Your main concern is about vertical distribution of the climate? For that, there are simpler ways of modeling which doesn’t require CFD. In fact, that is already taken into account by many existing sophisticated-enough crop models.
CFD is a costly overkill if that is the primary concern.
I believe that dynamical systems combined with machine learning scales while CFD doesn’t.
Here are why. Compared to a time-derivative dynamical system, CFD models the climate distribution spatially as well. This is advantageous if you have an environment in which the state variables (note
light is not a state variable) such as temperature, humidity, CO2 concentration, etc. are highly non-uniform across the environment.
- Are they indeed non-uniformly distributed? Yes
- Are they highly nonuniform? Depends on the farm structure (and less so if you’re talking about a well-built farm in the middle east)
- Can that problem be corrected by real-time data and machine learning? Absolutely.
And there’s significant cost for doing CFD. Because it models the spatial distribution, you’re required to model the geometry of the farm (aka
mesh generation) as well. Mesh generation is costly and doesn’t scale from farm to farm.
Despite so, CFD for greenhouse modeling is being investigated by a few research labs.