Suppose we have a latitude/longitude n by n grid where at each point there are p atmospheric variables (like temperature, humidity, etc.). What would be the best way to reduce this to at least an n by n array, to then flatten it into a 1 by nn array.
The end goal is to use this as conditioning in a diffusion model, and I am trying to stick as close as possible to the tokenized format of text prompts.
PCA, UMAP, or something else?