I'm not into those things, but you can check
here for more info. Tiling order aside, you can get normal map like this from channels of original RGBA image:
norm_r = orig_b
norm_g = orig_a
norm_b = filled with 255
Heightmap would require some math, but I guess it's also possible to convert normal map into heightmap if needed.
Elevation and normals are packed into 512px × 512px × 32bit tiles. The first byte and the fourth byte are the 8bit coefficients of the normal. The second and third bytes encode a 16bit elevation/height.
I've already managed to stitch these tiles together, since I've figured that the map is 8x8 tiles, with each tile being 512x512px.
So, if the image is in RGBA, RA is the normal map, and GB are the elevation map?
EDIT: Below are the results my code naively combining the images together. I'll note that all of them are very noisy and don't look like normal or height maps. (Note how the purple and green in the original alternate from green to purple from lower elevation to higher elevation.)
Full stitched original:https://pomf2.lain.la/f/iycsr6xq.png"Normal" map: (takes B and A channels)https://pomf2.lain.la/f/vixw2qy8.png(Very much not a) Height maphttps://pomf2.lain.la/f/7paw6eca.png