Astrophotography is fundamentally a problem of signal versus noise.
Every imaging decision — exposure length, focal length, binning, camera choice — ultimately affects Signal-to-Noise Ratio (SNR).
Signal is the number of photons from the astronomical target that are recorded by the sensor.
Examples of signal include:
Nebula emission
Galaxy light
Starlight
Faint dust and reflection
Signal is measured in electrons generated within each pixel.
If one exposure records S electrons of signal:
Two identical exposures record 2S
Ten exposures record 10S
One hundred exposures record 100S
When stacking multiple sub-exposures, the total collected signal increases linearly:
Total Signal=S×N\text{Total Signal} = S \times NTotal Signal=S×N
Where:
SSS = signal per exposure
NNN = number of exposures
Signal increases linearly with total integration time.
Noise is random variation that obscures the signal.
Unlike signal, noise does not accumulate linearly.
Most noise sources in astrophotography follow Poisson statistics, where:
Noise = sqrt(counts)
This square-root behaviour is the key to understanding stacking and integration.
Photon arrival is random. Even a perfect sensor cannot avoid this.
σshot=S\sigma_\text{shot} = \sqrt{S}σshot=S
Shot noise increases as signal increases.
Sky glow from light pollution, airglow, or moonlight contributes background electrons.
For a signal of S electrons, the associated shot noise is equal to the square root of the signal:
sigma_shot = sqrt(S)
In most modern imaging, sky noise dominates once exposures are long enough.
Thermal electrons generated within the sensor follow Poisson statistics.
The associated dark current noise is:
sigma_dark = sqrt(D)
Cooling reduces the amount of dark current but does not change its statistical behaviour.
In DSLRs and uncooled one-shot-colour cameras, dark current increases linearly with sub-exposure length, while its random (shot) noise increases with the square root of the dark current. In warm conditions, long sub-exposures therefore raise the dark-noise contribution to the total noise budget.
While purely random dark noise still averages down with increased total integration, high dark current in uncooled cameras is often accompanied by fixed-pattern artefacts (hot pixels, banding, amp glow) that do not average down efficiently. In this regime, extending individual sub-exposures tends to degrade image quality more than it improves SNR.
For these cameras, it is usually better to aim for sky noise at roughly 2–3× the read noise and accept a modest read-noise integration penalty, rather than allow dark-current-related artefacts to dominate.
Read noise is introduced each time the sensor is read out.
Unlike shot noise, it does not depend on signal level or exposure time.
Read noise = RN
Read noise is added once per sub-exposure and cannot be reduced by longer individual exposures, only by stacking fewer, longer subs.
Read noise is:
Independent of exposure length
Fixed per sub-exposure
The primary limitation in very short exposures
The total noise in a single pixel for one sub-exposure is the quadrature sum of all noise sources:
sigma_total = sqrt( S + B + D + RN^2 )
Where:
S is the signal electrons
B is the sky background electrons
D is the dark current electrons
RN is the read noise
Shot noise from signal, sky, and dark current all follow Poisson statistics, while read noise is added separately by the electronics.
When stacking N identical exposures:
Signal increases linearly:
S_total = N × S
Noise increases as the square root:
When stacking N sub-exposures, noise adds in quadrature.
The total noise becomes:
sigma_total = sqrt(N) × sigma
Where:
sigma is the noise in a single sub-exposure
N is the number of stacked sub-exposures
Signal-to-Noise Ratio (SNR)
The resulting signal-to-noise ratio after stacking is:
SNR = (N × S) / (sqrt(N) × sigma)
SNR = sqrt(N) × (S / sigma)
This shows that SNR improves with the square root of the total integration time, not linearly.
Because of the square-root law:
4× more integration → 2× better SNR
9× more integration → 3× better SNR
16× more integration → 4× better SNR
This is why deep images require hours, not minutes.
There are no shortcuts.
At very short exposures:
Read noise dominates
Stacking is inefficient
At longer exposures:
Sky noise dominates
Read noise becomes negligible
Sky-Limited Exposure Rule
A common rule of thumb for sub-exposure length is to expose until the sky background noise dominates read noise:
sigma_sky ≳ 2–3 × RN
When this condition is met, read noise becomes a minor contributor and further increases in sub-exposure length provide diminishing returns.
Once this point is reached:
Longer sub-exposures provide little SNR benefit
Only more total integration time improves the image
This is why optimal sub length depends on sky brightness and f-ratio, not focal length alone.
Understanding signal and noise explains:
Why oversampling destroys faint detail
Why long focal lengths punish DSLRs
Why binning can improve real signal
Why CFALD increases usable luminance SNR
Why “just take longer subs” is often wrong
Why integration time always wins
Astrophotography is not guesswork.
It is statistics applied to photons.