A star-shaped patch in an astronomical image contains more than starlight. It also contains sky background and the effects of the detector. Aperture photometry estimates the source signal by adding the recorded values inside a chosen region and accounting for the background beneath that same region. The important question is how much of the sum can reasonably be assigned to the star.
This guide works through sky subtraction, aperture size and background uncertainty using small, checkable numbers. It then changes the sky brightness and the masking pattern to test whether the explanation still works. You need ordinary arithmetic first; the uncertainty calculation comes after the measurement has a clear meaning.
eduKateSG · The Gold Standard of Astronomy
Choose your reading route
Understand the measurement · Chapters 1–2. Separate the star signal from the recorded sum.
Choose and check the model · Chapters 3–5. Examine sky regions, aperture area and uncertainty.
Test and continue · Chapters 6–8. Use changed images, FAQs and connected guides.
Full chapter index · Practice and evidence checks · Return to Astronomy
Chapter index
Understand the measurement · Chapters 1–2
Choose and check the model · Chapters 3–5
Test and continue · Chapters 6–8
Imagine placing a circular boundary around a star in an image. Every included pixel contributes its recorded value. A brighter-looking patch may mean a brighter star, a brighter sky, a longer exposure or a different detector response. The sum alone does not distinguish those possibilities.
For the first example, use calibrated, linear image values expressed as counts. Assume one exposure, no saturated pixels, and a locally uniform background. Let S be the sum inside the source aperture, A the effective number of included pixels, and b the estimated background counts per pixel. The net aperture signal is F = S − Ab.
The units explain the operation. Counts per pixel multiplied by pixels gives background counts inside the aperture. Subtract that total from the recorded counts. Subtracting b just once would remove the background from only one pixel.
F describes the source signal retained by this aperture. It is not automatically the total light from the star, a calibrated physical flux or an intrinsic luminosity. Those are further questions requiring further information.
Suppose an aperture includes exactly 100 whole pixels. Their sum is 14,000 counts. A suitable nearby region estimates a background of 100 counts per pixel. The background underneath the source aperture is therefore 100 × 100 = 10,000 counts.
The retained source estimate is 14,000 − 10,000 = 4,000 counts. A mistaken calculation of 14,000 − 100 would produce 13,900 counts. The mistake is a mismatch between a per-pixel quantity and a whole-aperture quantity.
| Quantity | Value | Meaning |
|---|---|---|
| S | 14,000 counts | Source plus background in the aperture |
| A | 100 pixels | Included area for this integer-pixel example |
| b | 100 counts per pixel | Estimated local background level |
| Ab | 10,000 counts | Background allocated to the aperture |
| F | 4,000 counts | Net signal retained in the aperture |
Explain the result before moving on: the image did not contain only 4,000 recorded counts. It contained 14,000 counts in this region, of which the model assigns 10,000 to background. The source estimate depends on that assignment being reasonable.
A nearby ring, called an annulus, can estimate the background around a star. Photutils documents this approach and the need to scale the estimated level by the actual unmasked aperture overlap area. The geometry and the background estimator are parts of the measurement, not decorative settings.
Suppose a neighbouring source contaminates the ring and raises its estimated level from 100 to 110 counts per pixel. With S = 14,000 and A = 100, the reported net signal becomes 3,000 counts. The extra ten background counts assigned to each of a hundred pixels remove a thousand counts from the star estimate.
A useful inspection asks whether the ring includes other stars, extended emission, bad pixels or a background gradient. A robust statistic can resist some contamination, but it cannot make an unsuitable region represent the sky beneath the target. A ring lying on a different part of a nebula may answer the wrong background question.
Record the chosen regions and exclusions. Compare reasonable alternative background regions and inspect the image itself. A substantial change in the inferred star signal is evidence that the background choice needs attention, rather than permission to select whichever estimate looks most attractive.
A larger aperture captures more of the spread-out stellar image, but also more background. Consider two measurements of the same star in one exposure. Keep the estimated background at 100 counts per pixel.
| Aperture | Area | Recorded sum | Background | Net signal |
|---|---|---|---|---|
| Smaller | 100 pixels | 14,000 | 10,000 | 4,000 |
| Larger | 144 pixels | 18,800 | 14,400 | 4,400 |
The larger sum is 4,800 counts higher, but its net source estimate is only 400 counts higher. The added region contains substantial sky as well as additional source light. Compare the net values, their uncertainties and possible contamination before deciding which aperture serves the question.
