Comisión Tercer Informe Bi-mestral (third bimestrial report) Nicolas A. Mari, Giovanni Laneve, Ximena Porcasi Lic.Nicolas A. Mari Dr. Giovanni Laneve - 1 -
Comisión 5. Processing of SAR data: The intention of the third bi-mistral report is to provide a baseline methodology for processing Cosmo Sky-Med (CSM) SAR data. The main idea is to show the technical aspects to take in consideration when we need to process CSM from Level 1A data. The methodology is presented according to the technical specifications of the Cosmo SkyMed products within its regular production chain, and in line with practical applications related to the mapping of natural and forested areas. The ultimate goal of this report, is to create a basic guide of processing SAR data presented over diverse surface properties. Technical information is provided from diverse sources and in particular from the SARscape software module. 5.1 Material and Methods Cosmo Sky-Med data The available data corresponds to an image covering the area around the town of Reggio Calabria, located in the south end of the Italian peninsula, acquired the 20 of October 2008 with a right ascending look side, with west hill shades faced toward the sensor (Figure 1). The region is characterized by a mountainous landscape with diverse land cover types, with long term agricultural activities. Main vegetation types are oaks and pines, principally located at higher elevations. There is also vineyards, olives and citrus fruits historically cultivated. The intervention of man made terraces are more predominant in lower regions near the coast line. - 2 -
Temp. max. media ( C) Precipitation (mm) Comisión Figure 1: Acquired scene with limits of the Parco Nazionale Dell aspromonte. Red Point indicates the location of Reggio Calabria meteorological station. In order to have a global idea of the condition of the vegetation types at the acquired date, we looked for historical climatologic data (Figure 2). October is a month where the vegetation is starting to peak after getting out from a deficit water period (April-September), characterized by high maximum mean temperatures and little relative mean precipitation levels (Figure 2). At this stage, from September to April the increasing precipitation and mild temperatures, provides sufficient humidity to vegetation and crops. 35 90 30 25 20 80 70 60 50 15 10 5 0 January February March April May June Temperature July August September October Precipitation November December 40 30 20 10 0 Figure 2: Climatological data (1971-2000) from Reggio Calabria. Red bar indicates the month of the acquisition image. - 3 -
Comisión The instrument mode of the acquired image is Ping Pong, with medium spatial/swath resolution with co-polarized HH-VV polarization. The ping pong mode implements a strip acquisition by alternating a pair of Tx/Rx polarization across bursts by mean of an antenna stirring. The signal pol is alternated between two possible ones : VV, HH, HV and VH. The swath of the Satellite acquisition is 30 km with a spatial resolution of 15m. Table 1: Description of the images and their characteristics for processing. Region Code Product Type Instrument Mode Polarization Look Side Level Calabria Cl-1 SCS_B (Single look Complex Slant) PP (Ping Pong) CO (Co polarized) HH-VV RA (Right assending) level 1A 5.2 Methodological scheme As indicated above, the idea of this report is to provide an easygoing methodology for obtaining SAR product of CSM ready to use. For this purpose we used the ENVI SARscape 4.2 (c) software licensed by SARMAP s.a. This software is a straightforward and intuitive module that permits the processing of many standard formats including CSM data. Concerning the characteristics of the available data, and with the objective of generating a product oriented to be used in forestry applications, we developed a methodology based on the work presented by Holecz et al., 2010. The contribution of this paper, and the use of optical data to evaluate results is the main support of our approach. The methodological scheme is presented as follows in Figure 3. The flowchart is divided in two main areas: The processing area, and the evaluation area. The processing area is extended from the import step to the filtering for speckle reduction. The evaluation area comprehends the use of optical data as reference. - 4 -
