Sentinel-1 SAR calibration: VH band deviation

We successfully integrated Sentinel-1 backscatter computation based on OTB in our openEO backend: https://github.com/Open-EO/openeo-geopyspark-driver/blob/master/openeogeotrellis/collections/s1backscatter_orfeo.py#L327

The workflow is derived from S1Tiling, and uses SARCalibration and Orthorectification to compute Sigma0 for a given area of interest.

We now compared the results to two other implementations. This shows that the correspondence is very good for the VV band, but not entirely exact for VH. This is shown in the figure below. Would it be possible to improve VH to get an equally good result as VV?

OTB version: 7.2.0, running on CentOS 7.