I'm working on a generalization process of probabilistic calculated raster data (rockfall extend, data shows probability of being hit by a rock in %; I used binary data, 1 for "one hit or more", 0 for "no hit").
Output should be well legible without "single lines" and if possible very smooth edges (you can see the many single lines in the input picture). As it will be used for batch processing, a solution as automatic as possible is prefered.
To some extend I got satisfying output data using resamplings, neighborhood analysis, vectorization or even just deleting values < 1% (these might be regarded as statistical mistake). You can see this in the following picture.
But still, there are some logical mistakes, some edges are not well rounded or do not really fit to e.g. valleys.
Thus, I'd like to connect the binary data with the DEM. I added a picture with contour lines to make more reasonable, what I try to explain. I wish that the edges follow a more or less logical line especially on the "lower" left line of the biggest patch.
I use QGIS, SAGA, GRASS. I tried some other tools and plugins (e.g. GDAL sieve), which also didn't help me. Further I considered using GRASS GIS watershed tools, but I wasn't able to connect this to my binary patches and gain proper data.
Finally, I am looking for a "GUI-solution".