Post-processing generates the output SLC and GRD products as well as quicklook images. Post-processing consists of post-processing range processing, post-processing azimuth processing and post-processing output processing as shown in the figure below.
The processing is applied to each sub-swath for IW and EW and for each azimuth block consisting of a burst for IW and EW and an entire vignette for WV.
 |
Post-Processing Range Processing
Post-processing range processing is performed on each range line within an individual burst of a sub-swath for TOPSAR or vignette for WV.
Post-Processing Azimuth Processing
Post-processing azimuth processing is performed on each azimuth line for each range look within an individual burst of a sub-swath for TOPSAR or vignette for WV.
Post-processing GRD Output Processing
For GRD products, the post-processing output processing consists of optionally removing the thermal noise, de-burst and merge for TOPSAR mode, generation of quicklooks and writing to the output file format. Every data-take contains two noise acquisitions, one at the beginning of the data-take and a second acquisition at the end. In addition, noise-equivalent acquisitions are represented by the travelling echoes after each interleaved calibration sequence all along the data-take. For Level-1 GRD products only, the thermal noise contribution is estimated and optionally removed to improve the quality of the image.
The thermal noise contribution is reshaped in a range and azimuth varying fashion by the range and azimuth varying radiometric corrections applied by the SAR processor. In particular for multi-swath products, the noise can be different between swaths causing an intensity step at swath boundaries (See Thermal denoising of products generated by the Sentinel-1 IPF).
For a particular azimuth time, thermal noise is estimated in slant range coordinates as follows:
1) Calculate the range spreading loss vector.
2) Calculate the elevation beam pattern vector.
3) Apply the scalar contributing factors.
For TopSAR modes a further contribution is present along the azimuth direction (scalloping effect) and consists in the azimuth elementary beam pattern.
For SLC products, the noise estimation vectors are then only saved in the annotation to allow for later use in removal and as input for Level-2 processing.
For GRD products, the thermal noise vectors are converted to ground range coordinates and applied to the data by subtracting the noise from the power-detected image.
When calibrating the product to β, σ or γ, the noise vector must be scaled by the corresponding calibration Look-Up Table (LUT) (β, σ or γ or dn, respectively):
where, depending on the LUT selected to calibrate the image data:
Once the calibrated noise profile has been obtained, the noise can be removed from the GRD data by subtraction.
For any pixel i that falls between points in the LUT the value is found by bilinear interpolation.
During TOPSAR sub-swath merging, the noise vectors are also merged into one vector following the same strategy as for merging two adjacent ground range images.
Within the annotation of the resulting product, the noise vector annotation data set contains the thermal noise estimation along azimuth. This can be used to reverse the thermal noise correction in products in which it has been applied or can be used for applying the correction in products in which it has not.
TOPSAR Debursting and Sub-Swath Merging
The TOPSAR modes acquire data in bursts and for several sub-swaths. For GRD products, the bursts are concatenated and sub-swaths are merged to form one image.
Bursts overlap minimally in azimuth and sub-swaths overlap minimally in range. Bursts for all beams have been resampled to a common grid during azimuth post-processing.
For merging the sub-swaths, the optimal cut is determined taking half the overlap between them, considering only the valid samples of each line. Samples from two consecutive sub-swaths are put side-to-side according to the optimal cut position (without performing sample 'blending').
In the azimuth direction, bursts are merged according to their zero Doppler time. Note that the black-fill demarcation is not distinctly zero at the end or start of the burst. Due to resampling, the data fades into zero and out. The merge time is determined by the average of the last line of the first burst and the first line of the next burst. For each range cell, the merging time is quantised to the nearest output azimuth cell to eliminate any fading to zero data.
Application LUT Scaling
Application LUTs are used to apply a range-dependent gain function to the processed data prior to generation of the final image output. The application LUT scaling is used to optimise the radiometric scaling of the main feature of interest, while optimising the available dynamic range in the output product and to compensate for changes in the radar backscatter with changing incidence angles. LUT's that could be used include:
- Point target application LUTs - suited to applications involving scattering from bright points targets. Typically these LUTs provide poor quantisation over areas of very low backscatter.
- Sea, land, mixed and ice LUTs - suited to the thematic applications they describe in which low backscatter features are expected. Typically, bright targets will saturate. Values vary with incidence angle.
- General application LUT - typically very bright targets will be saturated.
During satellite Commissioning Phase and related calibration activities, a general purpose LUT has been defined for each acquisition mode, allowing to avoid saturation of both brighter and darker targets and at the same time to reduce as much as possible the quantization error.
Level-1 SLC and GRD images are then scaled to 16 bit and saved in GeoTIFF file format.
Quicklook Generation
Quicklooks are lower resolution images of the product data used to preview the data. Quicklooks are generated by power detecting, averaging and decimating in both azimuth and range directions by a configurable amount. IW and EW SLC product quicklooks are first de-burst and merged.
The Red channel is the co-polarisation amplitude, the Green channel is the cross-polarisation amplitude and the Blue channel is the ratio of Green to Red channels. So for a HH/HV dual polarisation product, Red is HH, Green is HV. Separate scaling of each of the RGB channels is used to give a false colour image where the land is green and the ocean is generally blue (or sometimes red under higher wind conditions).
Quicklook images are scaled to 8 bit and saved in PNG file format.