Comisión Figure 3: General flow chart of the methodology. Modified from Holecz et al., 2010. 5.2.1-Import Data The process starts with the ingestion of Level 1A data and its conversion to subsequent levels. SARscape Import Data module is adapted to import standard formats, including COSMO Skymed data (DGM, GEC, GTC, SCS). The import function requires only the input Sensor to be entered. The relevant data type is automatically identified. The input file to be used has an extension called.h5, and the output is a.sml file, and a *_slc file. The content of the.sml file can be seen in the Sarscape module from the View files function choosing the header file option. This file contains all the needed information that will be required later on. The *_SLC file is the input of the next processing step. 5.2.2-Multi-looking The SARscape multilooking application is found in the Basic module. The input file to process must be the *_SLC file. The module indicates the number of Azimuth looks and Range looks which are displayed as Default Values (DV). DV values must be set before starting the whole - 5 -
Comisión process, and are dependant of the characteristics of the image. The DV list can be seen from the main SARscape module. Options for CSM products can be chosen between the VHR, HR and MR. These options are related to the spatial resolution of the product, being Very high resolution (better than 10m), High resolution (between 10m and 30m), and Medium resolution (coarser than 30m) respectively. To determine which the best resolution of our product is we must understand what does mean the multi-looking process it self: The multi-looking process refers to the partition of the radar beam into numerous sub-beams. Each sub-beam, are called looks, and represents an independent portion of the illuminated scene. The multi-looked image is obtained by averaging over range and/or azimuth cells, resulting on a lower resolution image but with reduced speckle. The number of looks must be defined in a proper way in order to obtain approximately squared pixels, considering the ground range resolution. The number of looks is a function of pixel spacing in azimuth, pixel spacing in Slant Range, and incidence angle (1). Pixelspaci ngrange GroundRang ere solution : (1) SIN( incidencea ngle) It is important to consider the possibility of committing over or under sampling effects that will be evident in the final geo-coded product. To avoid these possible effects, it is recommended to generate a multi looked image corresponding to approximately the same spatial resolution of the final geo coded product. The values to get the ground range resolution from formula 1 are provided in the _SLC.sml file. To resume this step, the final resolution of the multi-looked image will be the result of computing the ground resolution in range and azimuth through formula 1, that provides the number of looks needed to compute the multi-looking process. As mentioned above, the ping pong mode has two alternated signals (HH and VV in our image). So the multi-looking process will be independent for each signal, providing two independent images (Figure 4) - 6 -
Comisión Figure 4: Multi-looked scenes. At the right HH polarization and in the left VV polarization. The oblique perspective of the SAR sensor, in this case is a RA look side, determines a terrain distortion which results in tall objects being displaced towards the sensor. This phenomenon is called Layover. The ordering of surface elements on the radar image is the reverse of the ordering on the ground. Generally, these layover zones, facing radar illumination, appear as bright features on the image due to the low incidence angle (ESA Earthnet online, Radar Course III). These distortion effects are corrected later on, in the geocoding step with the use of a Digital Elevation Model (DEM). 5.2.3 Gamma DEMAP filtering (Gamma Distribution Entropy Map) The speckle reduction is one of the principal needs in processing SAR data. The understanding of radar speckle is fundamental to make radiometric enhancements over SAR images. The Gamma DEMAP filter is an adaptive filter that is conceived for improving classification and/or for the enhancement of the SAR image, it gives superior results than other filters in areas exhibiting mixed textures or in presence of strong relief as it is in the Reggio Calabria area. It is important to keep in mind that the ideal filter will be the one which reduce speckle with minimum loss of information. In this sense, adaptive filters take into account the local properties of the terrain backscatter or the nature of the sensor, whereas non-adaptive filters are not. In this way, the restoration of the local textural properties of the scene does not depend on any assumption - 7 -
