Interesting, as this hasn't been my experience.

I *always* get more detail from fewer longer exposures than many short (total time being equal)

Must be something else at play here.

What camera are you using? This depends a lot on how much read noise the camera has. Some cameras just require longer exposures, while others allow and even require shorter exposures.

Noise is an interesting thing, in that it is at it's core a description of random and unknown variations around an expected value. The interesting thing about random and unknown variations is when you combine one random variation with another, the larger variation will dominate the overall variance from the expected measure. When we say we have noise of say 10e-, we are talking about a mean deviation from this expectation, but actual variation could be larger or smaller. This deviation could also occur as a positive or a negative relative to the expectation. If we have another noise, of say 2e-, this noise will also result in random, unknown deviations, however these will combine with the larger noise term. We cannot simply add 2e- to 10e- and say we now have 12e- read noise, though. The deviations are random, unknown, and can be positive or negative. We might have a +10e- deviation combined with a -2e- deviation, giving us a total deviation of 8e-. Or we might have a +5e- deviation combined with a +2e- deviation for a total of 7e-, o even +10 and +2 for 12e-. Or we might have a -2e- deviation combined with a +2e- deviation giving us a total deviation of ZERO! Noise terms add in quadrature, which accounts for this odd way in which noise terms combine.

If we combine a 10e- noise with a 2e- noise, we must first square these, then add them, then take the square root:

Ntotal = SQRT(10^2 + 2^2) = SQRT(100 + 4) = SQRT(104) = 10.2e-

By combining our 2e- noise term with our 10e- noise term, the "impact" of the 2e- noise has been greatly diminished!! In effect, it represents an actual impact to the total noise in our image of a measly 0.2e-!!

Because of this interesting aspect of noise, we can conclude that there comes a point of diminishing returns on exposure. If we continue to expose well beyond the point where one particular noise term...that being the total shot noise from light entering the scope in our background signal...then other noise terms effectively lose their "impact", they are effectively meaningless. We call this "swamping", and we usually refer to read noise as the term to be swamped. We target read noise, because we get one "unit" of read noise in every single sub exposure, it is **count-dependent**. ALL other noise terms...object shot noise, skyfog shot noise, even dark current shot noise...are purely **time-dependent**. It matters not how many subs you have, for a given total integration...say 5 hours...these other noise terms will sum up the same no matter whether you acquire a single 5 hour exposure, or 18000 1 second exposures. In fact, you can combine all the shot noise terms, and if they are together sufficient to swamp read noise, then read noise is swamped. Dark current is usually trivial overall, so usually we just use the background sky signal to determine whether we have swamped read noise. And, this swamping needs to be done on a sub-exposure basis...you need to expose enough to swamp read noise EVERY SUB.

So, how long you MUST expose for depends very much on how much read noise you have. A commonly used "swamp factor" is 10xRN^2. This means, you want the background sky signal to be 10x as large as the read noise squared. So, if you have a camera that has, say, 10e- read noise, then you want your background sky signal to be 10x stronger than 10^2, which is 10x100, which is 1000e-! That is quite a lot of signal, and depending on your image scale and aperture, it may indeed take a very long time before you expose enough signal to achieve this goal. On the other hand, if you read noise is just 2e-, then you need a background signal of 10x2^2, which is 10x4, which is just 40e-. Purely from a background signal standpoint, assuming all else being equal, you would need exposures just 4% as long as those necessary for a 10e- read noise camera. It may be that you only need 5 minute exposures with 2e- read noise, but 125 minute exposures with 10e- read noise. Things get more complex once you start considering differences in pixel size...and even more complex when you start considering differences in aperture and focal length. Even with low read noise, imaging with a smallish aperture at f/10 might still require very long exposures... Conversely, imaging with a very big aperture at f/2 with high read noise might require very short exposures.

In the end, what matters is that you make the backround sky shot noise a much larger noise term than the read noise. If you do that, then you will have exposed sufficiently enough to render read noise largely moot. Interesting thing about stacking...the stack will have the same relative difference as a single sub. Previously, I demonstrated that swamping 2e- read noise with 10e- shot noise reduced the impact of the read noise to a mere 0.2e-. This ratio would hold in a stack. If we stacked 100 subs, we would have: SQRT(100 * (10^2 + 2^2)) = 102e-. This noise is exactly 10x larger than the single sub, and the relative impact of read noise is the same...2e- vs. 100e-, 0.2e- vs. 10e-, it's 0.02 either way.