@askaniy.bsky.social I don't remember the details since I wrote the code a long time ago and parts of it are based on code I found somewhere. But this is from comments in the code: "ML_XYZtoRGB; D50/2 white point". So the conversion seems to be via CIE XYS space.

(4/4) In the improved images I also used a better method to combine the color channels to get an RGB image. Instead of 'channel mixing' to get synthetic red & blue I used Jupiter's visible light spectrum to estimate the spectrum at each point in the map and then computed RGB color from the spectrum.

(3/4) Compared to real green images, the green error drops down to ~25% of the error one gets when using the simpler weighted average - a major improvement. Color variations are more pronounced in the improved versions. In particular, variations in the strength of the red colors are more obvious.

(2/4) Because there is no green data, synthetic green images must be generated. Typically a weighted average of orange and violet has been used but in the improved version of the map I used a much better (and more sophisticated) way to generate synthetic green data.

(4/4) In the improved images I also used a better method to combine the color channels to get an RGB image. Instead of 'channel mixing' to get synthetic red & blue I used Jupiter's visible light spectrum to estimate the spectrum at each point in the map and then computed RGB color from the spectrum.

(3/4) Compared to real green images, the green error drops down to ~25% of the error one gets when using the simpler weighted average - a major improvement. Color variations are more pronounced in the improved versions. In particular, variations in the strength of the red colors are more obvious.

(2/4) Because there is no green data, synthetic green images must be generated. Typically a weighted average of orange and violet has been used but in the improved version of the map I used a much better (and more sophisticated) way to generate synthetic green data.