# SAR calibration

## Description

The Radar Product Calibration (SAR-Calib) service performs the radiometric calibration of multi-sensor SAR GRD data and computes the radar backscatter coefficient at each polarization.

graph TB c(cos2) --> d[Dataset] d --> i[input] subgraph Inputs i end subgraph SAR Calibration i --> sar((SAR<br>Calibration)) end subgraph Outputs sar --> o1[SAR<br>calibrated<br>product] end

The SAR-Calib service uses SNAP Graphs or plain matrix calculations to apply the processing steps shown in the above workflow. The output of the service is a Stac Item with the $$sigma_0$$ values provided in decibels and the overview/s product/s.

## Workflow

In the ESA Charter Mapper, the Radar Product Calibration service performs the radiometric calibration of multi-sensor SAR detected amplitude data and computes the radar backscatter coefficient at each polarization. The possible operations applicable with respect to the nature of the products are:

1. Orbit correction (if necessary),

2. Border noise removal (if necessary),

3. Calibration,

4. Multilooking (if necessary),

5. Speckle filtering (only for overview generation),

6. Terrain correction (if not already geocoded),

7. Conversion to dB,

8. Image stretching and RGB composite (only for overview generation),

9. Creation of output products.

This service runs systematically during the acquisitions ingestion process, and is primarily meant to process focused SAR data that has been detected, multi-looked and projected to the ground range using an Earth ellipsoid model (e.g. WGS84) and a scene-averaged value of terrain height (e.g. ICEYE GRD data).

The first step, Orbit Correction, is built on SNAP "Apply orbit" toolset and gathers and employs precise orbits which are necessary to improve the geocoding of SAR images. This step is built with SNAP and is currently available only for Sentinel-1 GRD products.

The second step, Border Noise Removal, is meant to clean out border noise patterns in near and far range zones from SAR data. This step is built with SNAP and is currently only available for Sentinel-1 GRD data (SNAP “Sentinel-1 Remove GRD Border Noise)1.

The third step, Calibration, performs the radiometric calibration of SAR detected image to convert DN into a backscatter coefficient (radar brightness, beta nought $$beta_0$$ which represents radar reflectivity per unit area in slant range, or the other forms of average backscatter coefficient: sigma nought $$sigma_0$$, and gamma nought $$gamma_0$$). SAR detected images are commonly calibrated into sigma nought $$sigma_0$$ which represents averaged radar cross section per unit ground area in $$m^2/m^2$$. Sigma nought includes the influence of the terrain into the backscatter signal. In the computation of sigma nought, the local terrain slope derived from the DEM auxiliary data is often employed to derive the local incidence angle $$theta$$. For some missions (Kompsat-5 or TSX-1) local incidence angle values over the entire scene can be directly extracted through the decryption of Geocoded Incidence Angle Mask (GIM) included into the source product package2. Such local incidence angles are then used to obtain Sigma nought from the source DN values rescaled with rescaling/calibration factors included into product metadata. The radiometric calibration step is built using either SNAP or GDAL according to the different types of SAR mission.

The fourth step, Multilooking, is employed to reduce the speckle present in oversampled intensity images obtained from SAR data. Averaging neighboring pixels in intensity images is in fact a common practice for noise smoothing. Indeed, this speckle effect is present because SAR images are often provided oversampled by a factor of N, and the sample-rate above the Nyquist rate, to avoid Aliasing. Thus, the multilooking factor R is computed as following:

$R = \frac {res_{range} \times res_{azimuth}} {spacing_{range} \times spacing_{range}}$

where $$res_{range}$$ is the range resolution in (m), $$res_azimuth$$ is the azimuth resolution in (m), $$spacing_range$$ is the range pixel spacing, which is the distance between adjacent pixels perpendicular to the flight path in (m), and $$spacing_azimuth$$ is the azimuth pixel spacing, which is the distance between adjacent pixels parallel to the flight path in (m). The computation of (R) allows defining the NxN window over which to average the source calibrated intensity image. Multilooking reduces the presence of Speckle in oversampled SAR intensity images.

The fifth step, Speckle Filtering, is about SAR image despeckling, which consists of a radiometric enhancement required for a better interpretation of features into input SAR images due to their typical grainy salt-and-pepper appearance. To remove granular noise in SAR imagery multiple speckle filtering algorithms are available in the literature 3, 4, 5, 6,7. This processor employs the Lee Sigma filter4,5. The Lee’s method is widely employed cause it preserves edges while filtering averages of the image. This step is currently built only using SNAP and is applied only for generating Overview products.