A growth curve plots background-corrected signal against aperture size. It can help examine how the measurement approaches a stable retained signal. Crowding, gradients and noise can make this difficult. Aperture corrections or a model of the point-spread function require their own evidence; enlarging a circle is not a universal cure.
For a fractional-pixel boundary or a masked image, use the area consistent with the actual summation weights. If an implementation omits pixels, blindly using the full geometrical circle can subtract sky that was never included in the sum.
Assume A = 100 is exact, the source-aperture sum has standard uncertainty u(S) = 130 counts, and the estimated mean background has standard uncertainty u(b) = 2 counts per pixel. For this example only, take S and b as independent and let u(S) include the relevant noise already present in that aperture.
The background estimate contributes A × u(b) = 200 counts of standard uncertainty. First-order propagation gives u(F) = √[u(S)² + A²u(b)²] = √[130² + 200²] ≈ 239 counts. Relative to F = 4,000 counts, that is about 6.0%.
The scatter among individual background pixels is different from uncertainty in the estimated background level. The latter depends on how the estimate was constructed. Correlated pixels, gradients, rejected pixels and use of a median can change the calculation. The simple formula above does not establish that every real image has independent pixel errors.
Subtraction removes the estimated background level; it does not erase the random noise that arrived with those photons. Avoid counting an uncertainty contribution twice if a supplied error model already includes it. NIST’s propagation guidance provides the general framework, including covariance when independence does not hold.
Changed-context check: did the star brighten?
In a second exposure with the same duration and response, let the background rise to 130 counts per pixel. The same 100-pixel aperture now sums to 17,000 counts. Predict the source estimate before reading the answer.
The answer is 17,000 − 100 × 130 = 4,000 counts. The raw sum increased while the corrected aperture signal stayed the same. An uncorrected comparison would have mistaken a sky change for a source change. A real comparison must also examine transparency, calibration and uncertainty.
Changed-context check: did masking recover the whole star?
Now an aperture retains only 90 pixels and sums to 12,600 counts, with b = 100. Its net signal is 3,600 counts. Using an area of 100 would incorrectly give 2,600. Using the correct area fixes the sky subtraction, but it does not restore source light in the missing pixels.
Before accepting a result, preserve the image identity, exposure, calibration, aperture centre and shape, overlap area, mask, background method and uncertainty model. Inspect saturation and nearby sources. Check that repeated measurements use compatible definitions. These records make a discrepancy investigable.
Is a negative net signal impossible?
A noisy estimate can be negative when the background-subtracted signal is weak. That is an estimate relative to a background model, not a claim that the star emits negative light. Do not replace such values with zero without considering how that changes the analysis.
Does a larger telescope aperture mean the same thing as this aperture?
No. Here, the photometric aperture is a selected region in the recorded image. The telescope aperture describes its light-collecting opening. The two affect different parts of the measurement chain.
Can I compare counts from different exposures directly?
Only after checking duration, detector response, filters, calibration and observing conditions. Dividing by exposure time may be one step; it does not resolve every difference between images.
Does precise sky subtraction establish a planet transit?
No. It improves one measurement step. A transit interpretation still needs timing, comparison stars, instrumental checks and competing explanations. The existing transit guide follows that wider inference.
Try explaining the first example without the formula: add everything recorded inside the boundary, estimate how much sky lies beneath that same area, and retain the difference with its uncertainty. Then let someone change the area or the sky level and ask you to predict the effect.
That is the useful transfer. You are learning to keep an extensive quantity, a per-pixel level and a retained source estimate separate. The same discipline will help when reading a light curve, inspecting a telescope pipeline or deciding whether a numerical change deserves a physical explanation.
Subject and learning map
The Gold Standard of Astronomy — the connected subject guide.
Education: learning that transfers into life — explain the relationship again when the setting changes.
How X Works — continue through the wider subject library.
From photometry to a time-dependent claim
Vetting an exoplanet transit candidate — distinguish a measured dip from a planetary explanation.
Finite exposure averaging — check which variability an exposure averages away.
Daily aliases — separate sampling patterns from candidate periods.
Observation and the image
Astrophotography and a moving sky — connect image formation with observing choices.
Reliable astronomical observation — keep provenance and interpretation checkable.
Authoritative sources and scope
Sources checked on 21 October 2026. All numerical cases in this article are original, simplified teaching examples. They are not reported experimental results or official economic statistics.
Photutils: Aperture Photometry — developer documentation for background subtraction, overlap area and aperture measurements.
NIST TN 1297: Law of Propagation of Uncertainty — the first-order uncertainty framework and covariance terms.
Contents · Previous chapter · Continue with the subject guide