrated module approach will enable to comprehensively process the SAR multi-temporal series in terms of both intensity (e.g. accurately filtered, geocoded and calibrated images outputted by the Basic module) and interferometric products (e.g. coherence images, interferograms, DEMs and Displacement maps generated by the Interferometry module). Both the Basic and the Interferometry modules will be then essential to generate those SAR reference products, which are crucial to properly interpret and present the results derived from the use of the Persistent Scatterers module. sarmap, September

25 Standard Persistent Scatterers outcomes. Standard Persistent Scatterers outcomes: maps (top) and PS history (bottom) sarmap, September

26 9. Quality Assessment Tool The correct use of Earth Observation data implies a rigorous data calibration in geometric, radiometric and for SAR systems polarimetric, and interferometric terms, in particular if the primary goals are: - To analyze multi-temporal images acquired from the same sensor; - To analyze images acquired from different sensors; - To generate indexes; - To infer bio- or geo-physical parameters. Despite the large effort over the past two decades of the different calibration/validation working groups, the quality of pre-processed images i.e. the generation of co-registered and/or ortho-rectified radiometrically calibrated products still remains today questionable. The main reasons are: - Inappropriate models (often limited to empirical models); - Limitations of software tools; - Inappropriate input images; - Inappropriate sensor parameters; - Inaccurate platform parameters; - Inaccurate Ground Control Points (GCP) or Tie Points (TP); - Inaccurate Digital Elevation Model (DEM). In summary, in order to quantitatively assess the geometric, radiometric, and polarimetric quality of SAR products, a customized Quality Assessment Tool (QAT) based on a solid methodology is proposed. The overall architecture of the Quality Assessment Tool system which supports the quality assessment for ENVISAT ASAR (IM/AP/WS), Radarsat-1 and 2 (FB, SB, WB, EB, ScanSAR) is illustrated in the Figure below. The following main packages can be identified: - A first one implementing the Human Machine Interface; - A second package responsible of the ingestion of the different products and of the execution of all supported test; - A database package storing all configuration settings, reference and nominal parameters and GCP measurements as well as the results of the analysis of different SAR products; - A database connection layer, that is responsible of interfacing the results of the different processing modules with the database. sarmap, September

27 Overall architecture of the QAT package Following quality parameters and assessment procedures are supported: - Background removal and image interpolation - Impulse Response Function measurements - Spatial resolution in range and azimuth - Peak Side Lobe ratio - Range and azimuth Peak Side Lobe ratio - Spurious sidelobe ratio - Integrated Side Lobe Ratio - Derivation of sigma nought and radiometric calibration - Sigma nought derivation for ASAR ground range projected products - Sigma nought derivation for ASAR single complex slant range projected products - Beta nought for derivation or Radarsat-1 detected products - Beta nought derivation for Radarsat-1 single look complex products - Conversion from beta nought to signma nought - Calibration constant determination - Radiometric analysis - Radiometric accuracy - Radiometric resolution - Equivalent Number of Looks - Radiometric stability sarmap, September