The step six, Terrain Correction, performs the geocoding of SAR data in slant range geometry by using a Range-Doppler approach. Terrain correction is needed to remove distortion into SAR images caused by the side-looking of SAR sensors. In this step the Range Doppler orthorectification method (Small and Schubert, 2008)8 is employed, as given into the SNAP Range Doppler Terrain Correction tool, to obtain geocoded and terrain corrected SAR data.

The seventh step, Linear to dB, consists of a logarithmic scaling of linear values of Sigma nought. This is needed because the SAR intensity can vary many orders of magnitude. Thus, after this final step backscatter values are expressed in dB.

The step eight, Overview Image Creation, generates full-resolution overviews from calibrated radar EO data. Output visual products are given as grayscale single band or RGB band composite (using multiple polarizations, when available). Hereinafter is described how RGB composites are made in this service. In the case of dual-pol data, VV&VH or HH&HV, the RGB is created with R=co-pol, G=B=cross-pol. This first representation highlights mainly urban areas, the different orientation of buildings, and vegetation. Instead, in case of full-poll data the RGB is derived as follows: R=HH, G=HV, B=VV. This second representation improves the dual-pol representation, highlighting better volumetric scattering, bare soils and urban areas. This step employs an image stretching, which is applied to consider only Sigma nought within predefined minimum and maximum values while generating the Overview image product (see Table 2).

All these eight steps are mapped in the following Table 1 which provides a complete outlook of the processing chain applied for a selection of the current Charter SAR payload. The full list of the ESA Charter Mapper supported SAR sensors is available in the Satellite configuration table. Table 1 indicates the software / tool employed in the steps required for the processing of each SAR mission. Most of the columns in Table 1, except for the Speckle Filtering and the Overview Creation, indicate the steps of the primary chain required for the creation of the physical meaning products (Sigma Nought product encoded in Float-32). Instead, the Speckle Filtering and the Overview Creation columns indicate the steps of the secondary chain which generates the visual products (Overview product encoded in UInt-8).

Mission Sensor Lev Prod Data type Mode Orbit Corr. Border noise rem. Calibr. Multi looking Speckle filter Terrain corr. Linear to dB Overview Creation
Gaofen-3 SAR-C L2A SGC Geocoded FSI GDAL GDAL SNAP GDAL GDAL
SAOCOM-1A SAR-L L1D GTC Geocoded TW GDAL GDAL SNAP GDAL GDAL
RCM-1/2/3 SAR-C GRD Geo-referenced 30M, QP SNAP SNAP SNAP SNAP SNAP GDAL
TSX-1/TDX-1 SAR-X L1B EEC Geocoded SM SNAP GDAL SNAP SNAP GDAL
Sentinel-1 SAR-C GRD Geo-referenced IW, EW SNAP SNAP SNAP SNAP SNAP SNAP SNAP GDAL
ALOS-2 PALSAR-2 L1.5G Geo-referenced SM GDAL GDAL SNAP GDAL GDAL
KOMPSAT-5 COSI L1D GTC Geocoded EW, EH, UH GDAL GDAL SNAP GDAL GDAL
ICEYE-X4/5/6/7 SAR-X4 L1B GRD Geo-referenced SL SNAP SNAP SNAP SNAP SNAP GDAL

Table 1 - General outlook about the processing steps required for the creation of physical meaning and visual products in the ESA Charter Mapper's Radar Calibration systematic service.

In the creation of all SAR products derived from the systematic SAR calibration (either the geophysical asset or the overview one at 8 bit) an image stretching is applied to consider only Sigma nought within predefined minimum and maximum values. This helps optimize visualization on the screen of the assets. The minimum and maximum values in dB included in Table 2, are specific to mission, SAR band (or frequency) and polarization.

Mission SAR-band Co-pol (VV/HH) [min,max] in dB Cross-pol (VH/HV) [min,max] in dB
ICEYE X [-22,2] [-27,-3]
Kompsat-5 X [-22,2] [-27,-3]
TerraSAR-X / TanDEM-X X [-22,2] [-27,-3]
Gaofen-3 C [-20,0] [-26,-5]
RCM C [-20,0] [-26,-5]
Sentinel-1 C [-20,0] [-26,-5]
ALOS-2 L [-27,0] [-35,-5]
SAOCOM-1 L [-27,0] [-35,-5]

Table 2 - Signal dynamic ranges used for stretching Sigma Nought during the generation of SAR overviews products. The values in dB are given for multiple missions, SAR-bands and polarizations.