Comisión regarding the form of its statistical distribution. Optimal results can be achieved by selecting the appropriate filter depending on the land morphology and scene texture, this last to be considered also in relationship with the data spatial resolution. The Gamma DE MAP filter is located in the main SARscape module from the Single channel detected option, the Input file is _pwr. As any filter, estimating the values of input parameters can be a trial and error. Nevertheless, some tips are recommended: The size of the moving window must be set considering the characteristic of the image, for high detailed regions with high variations in grey tones is preferable to use small windows, and for homogenous regions with little variations, a window of higher size. The other parameter to consider is the Equivalent Number of Looks (ENL). The SARscape Help module refers to the ENL as equivalent to the number of independent Intensity values averaged per pixel during the multi-looking process. This parameter can be easily estimated over a homogeneous (stationary) sample in Intensity data according to (2): ENL = mean²/ standard deviation². (2) To tune the strength of speckle filtering and the level of preservation of scene details, it is preferable to adjust the value of the ENL, rather than to change the size of the processing window: To reduce the strength of speckle filtering, with the aim to preserve the thinnest details of the scene, enter a ENL value slightly higher than the calculated one; inversely, to improve the filtering of the speckle (possibly at the cost of the thinnest details of the scene...), enter a ENL value slightly lower than the calculated one. In our exercise, parameters were left with their default values. It is shown a subset of the entire image as an example to compare non filtered and filtered images (Figure 5) - 8 -
Comisión A B Figure 5: Subsets of non-filtered (A) and filtered (B) data Since the effect of the filter in this case is poor visible to the human eye, a horizontal profile (x axis) is drawn for each scene (Figure 6). It can be seen a reduction of the maximum values, with less variability between peaks. This is the result in the reduction of speckle with little impact on the overall texture of the images. Figure 6: Horizontal profiles describing the effect of the Gamma DEMAP filter. Left profile represents scene A, and right profile represents scene B. 5.2.4 Geocoding, Radiometric Calibration and Normalization This step it must be considered as the central procedure from the flow chart. To understand its purpose, some important concepts must be revised, let s start with Geocoding: - 9 -
Comisión When we refer to Geocoding, Georeferencing, Geometric Calibration, and Orthorectification we are talking about the same thing, these are synonyms. All of these definitions mean the conversion of SAR images - either slant range (preferably) or ground range geometry (Figure 7) - into a map coordinate system (e.g. cartographic reference system). When the conversion is done with the use of a DEM, we refer to Terrain Geocoding, and when this process is performed without the use of a DEM it is referred as Ellipsoidal Geocoding. Because SAR systems cause nonlinear compressions (e.g layover effects) they cannot be corrected using polynomials as in the case of optical images, where (in the case of flat Earth) an affine transformation is sufficient to convert it into a cartographic reference system, the only appropriate way to geocode SAR data is by applying a Range-Doppler approach (Meier et al, 1993):. R S P 2 f ( Vp Vs) Rs (3) fd C Rs Where Rs is the slant range, S and P are the spacecraft and backscatter element position, Vs and Vp are the spacecraft and backscatter element velocity, f is the carrier frequency, C the speed of light and fd is processed Doppler frequency. Using these equations, the relationship between the sensor, each single backscatter element and their related velocities is calculated and therefore not only the illuminating geometry but also the processors characteristics are considered. This complete reconstruction of the imaging and processing geometry also takes into account the above mentioned topographic effects (foreshortening, layover) as well as the influence of Earth rotation and terrain height on the Doppler frequency shift and azimuth geometry. - 10 -
