A new paper on incident angle (IA) variation of texture features for sea ice classification has been accepted in the Remote Sensing special issue on Remote Sensing of Sea Ice and Icebergs:

“Incident Angle Dependence of Sentinel-1 Texture Features for Sea Ice Classification” by J. Lohse, A. P. Doulgeris and W. Dierking. Remote Sensing 2021.

The full text can be accessed here.

Image texture is often used in automated sea ice classification to resolve ambiguities in the backscatter intensity from different ice types as well as the influence of varying sea states on the backscatter intensity from open water areas. The incorporation of texture into a classifier that accounts for per-class IA variation of its features (Lohse et al., 2020) requires knowledge of the dependence of texture features on IA. In this paper, we investigate this dependence for different ice types and a range texture parameter settings.
We find that texture should preferably be calculated from backscatter intensity in dB – the variation with IA is then almost negligible. For our training and validation data set, this is true independent of the tested parameter settings. However, the parameter settings do affect the separability of the different ice types. Most importantly, we find that larger texture window sizes improve class separability. As a larger window size reduces the effective spatial resolution of the results, there is an inherent trade-off between classification accuracy and spatial resolution.
The main improvements of including texture features in addition to backscatter intensities are in a generalized separation of sea ice and open water, as well as the correct classification of multi-year ice against young ice with frost flowers in refrozen leads.