Friday, February 3, 2017

Plate Tectonics - 9

As I mentioned in my last post on this topic, I believe the erosion algorithm for the Plate Tectonics simulator needs to be tweaked.  I've been analyzing the current erosion code and discovered that unlike the original version of PlaTec I worked with that basically used a smoothing algorithm, the version I'm working with now looks to be using the water drop erosion algorithm, akin to what is described as hydraulic erosion in Musgrave's work. In addition to the erosion proper, the function that implements the erosion also introduces some random noise (variations in elevation) every time the erosion function is called. The difference is "splotchy" without random variation vs. "grainy" with random variation, as seen in the image below.  Have I discovered the need for yet another parameter?

The current set of parameters is already pretty large. I'm still working all of these parameters into the code base. A lot of them are currently constants. Now they need to go into/come out of the parameters window and get passed to the simulation code.

Parameter chaos.  Perhaps grouping them into labelled groups and changing some into sliders will improve things?
In many case, many times just leaving the values as-is works great.  But not always.  It depends on your goal.
And I still need to back and fix the problem that I posted about previously, the "caterpillars" and over-eroded mountains.  That's why I started looking at the erosion code in the first place.  I basically want the land bordering on ocean to erode faster than the mountains.  The "caterpillar" islands form from too little erosion of islands in the oceans formed by plate movement.  Attempting to prevent the "caterpillars" by increasing the frequency with which erosion is applied causes the mountains to erode too quickly instead.

Conceptually, I think it is as simple as a scalar factor for the amount of material to move when the erosion algorithm moves material from a "high" cell to neighboring "low" cell(s).  My first approach would be that the scalar could equal the sum the number of neighbor cells that are below sea level, divide by four (von Neumann neighborhoold, not Moore), plus 0.5.  Hence a "high" cell that was inland would erode at half the nominal rate, a single-cell island would erode at 1.5 times the nominal rate, and cells belonging to larger islands or on the coast would erode at an intermediate rate. This actually makes some sense within the context of the simulation, as well, as wave action and rainfall would result in more erosion than rainfall alone. We'll see what it takes to implement that once I fully understand the existing implementation of the erosion algorithm, then how well it works in practice.

No comments:

Post a Comment