Hi Reno,
I know it may sound a bit counter productive, however I had similar situation: coverage of 1sqKm of archaeological site area, with olympus e-pl5 camera. Flight overlap (which is always a thing to consider) was 80% forward, 60% sideways, forested area (no running water though).
I would have preferred more side overlap, however I was just hiring a crew and I presumed that they knew well enough. Photos were always in nadir, and all bad ones have been eliminated.
On multiple occasions I failed generating aligned data with high parameter, with more settings used than i care to remember. What was successful was, setting alignment on medium, with some standard keypont/tiepoint relation. Now> I am still in procedure of establishing if this produced good quality, or good enough quality dataset. Perhaps you would like to try it out. So far, I saw no reason to be dissatisfied - i believe that using GCP and bundle adjustment after the alignment operation, corrects any erroneous reconstruction.
And it also, by the looks of screenshot, you might have issues with overlap of the photos. More dense camera network should solve the things.
In any case, test some options, im sure there will be a simple solution.
best regards