Comisión Figure 7: Slant Range Vs. Ground Range. (Modified from http://www.eosnap.com/earth-observation). Radiometric Calibration Radars measure the ratio between the power of the pulse transmitted and that of the echo received. This ratio is called the backscatter. Calibration of the backscatter values is necessary for intercomparison of radar images acquired with different sensors, or even of images obtained by the same sensor if acquired in different modes or processed with different processors. In order to avoid misunderstanding, note the following nomenclature: Beta Nought (ß ) is the radar brightness (or reflectivity) coefficient. The reflectivity per unit area in slant range is dimensionless. This normalization has the virtue that it does not require knowledge of the local incidence angle (e.g. scattering area A). Sigma Nought (σ), the backscattering coefficient, is the conventional measure of the strength of radar signals reflected by a distributed scatterer, usually expressed in db. It is a normalized dimensionless number, which compares the strength observed to that expected from an area of one square metre. Sigma nought is defined with respect to the nominally Horizontal plane, and in general has a significant variation with incidence angle, wavelength, and polarization, as well as with properties of the scattering surface itself. Gamma (γ) is the backscattering coefficient normalized by the cosine of the incidence angle. - 11 -
Comisión The radiometric calibration of SAR data is carried out by the radar equation law. It involves corrections for: The scattering area (A). Each output pixel is normalised for the real illuminated area of each resolution cell, which may change depending on topography and incidence angle. The antenna gain pattern (G²). The antenna gain variations in range are corrected taking into account the actual topography (Digital Elevation Model) or the reference ellipsoidal height. The antenna gain can be expressed as the ratio between the received signal and the transmitted signal or by comparing a real antenna to an isotropic antenna; it is measured in db. The range spread loss (R³). The received power (backscattered signal) is corrected by taking into account the sensor-to-ground distance variation from the near range to the far range. The formula applied for the radiometric calibration is (Holecz et al., 1993) (4): P d Pt G A A 3 E t ( el, az) Gr ( el, az) Gr. G p 0 3 3 4 R Ls La P sin r ir P a cos ia P. (4) n where P d is the received power for distributed scatterers, Pt is the transmitted power, P n is the additive power, radar receiver, A G is the transmitted and received antenna gain, G p is the processor constant, R is the range spread loss, E G is the electronic gain in el is the antenna elevation angle, az is the antenna azimuth angle, ir is the local incidence angle in range, ia is the local incidence angle in azimuth, L are atmospheric (a) and system (s) losses. In order to properly determine all required geometric parameters, which are used in the radar equation and especially for the calculation of the local values a Digital Elevation Model must be inputted; for this reason the calibration is performed during the data geocoding step, where the required parameters are already calculated. The calibrated value is a normalized dimensionless number; the corresponding value in db units can be calculated by applying 10*log10. The calibrated value can be generated as Sigma Nought, Gamma Nought and Beta Nought, by setting the relevant flag in the Default Values>Multilooking-Geocoding (Backscatter Value section) (Figure 8) - 12 -
Comisión Normalization Even after a rigorous radiometric calibration, backscattering coefficient variations are clearly identifiable in the range direction. This is because the backscattered energy of the illuminated objects is dependent on the incidence angle. In essence, the smaller the incidence angle and the wider the swath used to acquire an image, the stronger the variation of the backscattering coefficient in the range direction. Note that this variation is an intrinsic property of each imaged object, and thus might be compensated, but it may not be corrected in absolute terms. In order to equalize these variations in range, a correction factor based on a modified cosine model (Ulaby and Dobson, 1989) is applied to the backscattering coefficient according to: 0 0 norm cal cos9 ( cos9 normalizationangle (5) incidencea ngle ) n Where n is a weighting factor, typically ranging from 2 to 9 depending upon the image acquisition mode (i.e the larger the incidence angle difference from the near to the far range, the higher n factor shall be set); 9 is the SAR incidence angle in the scene center; normalizationangle 9 is the pixel based SAR incidence angle (it varies depending on the Azimuth/Range incidencea ngle pixel position). The incidence angles, which are taken into account in the normalisation process, are referred to the ellipsoid. The n factor can be set in the Default Values>Normalisation Factor. If a 9 different from the scene center angle has to be used, it can be specified in normalizationangle the Default Values>Normalisation Angle. It must be noted that the same normalisation approach is followed when dealing with backscatter values expressed as Gamma Nought or Beta Nought. - 13 -