28 - Noise equivalent sigma nought - Ambiguity analysis - Azimuth ambiguity location offset - Range ambiguity location offset - Point target ambiguity ratio - Geometric analysis - Level 1 products geocoding - Absolute location accuracy - Sampling Window Start Time bias - Ground Control Point geocoding and residual analysis - Swath width and position - Channel co-registration - Multi-temporal co-registration - Ratio - Polarimetric analysis - Polarimetric Phase Difference and polarimetric coherence - Polarimetric signature The Quality Assessment Tool is composed by three logical parts: - All what concerns Graphical User Interface (GUI), data, and information display is integrated within the ENVI environment (Window and Linux ). This solution allows to obtain professional tools for data visualisation both for raster and vector data in standard formats, and it enables to exploit the existing basic analysis tools already available within ENVI. This layer has the purpose of driving the execution of the different Quality Assessment modules in an interactive way as well as displaying their results, in terms of images, textual data and plots. - All modules related to initial data import and assessment are available as executable modules, originally written in C++ and that may be invoked both through the mentioned GUI and as command-line modules. Data import, in particular, has the purpose of interfacing the QAT system with SAR data originating from different platforms, and hence stored and delivered in different formats. The data stored in these formats are analysed during import phase, all the necessary parameters are extracted and stored in a XML file with common syntax (documented in the delivered documentation). The binary data (image products) are stored in a plain matrix (BIL) file, and a correspondent.hdr file is generated, consistent with ENVI specifications, to allow the direct display of the binary data within the ENVI environment. All Quality Assessment modules deal directly with the data imported into a common metaformat. - The third layer of database has the purpose of intercepting all the results of the Quality Assessment modules and storing them within a history database, as well as storing reference and nominal values for important parameters. A part of the database is dedicated to storing information concerning Ground Control Points to be reused during a long-time assessment of SAR data quality over a same reference test-site. sarmap, September

29 Quality Assessment Tool Graphical User Interface. sarmap, September

30 10. Key Features Among the several functionalities included in the software package, the following key features distinguish SARscape from competing products: Phase-preserving SAR focussing: high precision SLC data can be obtained from SAR raw data, starting from either Image mode or ScanSAR products, using a ω-k algorithm. Automatic coregistration of multi-temporal SAR or optical image stacks: using cross-correlation techniques, pixel-accuracy is obtained without need of manual selection of corresponding points. Multi-temporal speckle filter: when several SAR acquisitions are available over a same area, this additional unique filtering functionality allows obtaining high speckle reduction while conserving all spatial and temporal information. A dedicate SAR filtering module, based on the most advanced Gamma and Gaussian distributed scene models, is suitable for products such as SLC, PRI, multitemporal data sets, etc. One-step coherence maps computation: all processing steps necessary to generate such products can be performed as just one operation. Rigorous geocoding of SAR data (from slant range or ground range geometry) based on Range-Doppler equations. Accurate terrain correction can be performed when a DEM is available; radiometric correction is performed by accounting for all instrument (e.g. antenna pattern, range spread loss) and terrain influences. Precise geocoding using nominal parameters and no Ground Control Points (GCPs) may be performed when accurate orbital information is available (e.g. ERS and ASAR). In particular, for ENVISAT ASAR data, DORIS orbits can be inputted in order to get a geocoding sub-pixel accuracy using nominal parameters (e.g. no GCPs). If orbital information must be refined, the number of GCPs necessary can be reduced to one, still obtaining accurate geocoding. The identification of Ground Control Points in the SAR image is supported by an archive. Radiometric and geometric calibration of optical data can be performed during geocoding, based on DEM information when available. A wide number of cartographic reference systems is available for the geocoded products; the user can extend it by providing all necessary parameters in easily accessible ASCII files. Batch processing: long sequences of processing steps can be collected in one batch command and run as a single operation. Script programming is available as well. The InSAR and DInSAR processing chain can be applied to both medium resolution (e.g. ASAR Wide Swath products) and high resolution (e.g. ASAR Image mode products) data. Tailored and dedicated data mosaicing routines are developed to join at the best amplitude as well as DEM images. High accuracy DEMs can be generated using the interferometric module. Differential interferograms can be generated using the interferometric module to estimate small terrain movements (in the order of few millimetres). External DEMs can be imported; USGS GTOPO-30 and SRTM-3 world-wide DEMs can be automatically extracted and mosaiced over the area of interest by simply inputting the reference slant range or geocoded image. Both Windows and Linux operating systems are supported. Integration in well-known powerful image processing (ENVI ) environments, allowing the user to exploit a wide base of tools together with the specific capabilities of SARscape. The user can select the preferred environment suiting at best with his applications, accessing the same processing routines and even sharing the same data between the two platforms. Standard open formats are used for all generated products. sarmap, September