No there's no way it could render that arch, except if it was modeled there (might be, one day). But also the cliffs aren't there - the resolution of source height data of 76m is not enough, the slope you can see there spans exactly 76m.
The solution is either to use higher-resolution data for these parts (or generally for the mountains where the shapes are characteristic and this problem pops up), or specifically mark the nodes so that the fractal algorithm knows it should use a different path, one that creates steeper cliffs. It will be probably a combined approach, the latter part creating also the characteristic erosion patterns for cliffs.
Also there's another problem with data for coastal regions - the shallow water parts actually contain heights >0 so it makes land when there should be water. The input data will have to be fixed.
Lastly, there's a bug with water plane at the horizon resulting in those black line segments.
Ayers Rock in northern climate:
Generally, terrain features below 76m miss proper erosion patterns, but the fractal algorithm will be handling it later by using pattern masks specific for the terrain types (cliffs, icebergs, canyons and different mountain types)