Comisión Figure 8: Top left is the Local incidence angle ( ) calculated from the DEM, Top right is the shadow and layover mask. In red is represented the layover regions. At the lowermost, is the geocoded, radiometric corrected and normalized image. ir - 14 -
Comisión 5.2.5 Filtering In order to compare the different effects of filters, we compare the Lee and Frost filters and the Anisotropic Non-linear diffusion (ANLD). The image with no filter is considered as reference. We use a subset of the entire image and took the basic statistics (Table 1). The visual interpretation gives us a first idea of how each filter operates. The ANLD is more robust in conserving the topographical forms. It has also a homogenous effect on vegetation types, which it is good to discriminate land cover types. The Lee + Frost filter is less effective for this example, with a blurry effect which can negatively impact on the classification of land cover types. When looking at the no filtered image, we can see the clear difference. It is evident the typical salt and pepper affect caused by speckle (Figure 9). - 15 -
Comisión Figure 9: Inter comparison between three processed subsets (RGB: HH-VV-HH). Top left, Anisotropic Non.linear diffusion (ANLD) Top right, Lee + Frost filter. And down no filter is applied. Table 1 shows the basic statistics for each subset, considering HH and VV polarizations. The effect of each filter practically has no difference between the subsets, but it can be seen that the ANLD has a less impact on the Standard deviation (Stdev), which is a sign of less reduction of speckle and the maintenance of forms and structures, which in our examples are the topographical characteristics of the area. For the purpose of the classification of land cover types, it is recommended to use this filter. Table 1: Basic statistics for the different filters HH VV No Filter ANLD Lee + Frost No Filter ANLD Lee + Frost Min 0 0,01 0,009 9 0,01 0,014 Max 255 5,127 3,264 255 4,6 2,263 Mean 78,24 0,102 0,102 81,07 0,11 0,119 Stdev 31,8 0,103 0,082 31,93 0,11 0,089-16 -
Comisión 5.2.6 Optical reference data The evaluation of the final SAR product is done by means of a visual comparison based on Optical IRS Orthofoto imagery (RGB= Visible). We consider the ANLD product as the most representative of the desire results, so this is the one to be compared (Figure 10) Figure 10: Comparison of the ANLD filtered scene with the IRS optical ortho image of 30m resolution. The first impression of this comparison is the effect of the topographical structures that the SAR data has, not seen in the optical scene. The subset corresponds to the borders of the Parco Nazionale dell Aspromonte, which is characterized by strong relieves and dense vegetation on its higher zones. If we pay attention to the optical scene, we can distinguish in green bright colors the most dense vegetation patches which are located in the upper zones. Turning our sight to the SAR scene, this upper zone corresponds with a purple color pattern which is the resulting combination of the HH polarization band, which is located in the Red and Green RGB false color composite. In this particular situation, the HH polarization has a stronger effect in discriminating this land covers, where the ascending path and the incidence angle, has a determinant effect. It is also important to point out, that the SAR scene was acquired in August, which as explained above, the vegetation - 17 -
Comisión conditions are almost in his maximum state. On the other hand, if we see the bright zones located in the bottom left corner of the SAR scene, we can appreciate that those are the effect of layover. The same regions are seen in the optical scene, represented with bright values. This is an indication of the altitude and bare soils of the regions, probably characterized by steep zones. These results aim to work with SAR and Optical data in a combined way. The representation of topographical features and the vegetation characteristics is a very important SAR output that can be integrated with optical data. References Meier E., Frei U., and D. Nuesch: "Precise Terrain Corrected Geocoded Images, SAR Geocoding". Data and System, Wichmann Verlag, 1993. Holecz F., E. Meier, J. Piesbergen, and D. Nüesch: "Topographic effects on radar cross section, SAR Calibration Workshop". Proceedings of CEOS Calibration Sub-Group, ESTEC, Noordwijk, 1993. A Family of Distribution-Entropy