Is exposure only a global adjustment for auto exposure modes? Is there anyway to run auto-exposure independently for R, G and B? I'm getting a headache thinking about how that would not work -- colors would shift, the metering may not work that way, etc. -- but please sate my curiosity and tell me anyway. Exposure is exposure (shutter speed vs. ISO speed vs. aperture), and affects all channels, and there is no way to adjust one over the other. They are not affected equally (but don't worry about trying to get into all of that as it's not helpful).
Actually, exposure is JUST related to shutter speed and aperture. ISO is not really an exposure factor...it simply amplifies the signal that the exposure creates.
The key things to understand are that RAW files are recorded in a logarithmic manner and the highlights contain a lot of detail (in terms of data) while the shadows do not. If you clip the highlights completely, the data is gone, but if you are near the clipping (i.e. Expose to the Right), it's easy to recover. If you clip the shadows, it's totally gone and if your photo is underexposed, the detail that you can recover is mushy and noisy.
This is also fundamentally wrong. RAW data is recorded
linear. Trust me on this
...I do astrophotography, with DSLRs, and the linearity of the data is absolutely critical to being able to process the data correctly. In PixInsight, the processing procedurs are also usually split, between processing in linear mode, and processing in non-linear (post-stretch) mode.
The non-linearity of what we see in a tool like Lightroom has everything to do with the tool, NOT the data. The data is linear, it is rendered to the screen via non-linear tone curves, and non-linear processing.
So the idea is to get the data towards the right without clipping any of the channels completely (though a few small "blinkies" are okay), especially with flowers or other saturated color objects. To do this, you MUST have an accurate histogram and White Balance has a huge effect on the accuracy. The closer the WB is to the conditions, the more accurate the histogram.
You can expose to the right right up to the clipping point. Since the data IS linear in a RAW file, if you expose your red channel to 2^14 - 1, then your right at, but not over, the clipping point. You will not have lost anything.
The in-camera histograms are usually based off of JPEG thumbnails, which are actually highly inaccurate. This is why some people use UniWB, to change the per-channel weighting, and force the JPEGs that the histograms are based on to more accurately reflect the real dynamic range and clipping point of each channel.
Without UniWB, you can usually expose a little more than the in-camera histogram and "blinkies" would lead you to believe.
The resulting photo will look overexposed, but when you take it into a RAW processor and drop the exposure a stop or so, the photo will look perfect and you have lots of detail. Plus, the shadows (i.e. saturated colors here) will also have more detail and latitude in terms of exposure and color adjustments.
There is one caveat...there is a very slight non-linearity to the response of the silicon in the sensor itself. That usually results in the uppermost levels near the clipping point tapering off in a small shoulder. It's best not to push exposure right up to the limit...i.e. 2^14-1. You want to keep your maximum levels just a little lower than that...2^14-10 or so is best. Otherwise, you'll notice that the highlights in those regions end up normalizing, becoming gray. You also start noticing very slight color shifts when you recover highlights that are right near the clipping point, as the processing algorithms are non-linear, and they will affect those upper upper highlights more than any other part of the signal.
Dropping the exposure by a stop or two is extreme. You want to ETTR, to maximize your use of the sensor's DR, but you don't want to push it too far. I'd say once you figure out where your real clipping point is, pull back by a third of a stop.