The script described here is a good starting point: https://rpubs.com/richardneilbelcher/isovist
I tried it on the generic example from the article and it works, so we can adapt it to our data using river edges instead of the central line and buildings instead of the polygons used in the article.

River space

• Data:

  • OSM buildings

  • river edges

  • corridor boundary to adjust extent

• Algorithm:

  • Using the first row of buildings next to the river and the river edge, determine the continuous space within a 500m buffer. This can be done with a viewshed from equally distanced points along the river edge. this might involve performance issues; optimisation of distance between points might be needed. Extent clipped by corridor boundary.

  • Optionally, the output can be segmented.

  • Can be optimised later to include obstacles other than buildings

• Output:

  • One continuous polygon or segmented polygon