using 32bit TIFF or 16bit RAW for post-processing

Question

What are the advantages of using a 32bit TIFF over using a 16bit raw file?

Answer 1

The main advantage is that 32-bit TIFF is a picture as opposed to being raw sensor data. Raw file as-is looks really terrible, not like a photograph at all. In order to make it look like a photograph, you must do heavy processing. TIFF as an image format has all the processing steps already done.

The disadvantages of 32-bit TIFF are two-fold:

• 32 bits per channel is way too much, if you convert a raw file typically there's only 16 bits and storing it as 32 bits at least doubles the disk space usage (actually it would probably triple it after demosaicing too so you probably have 6x the disk space usage). Converting from 16 bits to 32 bits doesn't insert any new quality. You can't make a camera sensor magically have 32 bits of dynamic range by converting the 16-bit raw file into a 32-bit TIFF file.
• TIFF has all the preprocessing done, so you are fixed with one set of parameters for interpreting the raw file. If you want to change denoising settings for example, it's much harder to do it afterwards (technically you could do a second denoising run, but it's better to do only one denoising run with the parameters you want than to do denoising after denoising, and if the denoising in the first place was excessive you can't reduce the denoising amount). Exposure adjustment might be possible if the photo is underexposed as 32 bits doesn't lose any dynamic range, though.

I'd say use 16-bit TIFF instead if you want a lossless image for editing. Retain the raw files though if you want to recreate that image with different raw processing settings. Finally for distribution you probably want to convert to JPEG or HEIF.

Answer 2

There aren't really any advantages to using a 32-bit TIFF instead of a 16-bit raw file for post processing, assuming the 32-bit TIFF came from processing a 16-bit (or, more likely, a 14-bit or 12-bit) raw file.

When you convert raw sensor data to TIFF, you "bake in" whatever demosaicing algorithm was used (interpolating color information from monochromatic brightness values based on the differences of adjacent photosites filtered by one of three color filters in the Bayer mask). You also bake in black point and white point to the extent that you can't re-expand either one after the fact. What you've discarded below you black point and above your white point is gone forever in the TIFF.

Comparing how many bits a raw file has to how many bits a TIFF image has is like comparing apples to oranges. The bits in each do not represent the same thing.