MAP Speckle Filters for Polarimetric SAR Data, and for Single or Multi-Channel Detected and Complex SAR Images Metodologías de Pre-procesamiento y Procesamiento Utilizadas en el Tratamiento Cuantitativo de Datos SAR Para el Estudio de Ambientes en el Bajo Delta del Río Paraná, Argentina ANEX Glosary* Acquisition Mode DEM DGM Product GEC Product GTC Product Incidence angle Level 0 data -Spotlight (Enhanced Spotlight, SMART) -ScanSAR (WideRegion, HugeRegion) -Stripmap (Himage, PingPong) Digital Elevation Model. Terrain height data given on a regular map grid. Synonymous with Level 1B Product (Detected Ground Multi-look) Synonymous with Level 1C Product (Geo-coded ellipsoid corrected) Synonymous with Level 1D Product (Geo-coded terrain corrected) It is the angle measured between the slant range direction and the normal to the tangent plane to the Earth surface in the specified point on ground L0 data (i.e. raw SAR telemetry) consists of time ordered echo data, obtained after decryption and before unpacking, and includes all UTCdated auxiliary and ancillary data (e.g. orbit data, satellite s position and velocity, geometric sensor model, payload status, calibration data) - 18 -
Comisión required to produce the other basic and intermediate products. Level 0 product consists of time ordered echo data, obtained after decryption and decompression (i.e. conversion from BAQ encoded data to 8-bit uniformly quantised data) and after applying internal calibration and Level 0 error compensation; this product shall include all the auxiliary data (e.g.: Product trajectography, accurately dated satellite s co-ordinates and speed vector, geometric sensor model, payload status, calibration data,..) required to produce the other basic and intermediate products. Level 1A products (also indicated as Single-look Complex Slant (SCS)), Level 1A consist of SAR focused data internally radiometric calibrated, in zerodoppler slant range-azimuth geometric projection, left at natural geometric Product spacing with associated ancillary data Level 1B products (also indicated as Detected Ground Multi-look (MDG)), consist of SAR focused data internally radiometric calibrated, de-speckled, Level 1B amplitude detected, projected in zero-doppler ground Product range-azimuth onto a reference ellipsoid or on a DEM, resampled at a regular spacing on ground with associated ancillary data. Level 1C class of products (also indicated as Geocoded Ellipsoid Corrected (GEC)) is constituted by input data projected onto a reference Level 1C ellipsoid chosen among a predefined set, in a regular grid obtained from a Product cartographic reference system chosen among a predefined set with associated ancillary data Level 1D class of products (also indicated as Geocoded Terrain Corrected Level 1D (GTC)) is constituted by input data projected onto a reference elevation Products surface in a regular grid obtained from a cartographic reference system chosen among a predefined set with associated ancillary data Local The angle between the radar beam center and the normal to the local incidence topography. The difference between the global incidence angle and the angle terrain slope. Of a SAR, the angle from the nadir at which the radar beam is pointed. Of Look angle a target, the angle between the SAR-nadir and SAR-target lines. Raw Product Synonymous of Level 0 Product SCS Product Synonymous of Level 1A Product * Extracted from the COSMO-SkyMed SAR Products Handbook. Standard Processing Levels: Level 0 (Raw) Level 1A (SCS): Single Look Complex Slant A) Focused data in complex Slant in Slant Range and zero Doppler projection. B) L0 to L1A: Gain receiver compensation Internal Calibration - 19 -
Comisión Data focusing Statistics estimation Data formatting into output Level 1B (MDG): Detected Ground Multilook A) Focused data, Detected, Radiometrically equalized and in ground range/azimuth proyection. B) L1A to L1B: Multilooking for speckle reduction Image detection (Amplitud) Elipsoid ground projection Statistics evaluation Data formatting Level 1C (GEC) Geocoded ellipsoid corrected A) Focused data, detected geolocated on the reference ellipsoid and represented in a uniform pre selected cartographic presentation. B) L1B to L1C: Multilooking for speckle reduction Ellipsoid map projection Statistics evaluation Data formatting Level 1D (GTC) Geolocated Terrain Corrected A) Focused data projected into a reference elevation surface in a regular grid obtained from a cartographic reference system. The image scene is accurately (x,y,z) rectified onto a map proyection with GCP and DEM. B) L1C to L1D: GEC processing + DEM for map projection. - 20 -