Neither B3D terrains, nor meshtiles are suited for your task. They are well suited for games, but not for simulation.
Use just one mesh with several layers of LOD(starting with full LOD), and then rebuild the mesh at regular intervals when needed. This would enable you to render one mesh(at LOD 5 perhaps), but at the same do the actual simulation on a high fidelity model(at LOD 0 perhaps), visualising it on the less accurate model.
In stead of storing the data in a single file, you could store each datapoint in a separate file, with height as well as all other data and references to the datapoints to the north, south, east and west. It would take much more nonvolatile memory, but would make it easier to expand the area of interest later.
The aim is to make a system which will not be hampered by artificial boundries like the area that your fire brigade covers etc, while rendering it with various levels of LOD.
I haven't done much fluid simulation, but I would probably simulate fluids by creating a number of particle emiters, emitting a large number of particles, and then model the path of the particles.
Use just one mesh with several layers of LOD(starting with full LOD), and then rebuild the mesh at regular intervals when needed. This would enable you to render one mesh(at LOD 5 perhaps), but at the same do the actual simulation on a high fidelity model(at LOD 0 perhaps), visualising it on the less accurate model.
In stead of storing the data in a single file, you could store each datapoint in a separate file, with height as well as all other data and references to the datapoints to the north, south, east and west. It would take much more nonvolatile memory, but would make it easier to expand the area of interest later.
The aim is to make a system which will not be hampered by artificial boundries like the area that your fire brigade covers etc, while rendering it with various levels of LOD.
I haven't done much fluid simulation, but I would probably simulate fluids by creating a number of particle emiters, emitting a large number of particles, and then model the path of the particles.