GLOBAL KNOWLEDGE ARK

Astronomy · Questions about the sky

What we see and what we infer

Follow a signal from a detector to an astronomical conclusion. Use a simple brightness comparison to see where calibration, assumptions and uncertainty enter.

Read in chapter order if these ideas are new. The quantities and vocabulary used here are introduced in the lesson.

Back to the chapter

After reading, you can

  • Distinguish a detector signal, a calibrated measurement and an inferred physical property.
  • Calculate a simple relative brightness change and state the conditions needed to interpret it.
  • Explain why one successful fit does not exclude every alternative explanation.
01

The detector records a signal

A camera image shows a faint star. You might ask how bright the star is, but the first numbers belong to the detector: its pixels have recorded an electronic response during an exposure. Those values include contributions from the target, the sky and the instrument. Reading a number from an image is not yet measuring how much energy the star produces.

The first distinction is between a signal at the instrument and a property of the source. A pixel's value depends on collected light, exposure time, wavelength response and detector behaviour. A calibrated measurement accounts for those effects as far as the method allows. An inference then connects the measurement to a physical quantity through a model.

Inference is not a synonym for guesswork. It is reasoning from evidence using stated assumptions and a relationship that can be checked. A carefully tested indirect measurement can be more informative than a visually striking image. The important question is how the conclusion follows, and how it could fail.

02

Separate the target from its background

Suppose you add the values in an aperture, a chosen patch covering the star's image. The total includes light from the star and background. Estimate the background per pixel from an appropriate nearby region, then subtract the estimated background contribution over the aperture. Instrumental offset, dark current and pixel sensitivity also require suitable calibration in real work.

In a simplified bookkeeping model, S = C − n b. Here S is the estimated net source signal in detector counts, C is the total aperture count, n is the number of pixels in that aperture and b is the estimated background count per pixel. The model assumes the chosen background estimate applies to the aperture. It does not automatically correct detector nonlinearity, crowding or changes in how much of the star's image the aperture contains.

Counts are instrument units. They are not automatically photons, watts or a star's luminosity. In the synthetic example C = 1,400 counts, n = 20 pixels and b = 20 counts per pixel. The estimated background is 20 × 20 = 400 counts, leaving S = 1,000 counts. The word estimated matters: b has uncertainty, and subtracting it does not make that uncertainty disappear.

03

A ratio can reveal change hidden by observing conditions

Now measure a suitable comparison star in the same image using a compatible method. Let T be the target's net signal and R the reference star's net signal. Their ratio q = T/R has no unit because both signals use the same detector unit. If transparency or exposure multiplies both signals by the same factor, that shared factor cancels in the ratio.

That cancellation has conditions. The detector must be operating in its linear range, the reference must be sufficiently stable, the measurement must treat background and image shape adequately, and the shared change must really affect the sources comparably. Different colours, positions or atmospheric paths can introduce residual differences. Real differential photometry checks these effects and often uses several comparison stars.

The synthetic table gives two different second-night possibilities. In case A both signals are halved; the ratio stays 0.50. In case B the reference is halved but the target falls to 400 counts; its ratio becomes 0.40. Relative to the reference, the target signal is then 0.40/0.50 = 0.80 of its first-night value, a 20% decrease. This identifies relative change in the measured band under the stated assumptions. It does not establish why the target changed.

Synthetic detector signals for equal measurement methods. These numbers illustrate reasoning and are not observations of a real star.
RecordTarget T (counts)Reference R (counts)Ratio T/R
First night1,0002,0000.50
Second night, case A5001,0000.50
Second night, case B4001,0000.40

Shared dimming and a changed ratio

Scroll sideways to see the full diagram. Its explanation is below.

Shared dimming and a changed ratio Synthetic example, not observations. Upper panel: first-night target signal is 1,000 counts and reference signal is 2,000 counts. Second-night case A has 500 target counts and 1,000 reference counts. Alternative second-night case B has 400 target counts and 1,000 reference counts. Blue solid bars with circular markers are the target; orange patterned bars with diamond markers are the reference. The count axis starts at zero. Lower panel: the unitless target-to-reference ratios are respectively 0.50, 0.50 and 0.40 on a zero-based scale. Case A halves both signals and leaves the ratio unchanged. Case B gives a relative ratio of 0.40 divided by 0.50, or 0.80, a 20 percent decrease. The comparison assumes a stable reference and valid, comparable measurements; it does not identify a physical cause. Shared dimming and a changed ratio Synthetic signals • cases A and B are alternative second-night records Net signals (counts) Target T Reference R 0 500 1,000 1,500 2,000 1,000 2,000 500 1,000 400 1,000 First night Second night Case A: both × 0.5 Second night Case B: T × 0.4, R × 0.5 Ratio q = T / R (no unit) 0.0 0.2 0.4 0.6 0.50 0.50 0.40 First night Case A Case B Unchanged ratio Lower ratio Case B: 0.40 / 0.50 = 0.80 → a 20% decrease in relative signal. Assumes a stable reference and valid, comparable measurements. A changed ratio alone does not identify the physical cause.
The synthetic first-night net signals are T = 1,000 counts and R = 2,000 counts, giving T/R = 0.50. Alternative second-night case A has T = 500 and R = 1,000 counts, so the ratio stays 0.50; case B has T = 400 and R = 1,000 counts, so the ratio is 0.40. Both chart scales start at zero. In case B, 0.40/0.50 = 0.80, a 20% decrease in relative signal, conditional on a stable reference and valid comparable measurements; the ratio does not identify a physical cause.
04

Uncertainty travels with the conclusion

A measured ratio has uncertainty from the target signal, reference signal, background and calibration. More independent measurements can reduce some random fluctuations. Repeating the same unrecognised calibration error can instead produce very consistent wrong results. Precision describes the spread or repeatability; agreement with the intended physical quantity also requires attention to systematic effects.

