ASTR 210 · Lecture 11
Reading anchor: OpenStax Astronomy 2e, Chapter 6.3 (detectors, CCD imaging).
Deep imaging and faint-source detection depend on combining many exposures, but stacking only improves a measurement if the images are registered to a common coordinate system and the noise in each frame is independent. Careless stacking can inject artifacts rather than removing them.
A single 2-hour-old exposure gives SNR = 12 for a faint source — barely a secure detection. Combining 9 such exposures, taken under similar conditions, should improve the detection significantly if the images are well registered.
First question: what here is directly measured, and what is inferred from a model?
Definition: The transformation and resampling that align multiple images to a common pixel or sky-coordinate grid.
Why it matters here: Registration ensures that real source signal lands at the same output location before frames are combined.
Relationships: Centroids or WCS solutions constrain shifts, rotation, scale, and sometimes distortion.
Boundary / common confusion: Registration after stacking cannot undo source smearing already introduced by misalignment.
Definition: A stack in which each output pixel is the median of the aligned input-pixel values.
Why it matters here: It rejects isolated cosmic-ray values without requiring their amplitudes to be modeled.
Relationships: For an odd number of frames, one extreme outlier does not change the central ordered value.
Boundary / common confusion: The median is robust but statistically less efficient than the mean for purely Gaussian noise and small stacks.
Definition: An iterative combination method that estimates a center and scatter, rejects values beyond a chosen sigma threshold, and recomputes the result.
Why it matters here: It retains much of the mean’s efficiency while rejecting cosmic rays and other outliers.
Relationships: The threshold is defined relative to a noise estimate and therefore depends on stack depth and the validity of that estimate.
Boundary / common confusion: Aggressive clipping can remove real variable or moving-source signal, especially with few frames.
Definition: An energetic-particle event that deposits charge in one exposure, often as a sharp bright pixel or track.
Why it matters here: Its single-frame nature distinguishes it from a stationary astronomical source and motivates robust stacking.
Relationships: Registered stacks expose cosmic rays as temporal outliers at a pixel while persistent source flux repeats.
Boundary / common confusion: Not every isolated bright pixel is a cosmic ray; hot pixels, satellites, and transients require different checks.
Definition: Random fluctuations whose values in one exposure do not predict their values in another.
Why it matters here: The square-root stacking gain assumes this independence across frames.
Relationships: Independent variances add, so averaging \(N\) comparable frames reduces standard error by \(1/\sqrt N\).
Boundary / common confusion: Flat-field errors, sky gradients, and guiding patterns repeated across frames are correlated systematics and do not average down this way.
Definition: The number of usable aligned exposures, or equivalently the total integrated exposure represented by a stack.
Why it matters here: It sets the expected random-noise improvement and the robustness available for outlier rejection.
Relationships: For equal independent frames, SNR grows approximately as \(\sqrt N\); total exposure grows as \(Nt\).
Boundary / common confusion: Nominal frame count overstates effective depth if frames differ greatly in quality or share correlated errors.
Definition: A documented rule for identifying and excluding measurements inconsistent with the expected distribution or temporal behavior.
Why it matters here: It prevents cosmic rays and defective pixels from dominating combined images.
Relationships: Median combination and sigma-clipping are two rejection strategies with different efficiency and sample-size requirements.
Boundary / common confusion: Rejecting points solely because they disagree with the desired result is not valid outlier treatment.
Read the stacking-gain curve alongside the cosmic-ray comparison.
Assuming that stacking automatically removes artifacts. Stacking only removes random, independent noise; it does not remove a systematic error (like an uncorrected flat field) present in every frame, and a naive mean combine does not remove cosmic rays.
Given nine simulated per-frame values for one pixel, one of which is a clear outlier, compute both the mean and median and discuss which combination method a careful reduction pipeline should use.
This unit is assessed through Problem Set 06 rather than a dedicated lab, using simulated per-frame values that mirror this worked example.
Stacking is a statistical operation that trades exposure count for signal-to-noise; it rewards careful registration and is easily defeated by sloppy combination methods.
references/openstax-astronomy-2e.pdf.materials/ASTR210/data/.