How Long Should Your Sub-Exposures Be?

Ask this in a forum and you will get three answers: 300 seconds because that is what the poster uses, “as long as your mount allows”, and a link to a calculator that wants eleven numbers you do not have.

There is a real answer, it is specific to your sky rather than your gear, and you can measure it on one ordinary night.

What a longer sub actually buys you

Only one thing: it dilutes read noise.

Read noise is added once per frame, when the sensor is emptied. A 600-second exposure pays it once. Twenty 30-second exposures pay it twenty times. That is the entire case for long subs, and it matters most on faint targets where read noise is a meaningful share of the total.

Everything else — photon noise from the target, photon noise from the sky — accumulates with total integration time, not with how you slice it. Ten hours is ten hours whether you took it in 60-second or 300-second pieces. (How stacking actually works covers why that is true.)

So the question is narrower than it sounds. It is: how long until read noise stops mattering?

The swamping criterion

Read noise stops mattering once the noise from the sky background is comfortably larger than it.

Sky background is not darkness. Under any real sky there is airglow, scattered moonlight, and — for most of us — a town. Those photons land on your sensor and bring their own shot noise, and once that noise is several times the read noise, adding more exposure to a single frame is buying you almost nothing.

The usual rule of thumb: aim for sky noise at three to five times read noise. Beyond five, you are at the point of diminishing returns and paying for it with everything in the next section.

Here is the part people miss. A bright sky reaches that threshold faster. Light pollution, which is bad for almost everything else, means your optimal sub length is shorter. Someone in a Bortle 8 suburb genuinely needs shorter subs than someone in a Bortle 3 field, and copying their settings from a dark-sky imager is copying the wrong number.

Measuring it, without maths

You do not need to know your camera’s read noise in electrons. Use the histogram.

Take a single exposure at your working gain and look at where the background peak sits, as a fraction of full scale. For most modern CMOS astro cameras, a background peak sitting somewhere around 10% to 20% from the left is comfortably swamped. Much further left and read noise is still in play. Much further right and you are wasting dynamic range on sky you will subtract anyway.

Then do this: take a 60-second frame and a 300-second frame of the same field. Subtract the sky level from each. If the 300-second frame’s background peak is five times higher, you are sky-limited already and the longer exposure gained you nothing but risk. If it is only twice as high, you were read-noise-limited at 60 seconds and the longer sub was doing real work.

Ten minutes of testing, once per site and gain setting, and you never have to guess again.

Why longer is not free

Four costs, and they are not small.

Saturated stars. Every extra second pushes more of the bright stars to full well. Once a star’s core clips, it is a white disc with no colour and no profile, and it stays that way through every subsequent step. Long-sub images of star-rich fields often have that faintly cartoonish look for exactly this reason.

Larger loss per accident. A satellite, a gust, a passing cloud costs you one frame. When each frame is five minutes, that is five minutes. At 60 seconds it is one.

Fewer frames for rejection. Statistical outlier rejection needs a population to work with. Below about fifteen frames, sigma clipping cannot reliably tell a satellite trail from a legitimately bright pixel — there is more on this in how many frames you need. A three-hour session in 300-second subs is 36 frames. In 90-second subs it is 120, and the rejection is meaningfully better.

Tighter tracking demand. Everything in why your stars are trailing gets harder as subs get longer. A mount that holds round stars for 90 seconds unguided may fail at 240.

Rough starting points

Measure rather than trust these. But if you need somewhere to begin, for a modern cooled CMOS at unity-ish gain:

Sky	Broadband (LRGB / OSC)	Narrowband
Bortle 8–9, city	30–60 s	180–300 s
Bortle 6–7, suburb	60–120 s	300–600 s
Bortle 4–5, rural	120–300 s	600–900 s
Bortle 1–3, dark	300–600 s	900–1200 s

Narrowband is longer across the board because a 3nm filter throws away most of the sky background along with everything else — it takes far more time to swamp read noise when you have blocked 99% of the light reaching the sensor.

A DSLR or mirrorless camera without cooling belongs one row shorter than its Bortle class suggests, because thermal noise accumulates within the frame and there is no set point holding it steady.

Gain, briefly

Higher gain lowers read noise in electrons, which shortens the sub needed to swamp it, at the cost of dynamic range. Most modern sensors have a step in the gain curve — the point where a high-conversion-gain mode kicks in and read noise drops sharply. Sitting just above that step is a defensible default and it is what most manufacturer “unity gain” recommendations are pointing at.

Do not chase gain to fix an exposure problem. Set it once, sensibly, and vary time.

What the stack does with it

Uneven sub lengths within one session are best avoided, because calibration frames have to match and rejection statistics assume a consistent population. If you must mix, keep them in separate groups and combine the results at the end rather than throwing everything into one stack.

Akastroid reads the exposure time from the FITS header and will say so when a folder contains frames of different lengths, rather than silently averaging a 60-second frame with a 300-second one and producing a result nobody can interpret. It is a small thing that saves an evening of confusion.

The short version

• Long subs buy exactly one thing: less read noise. Nothing else.
• Bright skies swamp read noise faster, so light pollution means shorter optimal subs.
• Check the background peak sits 10–20% from the left of the histogram.
• Longer costs you saturated stars, bigger losses per accident, and worse rejection.
• Take the shortest sub that swamps read noise, then put the night into frame count.