suggestion for Asobo/Blackshark:
machine learning with
- water polygon overlay in satelite/air image
- detect color range inside watermask
- detect foreign colors too different to average image colors within water mask
–> those are small islands, rocks, etc. and should be masked out of water polygon - detect similar colors to average within watermask, outside of watermask (maybe using a max range of deviation, like 500m)
–> these are areas lacking water polygon
–> create water polygon based on raster pixel… and merge them together with the original watermask (if necessary)