lily-pad VS SoilMachine

As for CPU vs GPU, I would never presume to judge your preferences here. You are absolutely right that it is easier to write for the CPU, and moreover, it is much easier to write code that can't be massively parallel. If time or possibility are your limiting factor, CPU is definitely the way to go. In fact, I was recently looking at this neat project that simulates real time fluids with solid objects obstructing the flow. It is done on the CPU, and I have no idea if it would be possible on the GPU. I even started my own simulation as a parallel CPU algorithm. I did not port it until I had a very solid understanding of what I was doing, but I will say the performance increase was truly astounding. I went from managing 200-300 iterations in 20 seconds on 12 SMT CPU cores, to processing 10,000 iterations with 16 times the data in just a few seconds. So if you have something that can be ported to the GPU, there is a lot to gain from it. This kind of performance boost can make impractical applications practical.

I have another program SoilMachine, which re-implements lakes with a new time-deferred automata system running on the layer map. If you wanna know how it works, you can read here. SoilMachine is built specifically to allow for multiple soil types with different properties.

I have a working simulator which is built for multiple layers called SoilMachine. Here is an article about how it works.

You can find an article explaining how this works on my blog here and a terrain-based implementation here. For more detailed source code on the LBM method, check out the examples here.

You can find the source code here (project name SoilMachine, reference implementation).

The main application I am looking at is doing procedural weather on simulated terrain to seed hydraulic and wind erosion systems with this repo.

In the future, with a good groundwater system you can distribute plants with different root properties realistically as a particle system and use this to further influence erosion parameters.

You can find the source code for this project here.

Sure, you can find the code for this terrain generator here. It's not done yet, I am currently experimenting with multiple water tables / ground water. And the GUI is incomplete - you can't elegantly define initial terrain and stuff yet or export. But it runs and you can choose the sediment parameters! It's a WIP.