When you say you encode different values in rgba channels, are you actually saving the rasters as image files then?

The previous poster is likely referring to something like a GeoTIFF, which more or less accepts an arbitrary number of bands (matrices).

Yep, for my purposes I tend to use pngs for simplicity, but the next step would be GeoTIFF

Isn't modeling turbulence in two-dimensional completely different phenomena from (real-life) three-dimensional turbulence?

I thought this is why it's impossible to simplify turbulence models by 'ignoring a dimension'?

for taking a computational fluids approach, I think yes, you're largely correct. Hurricanes have a class of models that are parametric, where you can get a 2 dimensional wind footprint based on some initial conditions; primarily central pressure, size (radius of max wind), and location. These models rely on empirical measurements to get simpler. NWS 23 (0) is an old model, it's the one I've implemented before. But there's been more refined parametric models that have come out, and updates to older ones, for example the work by Holland (1) and Wiloughby (2). Take this with a grain of salt, since I'm no expert in this field.

(0) https://library.oarcloud.noaa.gov/noaa_documents.lib/NWS/TR_... (1) https://opensky.ucar.edu/islandora/object/articles%3A17282/d... (2) https://journals.ametsoc.org/view/journals/mwre/141/6/mwr-d-...