Astronomy · Observatories, surveys and archives

Targeted studies and sky surveys

Connect a question to target selection, survey design and follow-up, using a fixed observing budget and an explicit repeated-exposure noise model.

Continue in the approved chapter order. Necessary vocabulary, quantities and assumptions are explained in the lesson.

Back to the chapter

After reading, you can

Distinguish a targeted study from a survey without assuming either is automatically representative.
Account separately for unique fields, visits, exposure and overhead.
Explain what repeated independent measurements improve and what their combination leaves unresolved.

Target selection is part of the evidence

A targeted study observes chosen objects or fields to answer a stated question. A survey uses a documented observing and selection programme over a declared region or sample. A survey need not cover the whole sky, and a targeted programme can contain many objects.

The distinction concerns the design and resulting claims, not a simple large-versus-small count. A deep survey of a small field can seek faint populations. A large target list can still be selected for unusual spectra, convenient visibility or prior detections.

Ask how a source became eligible, how observations were allocated, and which usable measurements entered the final analysis. These steps can differ. A source in the footprint may be too faint, crowded, masked or otherwise excluded from a particular catalogue or follow-up sample.

A conclusion about the measured targets becomes a conclusion about a larger population only with an appropriate selection and inference model. A survey label does not make a sample random, complete or free of bias.

Coverage and repeat observations compete within a declared budget

Consider an original synthetic programme with six charged hours, or 21,600 seconds. Every visit consists of a 300-second exposure plus 60 seconds of overhead, including the first and last visit. The budget therefore permits 21,600 / 360 = 60 visits.

Assume each field covers exactly one square degree and all fields are disjoint. Twenty fields with three visits each cover 20 unique square degrees. Five fields with twelve visits each cover five. One field with sixty visits covers one. Each programme uses the same 60 visits and 18,000 seconds of on-sky exposure.

The numbers deliberately omit weather, visibility, calibrations beyond the declared overhead, changing conditions and calendar gaps. They are an accounting model, not a schedule for an actual observatory. In a real survey, overlapping footprints must also be counted as a union rather than added as if disjoint.

Each allocation serves a different question. More exposure on fewer fields can improve some faint-source measurements, while wider sampling can reach more sky. The times of the visits determine which changes can be followed. Total exposure alone does not specify that timing.

Original six-hour model: 300 seconds of exposure and 60 seconds of charged overhead per visit; disjoint one-square-degree fields.
Fields	Visits per field	Unique area	Exposure per field
20	3	20 deg²	900 s
5	12	5 deg²	3,600 s
1	60	1 deg²	18,000 s

Combining repeats needs a signal and noise model

A coadd combines measurements from multiple exposures under a specified registration, calibration and weighting procedure. For a constant source under identical conditions, expected signal counts add. Independent random variances add too. Correlated errors, response changes and source variability require extra treatment.

Use the same synthetic 300-second visits with an expected source rate of 2 electron counts per second and sky rate of 8 over the whole extraction aperture. The aperture contains four pixels, each with read-noise standard deviation 3 electron counts per readout. Dark counts are zero in this model, the response is linear and unsaturated, and background estimation adds negligible uncertainty.

For N visits, expected source counts S = 600N, sky counts B = 2,400N, and read variance = 4N × 3² = 36N. The total model variance is S + B + 36N = 3,036N. Signal-to-noise ratio is S divided by the square root of that variance.

Three visits give S = 1,800 and variance 9,108, so the model ratio is about 18.86. Twelve visits give S = 7,200 and variance 36,432, so it is about 37.72. Four times as many identical independent visits doubles this ratio. It does not guarantee a particular detection decision or improve every aspect of the angular response.

Discovery and follow-up answer connected but different questions

A survey may identify candidates using images, spectra or changes between epochs. Follow-up then selects some candidates for additional measurements. That selection can depend on brightness, classification confidence, location, observing opportunities or the question being investigated.

The followed-up objects therefore need their own selection record. A high fraction of one class among them does not automatically describe the full survey, still less all objects in the Universe. Objects without follow-up are not necessarily false detections or members of a different class.

The Hyper Suprime-Cam Subaru Strategic Program illustrates a survey with Wide, Deep and UltraDeep design layers and deliberate overlap with other wavelength projects. Its layered structure is useful here without treating older design targets as current delivered performance.

When observations from different facilities are combined, retain their bands, epochs, responses, calibration and selection. A linked set of measurements can be more informative than any one alone, provided the dependencies and limits remain visible. The next lesson separates sky coverage, depth and repeat timing in more detail.

Worked examples

Account for the same observing budget two ways

Compare twenty fields observed three times with five fields observed twelve times in the synthetic programme.

Both allocations use 60 visits at 360 charged seconds each, giving 21,600 seconds.
The first supplies 900 seconds per field across 20 disjoint deg²; the second supplies 3,600 seconds per field across 5 disjoint deg².
Neither allocation specifies the visit dates, and neither is universally preferable without a question.

The second allocation places four times as much exposure on each field while covering one quarter of the unique area in this model.

Check the noise gain rather than the exposure count alone

Why does increasing the synthetic coadd from three to twelve identical visits double its model signal-to-noise ratio?

Source signal rises from 1,800 to 7,200 expected electron counts, a factor of four.
Independent total variance rises from 9,108 to 36,432, also a factor of four, so its standard deviation doubles.
Signal divided by standard deviation therefore increases by four divided by two.

The gain is a factor of two under the stated model. Systematic or correlated errors and variable signals can change what a real coadd achieves.

Try the reasoning

Must a survey observe the whole sky?
Show the answer and reasoning
No.
A survey declares its region or sample and observing and selection design; it can cover a small deep field.
Can a class fraction in selected follow-up targets automatically describe all survey detections?
Show the answer and reasoning
No.
Follow-up selection can change the composition of the measured sample.
In the synthetic noise model, why does the read variance include the number of visits?
Show the answer and reasoning
Each visit adds another readout of every aperture pixel.
With four pixels and independent per-read standard deviation 3, N visits add 4N × 9 to the variance.
Do sixty visits and their total exposure specify whether a brief transient was sampled?
Show the answer and reasoning
No.
The exposure windows and gaps, as well as signal and noise, determine what the programme samples.

What to carry forward

Documented selection connects an observing programme to the population it can describe.
Unique area, visits, on-sky exposure and charged time are different quantities.
Repeated measurements improve precision only through an appropriate signal, variance and dependence model.

References

Teaching resources and research supporting this lesson’s explanations and examples.