Earth observation data providers are using different synthetic aperture radar (SAR) processors to deliver Sentinel-1 normalised radar backscatter (NRB) products. Radiometric and geometric inconsistencies across NRB products are caused by different approaches within SAR processors and how they are configured by data providers. This includes differences in algorithms to project data from radar to ground coordinates, digital elevation models, multi-looking factors and speckle filtering. Consequently, evaluating the quality of available Sentinel-1 NRB products is convoluted and further confounded by the dependance of radiometry on geometry. To assist end users, this study developed an evaluation framework which held configurable parameters for three commonly used SAR processors (SNAP9, GAMMA and ISCE-3) constant, revealing their inherent differences.
The evaluation framework relates methodological differences in SAR processors to quantitative evaluation metrics for geolocation, radiometric and topographic quality. We apply this framework to three analysis ready data (ARD) workflows, configured to produce comparable NRB products; OPERA (ISCE-3), pyroSAR (SNAP) and pyroSAR (GAMMA). To further inform users, we also assess products generated from the widely used Hyp3 (GAMMA) and s1ard (SNAP) ARD workflows and consider the impact of atmospheric corrections.
When evaluating the geolocation quality of NRB products using the Surat Basin corner reflector array we find significant differences across the SAR processors. The OPERA (ISCE-3) ARD workflow delivers the most geometrically accurate and precise NRB products with a mean absolute error (MAE) of 6.53 +/- 1.24 m. SNAP based NRB products are the least geometrically accurate and precise with MAE of 14.26 +/- 5.30 m for the pyroSAR (SNAP) workflow. The geometric differences are mostly caused by procedures for handling digital elevation models (DEMs) and algorithms to project data between radar and ground co-ordinates. The inclusion of atmospheric corrections results in improved geometric accuracy (but not precision) with a reduction in the OPERA (ISCE-3) NRB workflow MAE to 2.47 +/- 1.22.
Next, we quantify radiometric differences over a gently sloping area of an Australian Queensland rainforest to minimise the impact of geometric differences. Minimal radiometric differences are observed between the OPERA (ISCE-3), pyroSAR (SNAP) and pyroSAR (GAMMA) workflows, and all NRB products agree with published values for similar land classes. The use of a larger multi-looking factor by the s1ard (SNAP) and Hyp3 (GAMMA) workflows disproportionally dampens dark speckle noise, which reduces the range disproportionately towards lower NRB values. These results highlight the impact of multi-looking on radiometric differences and the interpretation of radiometric differences in the decibel scale.
Finally, we quantify topographic quality over a rugged area of an Australian Queensland rainforest using the VH channel which is dominated by volume scattering. Between moderate local incidence angles of undulating terrain, relationships are comparable across workflows and are close to linear, indicating good performance in radiometric terrain correction. However, large differences are found over extreme terrain where layover and shadow persist. This is caused by differences in how the workflows perform the radiometric terrain correction, the geometric effects of atmospheric corrections and likely terrain deviations from the DEM. As a result, we recommend the use of layover and shadow masks to remove questionable data.
These results have been used to guide the construction of a new NRB product to be produced over Australia and Antarctica by Geoscience Australia. The evaluation framework will continue to be developed under the Committee for Earth Observation Science and will be accessible via a public GitHub repository. The application of the framework to other areas and NRB products will continue to build the community’s understanding of the performance of different SAR processors.
Presented at the 2025 IEEE International Geoscience and Remote Sensing Symposium.
