Calibration sources and comparison measurements

Choose the comparison for the effect it must constrain

Astronomical observations use many kinds of reference: a dark exposure, flat field, wavelength lamp, flux standard, comparison star or source-free sky region. They do not all test the same part of the measurement. A reference useful for detector bias cannot establish a spectral flux scale merely because both observations are called calibration.

Match the reference to the instrument mode, response range, observing conditions, region and scientific quantity. A standard measured in an incompatible configuration or outside the relevant calibration range may leave the needed correction unconstrained. A reference source also has measurement uncertainty; treating its tabulated value as exact can understate uncertainty in the target result.

A comparison used to establish the result and an independent check have different roles. In differential photometry a comparison star can set a relative brightness scale, while a separate stable check star tests whether that procedure behaves consistently. The AAVSO guide discusses choosing and monitoring those stars; agreement is a bounded check, not proof that every source or calibration effect is correct.

Instrument calibration plans distinguish supported configurations and additional reference observations. ESO’s plan illustrates why the configuration and temporal association matter. For a real observation, the appropriate instrument documentation determines the needed references; the calculation below is an original idealized design rather than a facility prescription.

Use an estimator that includes the reference observation

Let the source rate in a specified aperture be s = 20 electrons s⁻¹ and the background rate in the same effective aperture be b = 5 electrons s⁻¹. Stipulate a stationary background and matched response between the on-target and source-free off-target measurements.

Use integration durations ton and toff with ton + toff = 900 s. Assume independent Poisson counts Non with mean (s + b)ton and Noff with mean btoff. Ignore overhead, read noise, calibration uncertainty and source variability for this toy. The source-rate estimate is shat = Non/ton − Noff/toff.

Its expected value is s under the matched-background model. Its variance is Var(shat) = (s + b)/ton + b/toff = 25/ton + 5/toff, in squared rate units. The second term is the uncertainty cost of measuring the reference background; subtracting that estimate does not remove its random noise.

The known planning rates are used to choose the design, not claimed to be exact measured values from a future observation. If those rates are uncertain, the design comparison must account for that uncertainty or examine a range of plausible conditions.

Equal time is not automatically the best split

For a fixed total duration T, minimize V(t) = (s + b)/t + b/(T − t) over 0 < t < T, where t = ton. Differentiation gives −(s + b)/ton² + b/toff² = 0. Thus ton/toff = √((s + b)/b).

With the supplied rates this ratio is √5 ≈ 2.23607. The resulting durations are ton ≈ 621.8847 s and toff ≈ 278.1153 s. The rate variance is approximately 0.058178533, and its standard uncertainty is approximately 0.2412023 electrons s⁻¹.

An equal split, 450 s on and 450 s off, gives variance 25/450 + 5/450 = 0.066666667 and standard uncertainty approximately 0.2581989 electrons s⁻¹. The optimized split improves precision within this exact model because the on-target count has the larger rate and variance contribution.

The variance tends to infinity at either zero-time end for the positive rates used here, and its second derivative is positive inside the domain. The stationary split is therefore the minimum of this stipulated continuous allocation. Discrete exposure choices, readout costs, a different background estimator or time-variable sky create a different optimization problem.

Longer reference integration cannot repair the wrong background

Change one assumption: the true on-target background is 6 electrons s⁻¹, while the off-target background remains 5 electrons s⁻¹. Keep the source rate at 20. The expected estimated rate is now (20 + 6) − 5 = 21 electrons s⁻¹.

The one-electron-per-second offset is a background mismatch. Increasing both durations can reduce counting variation around 21, but it does not move that expectation back to the true source rate of 20. An uncertainty budget derived only from the matched-background model has omitted the effect that caused the bias.

A contaminated off-target region can cause the opposite problem by subtracting source light along with the background. Extended emission may fill an apparent sky region; a nearby star or spectral component may also enter it. Checking spatial extent and source-free coverage is a scientific requirement, not merely a way to obtain more reference pixels.

The NIRSpec background-strategy documentation illustrates the dependence on source extent, instrument configuration, subtraction method and available source-free regions. Its alternatives have different noise and matching conditions. None can be selected solely by transplanting this toy’s optimal time ratio.

Check standards sample what the calibration will be used for

A calibration fit can describe its input standards closely while failing at another brightness, wavelength, position or time. An independent check should exercise the regime where the result will be used. It can reveal discrepancies or drift that fitted residuals alone would conceal.

The NIST check-standard discussion emphasizes adequate sampling of the measurement process and retains uncertainty from the reference standards as a separate contribution. A few check observations do not exhaust every operating condition. Agreement where the process was tested should be reported with its range and limitations.

For an actual plan, distinguish which measurements establish the response, which estimate the background, which monitor stability, and which test the final procedure independently. Record the temporal relation, configuration, uncertainty and exclusions for each. If a check fails, investigate the model or conditions before treating extra exposure as a cure.

The purpose of comparison observations is to constrain the right influence while carrying their uncertainty into the result. The next lesson broadens the design choice from references and exposure time to which targets and observing windows can best answer a population or follow-up question.

Worked examples

Try the reasoning

1. Why does the source-rate variance include b/toff?

   Show the answer and reasoning

   The off-target background is measured with random counting uncertainty.

   A subtracted estimate is not an exact number.

2. Is a standard used to fit a calibration automatically an independent validation observation?

   Show the answer and reasoning

   No.

   The fitted data and an independent check have different roles.

3. Can an apparently empty region be used as background without checking whether the target’s extended emission enters it?

   Show the answer and reasoning

   No.

   Source contamination can make the subtraction remove part of the desired signal.

4. Does the √5 time ratio apply automatically when overhead, detector noise or time-variable background matters?

   Show the answer and reasoning

   No.

   Those changes alter the stated estimator, constraints or variance model.

What to carry forward

• References must constrain the relevant effect in the actual observing regime.
• A background estimate has its own uncertainty and can change the best exposure allocation.
• Independent checks can expose mismatch; more integration alone does not correct it.