This is the second version. For each tested interval it estimates the maximum possible terrain height by linear extrapolation of the interval border heights. The mere arbitrary parameter is the maximum possible slope, which might arise within the fractal mountain area. This algorithm is much faster than the previous one (at this magnification, it is somewhat slower at other scales, but not significantly, such that the overall performance is better), and does miss significantly less ridges. The great change of the appearance is easily visible in this image. The background, that is same in both cases, proves that both sceneries are identical.

Please not the fine ridge and its shadow at the lower left of the image.
Computation time: 000d 02h:59m:10s