Imagine two ratio estimates, 0.50 and 0.49. Without uncertainties you cannot decide whether their difference is convincing evidence of change. If the uncertainty in each is around 0.03, the difference is small compared with the measurement limits. If it is around 0.001 and the calibration is sound, it becomes a more interesting discrepancy. These uncertainty values are illustrative; no probability or significance level is being claimed.

A non-detection needs the same care. If an instrument cannot detect sources below a certain signal under the observing conditions, failing to see a source does not establish that its true signal is exactly zero. Report the sensitivity and the conditions. The absence of a recorded signal constrains what could have been present, rather than proving every possible source absent.

05

A light curve is evidence, not a label for the cause

A light curve is a record of measured brightness against time. A repeated dip could be caused by a planet crossing the star's face from our viewpoint, but eclipsing stars and other effects can also produce brightness changes. The shape, depth, duration, repetition, colour dependence and independent measurements help distinguish possibilities.

The forward question is: if this system and instrument model were correct, what would we expect to record? The inverse question is: given what we recorded, which systems and parameter values remain compatible? The inverse question can have several answers. A plausible model with a good fit therefore needs tests against alternatives and checks on the measurement itself.

Combining different evidence can narrow the possibilities. A transit constrains the relative sizes and viewing geometry under a stellar model. Measurements of spectral shifts can constrain line-of-sight motion, with stellar activity and instrument effects needing checks. Agreement between methods is valuable because their weaknesses and assumptions differ; it does not remove the need to inspect either method.

06

Say enough for someone else to follow

Compare “the star lost 20% of its energy” with “the target-to-reference signal ratio in this observing band fell by 20% in the synthetic example, assuming a stable reference and a linear, adequately calibrated measurement”. The second statement is longer because it identifies what the calculation actually established. It leaves luminosity, distance, extinction and the physical cause for further investigation.

The goal is not to attach every possible caveat to every sentence. Name the assumptions that could change the conclusion. Keep raw records, calibration information, processing choices and the path from measurement to inference available. Then another observer can test the same question, and a later you can understand why the result seemed convincing.

Worked example

From an aperture count to a defensible comparison

Using the synthetic values above, what can you conclude from a first-night net target signal of 1,000 counts and second-night case B?

  1. Check the first-night subtraction: 1,400 − 20 × 20 = 1,000 counts.
  2. Compare target with reference: first night q₁ = 1,000/2,000 = 0.50; second night q₂ = 400/1,000 = 0.40.
  3. Calculate the fractional decrease: (q₁ − q₂)/q₁ = (0.50 − 0.40)/0.50 = 0.20 = 20%.
  4. State the conditional result: the relative measured signal decreased, assuming stable comparison, comparable band and valid calibration.
  5. Do not infer a unique physical cause or a 20% change in total stellar energy output from this calculation alone.

The example establishes a 20% decrease in relative signal under its assumptions. It does not by itself establish a planet, a stellar luminosity change or a cause.

Try the reasoning

  1. The target and reference signals both fall to one-third of their original values. What happens to their ratio?

    Show the answer and reasoning

    It stays the same if the change is a shared multiplicative factor.

    (T/3)/(R/3) = T/R. This is the reason for the comparison, but additive backgrounds and unequal effects do not cancel in this way.

  2. With C = 1,400 counts and n = 20 pixels, what net signal results if b is mistakenly estimated as 25 rather than 20 counts per pixel?

    Show the answer and reasoning

    900 counts, which is 100 counts too low in this synthetic example.

    The subtraction becomes 1,400 − 20 × 25 = 900. A background error of 5 counts per pixel is multiplied across all 20 pixels.

  3. A planet-transit model fits a measured dip. Why is “therefore it must be a planet” too strong?

    Show the answer and reasoning

    Other systems and measurement problems may produce compatible dips.

    Check the data and competing predictions, and seek additional constraints. Fitting one model does not establish that every relevant alternative fails.

What to carry forward

  • A detector response, a calibrated measurement and a physical inference are different stages.
  • Ratios can remove shared multiplicative effects under explicit conditions.
  • Carry uncertainty and model alternatives along the full route from signal to conclusion.

References

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

  1. Stellar PhotometryAmerican Association of Variable Star ObserversTeaching resourceaavso.orgRead source
  2. The Nature of ScienceAndrew Fraknoi, David Morrison and Sidney C. Wolff · OpenStaxTextbook chapteropenstax.orgRead source
  3. The radial velocity method for finding exoplanetsESO / L. CalçadaTeaching resourceeso.orgRead source
  4. Guide to CCD/CMOS PhotometryAmerican Association of Variable Star ObserversCourse notes · PDFaavso.orgRead source
  5. Modeling Kepler Transit Light Curves as False PositivesNASA Technical Reports Server; primary BLENDER research recordResearch abstractntrs.nasa.govRead source
  6. Combat stellar activity in the hunt for young planetsEuropean Southern Observatory; Louise Dyregaard NielsenTeaching resourceeso.orgRead source