isomage

11-10-2011, 03:39 AM

Hello again, cartographers!

After a year-long, minecraft-induced hiatus from all things mappy, I started once again thinking about my old tectonic plate fakery (http://www.cartographersguild.com/showthread.php?4737-Tectonic-plate-fakery), and decided to see if I could improve matters; since I think I have improved matters, I thought I'd share the results.

As before, we start off with several randomly chosen points (I'll call these "nodes") on the sphere. This time, for each point on the sphere, we find the angular distance of the point from its nearest node, and normalize that value so that it is equal to 1 at the nearest node and drops to zero when it's equally close to that node and the second-nearest, after perturbing the point with some random noise; then that resulting value is used as a multiplier against a different noise field -- a ridged multifractal -- to obtain the elevation at that point. Thus we end up with the ridged multifractal field more or less untouched when we're close to a node, and fading off in elevation down to sea level between nodes.

There are two key points here: first, each node defines the center of a "continental plate", and neighboring continents, sharing a boundary, will occasionally have identifiable coastlines, as though they had broken apart; and second, since the positioning of the ridges is somewhat independent of the underlying continent position, the mountains will occasionally be coastal, rather than always being at the centers of the land masses as they would be if we used a simple height field only. Hence we can have some sense of an earlier supercontinent, and we get more realistic mountain ranges.

Here are some examples (the appearance of continental shelving is due only to the color palette and elevation-dependent shading):

After a year-long, minecraft-induced hiatus from all things mappy, I started once again thinking about my old tectonic plate fakery (http://www.cartographersguild.com/showthread.php?4737-Tectonic-plate-fakery), and decided to see if I could improve matters; since I think I have improved matters, I thought I'd share the results.

As before, we start off with several randomly chosen points (I'll call these "nodes") on the sphere. This time, for each point on the sphere, we find the angular distance of the point from its nearest node, and normalize that value so that it is equal to 1 at the nearest node and drops to zero when it's equally close to that node and the second-nearest, after perturbing the point with some random noise; then that resulting value is used as a multiplier against a different noise field -- a ridged multifractal -- to obtain the elevation at that point. Thus we end up with the ridged multifractal field more or less untouched when we're close to a node, and fading off in elevation down to sea level between nodes.

There are two key points here: first, each node defines the center of a "continental plate", and neighboring continents, sharing a boundary, will occasionally have identifiable coastlines, as though they had broken apart; and second, since the positioning of the ridges is somewhat independent of the underlying continent position, the mountains will occasionally be coastal, rather than always being at the centers of the land masses as they would be if we used a simple height field only. Hence we can have some sense of an earlier supercontinent, and we get more realistic mountain ranges.

Here are some examples (the appearance of continental shelving is due only to the color palette and elevation-dependent shading):