One problem we photographers continue to have is that terms get invented and then used in discussions and instruction without a clear definition. The elephant in the room for this is the term “dynamic range.”
Go ahead. Define dynamic range. I’ll give you a few moments to do so (scroll down when done).
I’ll bet that most of you came up with something akin to “noise floor up to sensor saturation.” Only two problems with that definition: (1) a number of sensors, including those in the Nikon D850 and Z7 models, are not linear as they approach saturation; and (2) you used another undefined term to define an undefined term: what’s a noise floor?
There is one definition for a type dynamic range (DR) that’s generally agreed upon: something called Engineering DR. That’s the number of stops between sensor saturation down to a signal to noise ratio of 1:1. Ever seen a signal to noise ratio of 1:1? There’s really no remaining and useful photographic information there that you can perceive. You’re actually in the basement, not at the floor. Heck, you may be in the ground below the basement it's so random at 1:1.
Why does Engineering DR use 1:1? For a number of reasons, but one is an attempt to separate quantum shot noise from electronic read noise (or light-produced noise versus sensor-produced noise). A sensor that is not getting any light at all, and thus should be producing a true black output value of 0, doesn’t produce such an exact value because of electronics-related noise.
Different types of noise produce different impacts on our data. Quantum shot noise is the randomness of photons. Time and area impact how much of that noise you’ll see. Electronics noise varies, with heat being just one of the critical additional factors that come into play.
While not perfect, one thing that has happened in image sensors is that electronics noise has been (mostly) “tamed.” Not eradicated, just minimized for the most part. Generate too much heat or leave an image sensor collecting data for too long an exposure, and you’ll start to see the electronic noise push past that minimization.
As you might be figuring out, there’s a lot going on here, and some of it might depend upon what we’re doing.
Then there’s another variable that comes into play: are you talking about the dynamic range of light produced in a scene (e.g. darkest to brightest point), or are you talking about the dynamic range your camera/film can capture?
I’ve lost the reference, but in the film era Kodak once went around measuring the range of light in a scene and came up with about 10.5 stops as being a typical outdoor high contrast scene maximum (I don’t recall that they included direct sun in their measurements). That’s a contrast ratio of 1500:1 (quick sneaky question: what’s the contrast ratio of the monitor you’re editing your image on, and did you know that photographic paper might have a contrast range of only 128:1?).
These days I see reports of dynamic range for our cameras that are all over the map. The camera maker says 15 stops, DxOMark says a “Screen” DR of 13.2 stops and “Print” of 14.5 stops, photonstophotos.com says 11.3 stops. Who’s right? Guess what, they can all be right, assuming that they use diligence in creating repeatable results. They just have different definitions of dynamic range.
See how “dynamic range” is a nebulous concept that none of us agree on? Yet we're arguing about it without defining the term!
Long-term readers will know that I’ve used the term “usable dynamic range” for almost two decades in my books. Yes, I have an entirely different method of measuring dynamic range than the ones I’ve already mentioned, a method I’ve used since I first started doing such testing for my now-defunct newsletter when the D1h/D1x came out. As sensors began producing less electronics noise, I lowered my “noise floor” definition somewhat, as random photons aren’t as visually objectionable as artificially produced noise layered on top of that.
Some of you may have noted when I started referring to noise in some cameras as “much like grain” as opposed to “clear electronic artifacts.” Nikon, in particular, concentrated on eliminating color noise while leaving some luminance noise in their JPEG renderings very early on. That’s because luminance noise looks a lot like film grain, which we're conditioned to, but color noise looks, well, artificial and electronic. Heck, have you even noticed that the Adobe raw converters automatically apply some color noise reduction by default? Same reason.
It’s not up to me to define dynamic range. The industry as a whole needs to address this, so that we’re all referencing the same data the same way. That’s not likely to happen, as everyone has a reason to do what they’re doing.
My advice to you—assuming you think “dynamic range” is something you need to know—is to pick one source that tests consistently and use only their numbers to try to compare how two products are different. Also: learn how that organization tests and reports their numbers, so that you can better understand what they actually mean.
Another way of looking at dynamic range is to consider thresholds. At the lower end of the exposure is a threshold below which you don't get usable information (noise exceeds signal enough that the signal's usefulness is at or near zero). At the high end of the exposure is a threshold beyond which you either get no additional data (saturation) or get unreliable data (non-linear). "Dynamic range" (for you) is the usable data between the two thresholds you define.
You may have noted that “assuming you think dynamic range is something you need to know” clause earlier. Personally, I stopped worrying about small differences in dynamic range years ago. The current state of image sensors is such that they fairly truly record the randomness of photons within almost the entire range of data collection I tend to make. If my shutter speed is in the 1/2 to 1/8000 second range and my ISO isn’t at an insane level, nitpicking the image sensor stopped being useful to me somewhere around 2011/2012. Yes, the D850 is a little better than the D800, but not really in the range I just mentioned. (Full disclosure: my “insane” ISO level changed between those cameras, mostly due to the point at which Nikon starts using scaling in the DN’s, which was 2000 on the D800, and is 8000 on the D850. The D800 tends to go purple at high ISO values.)
Where I do have to pay close attention to dynamic range differences still is in exposures longer than 1/2 second. The electronics noise is different among different cameras, and all kinds of noise nastiness can raise its head in long exposures, from hot pixels to stuck pixels to PRNU noise to…well, the list is actually pretty extensive. But I wouldn’t use the term “dynamic range” to differentiate between what two cameras are doing at 5 minute exposures.
