What sort of transformations could make a large sum of data of 3D arrays more compressible with zstd?

Question

There is not much sources on this,I would like ideas and ways of thinking that will enable my understanding.

Answer 1

Possibly nothing, unless there are patterns in your data to exploit. If your data is random numbers, then it will simply not be compressible no matter how you transform it.

As an example, if the entries of a 3D array are similar to each other along a given dimensions, then transposing the array (if necessary), so that elements in that dimension are adjacent could improve compression. If those elements are numeric and tend to be close to each other, then delta coding could improve compression.