Is there a way to convert UHD 4:2:0 video to HD 4:4:4?
Why do this?
I shoot green screen video. It takes longer to chroma key UHD video than HD, and my final output is HD anyway. So if I can convert UDH 420 to HD 444 and then chroma key, I should get the same result as chroma keying in UHD then exporting HD, except it should run faster.
Chroma keying creates an alpha channel in the footage in your timeline, so you don't need a source file with an alpha. It's just extra bytes that serve no useful purpose.
The only time you'd need to import a file with an alpha channel is if you intend to composite it as-is (no keying). The only time you'd need to export a file with an alpha channel is as a digital intermediate that will be composited elsewhere.
From what I undestand, you'd need to convert your YUV video to 444 RGB, in 16 or 32 bpc environment, do a bicubic downscale to HD with full RGB dithering and capture in a 444 RGB video codec. The luma will scale from 8 to 10 bit and the chroma from 4:2:0 to 4:4:4. Note: this won't fix banding in skies because it is a local domain transform only and is not aware of group pixel issues.
Prores and Cineform have a 444 RGB codec but your file sizes will be massive, perhaps even bigger than the original you are trying to save space with. and if the dither algorithm isn't perfect, you won't gain any advantage because normally, a 4:2:0 4k image will key better than a 4:2:2 image due the 4x more luma per chroma information.
As to getting the same power of 444 in a downscaled image, in a perfect world, you might only lose 5% information. This is because Prores and Cineform aren't truly lossless and you would need a 100% uncompressed video codec with ridiculously large file sizes like 16 bit TIFF/PNG/EXR. This is why you don't hear about people using it, its just not practical. If you can live with a 5% loss, it might be 15-20% better than 4:2:2 of same resolution, but all of this is a rough estimate and your mileage may vary.