davidson16807
commented
9 years ago The problem here has been a lack of models that could very precisely describe the supercontinent cycle. It's easy enough to say that supercontinents breakup due to heat build up and form from extroversion/introversion. Its another thing entirely to make predictions out of that. The model needs to very precisely express the conditions that trigger supercontinent breakup, in addition to the method by which continents disperse. This includes timings, speeds, directions, etc.
Right now the model uses the number of plates as a trigger for breakup. This is good because it guarantees there are never more than a certain number of plates on the planet, which saves on computer resources. This is bad because there is no guarantee on when breakup occurs. On average, the time to breakup is usually 0.75 - 1 billion years, which is much higher than what's observed on Earth. Still, spending so many processor cycles to simulate an uninteresting planet doesn't matter to me so long as the interface remains responsive - its the framerate that matters, and if plates get out of hand then framerates tank.
Another approach is the empirical way of doing things - the cycle repeats every 300-500 My, so base your plate speeds around that. Problem with that is now your plate speeds don't match reality. Mixed in with this is the fact that the simulation uses the extroversion model of supercontinent formation. The extroversion model has classically encountered problems with explaining how supercontinent cycles are so frequent. Its only used in this simulation because its so much easier to describe than the introversion model. The introversion model requires plate speed to be modeled by some nebulous sinusoidal function that's never provided in the literature, whereas extroversion lets me say the plates move at constant speed.
The important part:
I found a recent paper in Nature that discusses an alternative model for the supercontinent cycle (Mitchell Kilian Evans 2012). Its very much on the cutting edge, but it appears to be well grounded by observation. It makes the very concrete prediction that the center of a supercontinent will form 90 degrees away from the center of its predecessor. If I know where a supercontinent will form, I can base trajectories around that have a reasonable expectation that collision will occur. Extroversion/introversion also implicitly make predictions about where the center would be, but this is still with all their aforementioned downsides, plus I'd expect these predictions to have much less precision than what we see in the data. The orthoversion model has the potential to deliver realistic cycle durations and still maintain realistic speeds.
The problems I see with implementation are mostly technical and have to do with using 3d: finding a random point that is 90 degrees away from another given point, and then finding a set of speeds/euler poles that describe a path from one point to another that occurs in a given amount of time. I think I've got the first problem down. As for the second problem, I'm open to suggestions.
redferret
Didn't know where else to put this, so here goes.
The plates need to split more frequently. As it is, the plates tend to quickly settle in to a steady state with one gigantic slowly growing superplate and a small number of other plates, generally connected by transverse boundaries.
The result tends to be a dull, generally permanent supercontinent that usually survives the infrequent rifting events. Even the real supercontinents seem to have consisted of chance convergences of continental crust on multiple plates rather than a single hemispheric godzilla plate.
Other realism arguments: Heat build up under very large areas of thick continental crust seems likely to drive strong divergent flows that would break up such concentrations of continental material. My understanding is that the plate pattern follows an underlying convectional structure; it seems unlikely that a singly convection cell could form that is that large.
In addition to the realism arguments there is the aesthetic and technical argument against spending so many processor cycles regenerating what are essentially static maps with little interesting change.
The first solution would be simply to increase the probability or frequency of global splitting events. This is probably the easiest option to program, but aesthetically and realistically not preferable. Although supercontinents may realistically drive such concentrated rifting parties, in general it is probably more realistic if plates split and fuse individually over time.
The second solution, and, I think the better one, would be to have each plate check periodically for rifting events. Older and larger plates would be more likely to split than younger and smaller plates. This could be dealt with either by incrementing or decrementing each plate's cell_count whenever there is an interaction that involves gain or loss of cells. Another way would be simply to keep track of the age of each plate and adjust split probability up as it ages.