## Inputs

The inputs are geo-referenced or geocoded SAR detected images from multiple sensors (e.g. SAR L1 detected products like TSX-1 L1B EEC, or GRD for ICEYE, RCM and S1).

## Outputs

Radar Calibration product specifications can be found in the following tables.

 Radar Products Calibration Product Long Name Calibrated radar backscatter amplitude at [HH, VV, VH, HV] polarization in dB Short Name s0-db-B-PP (where PP is the polarization [HH, HV, VH, VV] and B the SAR-band [X,C,L]), e.g.: s0-db-x-hh, s0-db-x-hv, … , s0-db-l-vv Description Backscatter coefficient σ_0, or sigma-nought, for [HH, VV, VH, HV] polarization derived from a multi-looked and projected to the ground range product (GRD) of the SAR focused signal expressed in dB Processing level L1D Data Type Float 32 bit Band Single for each polarization Format COG Projection Native (or EPSG:4326 - WGS84 if not map projected) Units dB

Table 3 - Product specification for Sigma Nought Product from the ESA Charter Mapper Radar Calibration.

The second type of output product (see Table 4) is a Browse image given as full resolution RGBA composite (multibands GeoTIFF in COG format).

 Radar Full Resolution Browse Image Generation Product Long Name Full resolution RGBA composite or grayscale single band image from radar EO data Short Name overview-vv, overview-vh, overview-hv, overview-hh (grayscale), overview-dual, overview-full (false color composite) Description Grayscale single band geo-referenced image from single polarization or RGBA composites from multi polarization SAR data (including alpha band). Processing level L1 Data Type UnSigned 8-bit Integer Band 4 Format COG Projection Native (or EPSG:4326 - WGS84 if not map projected) Valid Range [0 - 255] Fill Value 0

Table 4 - Product specification for Browse Image Product from the ESA Charter Mapper Radar Calibration.

1. Ali I., Cao S., Naeimi V., Paulik C. and Wagner W., (2018), "Methods to Remove the Border Noise From Sentinel-1 Synthetic Aperture Radar Data: Implications and Importance For Time-Series Analysis," in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 11, no. 3, pp. 777-786, March 2018, DOI: 10.1109/JSTARS.2017.2787650

2. Fritz, T., Eineder, M.: TerraSAR-X Basic Product Specification Document, TX-GS-DD-3302, Issue 1.5, February 2008, available at: https://tandemx-science.dlr.de

3. Lee J. S., L. Jurkevich, P. Dewaele, P. Wambacq & A. Oosterlinck, (1994), "Speckle filtering of synthetic aperture radar images: A review, Remote Sensing Reviews", 8:4, 313-340, DOI: 10.1080/02757259409532206

4. Lee, J. S., Ainsworth , T. L., Wang, Y. & Chen, K. S., (2015), "Polarimetric SAR Speckle Filtering and the Extended Sigma Filter", IEEE Transactions on Geoscience and Remote Sensing, 53, 1150-1160, DOI: 10.1109/TGRS.2014.2335114

5. Lee J. S., Wen J. H., Ainsworth T. L., et al., (2009), "Improved Sigma Filter for Speckle Filtering of SAR Imagery," in IEEE Transactions on Geoscience and Remote Sensing, vol. 47, no. 1, pp. 202-213, Jan. 2009, DOI: 10.1109/TGRS.2008.2002881

6. Lopes A., R. Touzi and E. Nezry, (1990), "Adaptive speckle filters and scene heterogeneity," in IEEE Transactions on Geoscience and Remote Sensing, vol. 28, no. 6, pp. 992-1000, DOI: 10.1109/36.62623

7. G. Vasile, E. Trouve, Jong-Sen Lee and V. Buzuloiu, (2006) "Intensity-driven adaptive-neighborhood technique for polarimetric and interferometric SAR parameters estimation," in IEEE Transactions on Geoscience and Remote Sensing, vol. 44, no. 6, pp. 1609-1621, June 2006, DOI: 10.1109/TGRS.2005.864142

8. Small D., Schubert A., (2019), "Guide to S1 Geocoding", UZH-S1-GC-AD, Technical Note, Issue 1.10, 26.03.2019, 42p, available at: https://sentinel.esa.int