Explainers · 2026-07-11 · Patreon guide

Patreon for astrophotography creators: CMOS sensor QE and read noise, pixel scale and Nyquist sampling, sky-limited SNR and exposure strategy, Kappa-Sigma stacking algorithms, narrowband Hα OIII SII emission line imaging, PHD2 autoguiding and periodic error, atmospheric dispersion correction, and the Apple Tax in 2026

Astrophotography Patreons retain because the audience faces a calibration gap that YouTube cannot close: the time-lapse of a 10-hour integration does not contain the pixel scale calculation that determined the telescope-camera pairing, the sky-limited exposure formula that set the sub-frame duration, the Kappa-Sigma parameters that removed satellite trails from the stack, or the ADC Risley prism rotation angle that corrected atmospheric dispersion during a planetary window. The Patreon tier that keeps astrophotography patrons is the one with the worked sensor characterization, the exposure strategy spreadsheet, and the processing recipe — not the most dramatic final nebula image.

CMOS sensor quantum efficiency, read noise, and full well capacity

Quantum efficiency (QE) describes the probability that an incident photon generates a detectable electron in the sensor pixel — the fundamental efficiency of converting starlight to signal. Modern back-illuminated scientific CMOS (BSI sCMOS) sensors achieve peak QE of 85–92% near 550–700 nm; at Hα (656.3 nm), QE is typically 80–90%. QE drops below 40% in the ultraviolet below 400 nm due to silicon surface and electrode absorption in front-illuminated designs; BSI sensors eliminate the electrode layer above the photosensitive region, recovering short-wavelength QE significantly. A sensor with 80% QE at Hα converts 800 of every 1,000 arriving photons into photoelectrons; the remaining 20% are lost as heat. QE curves (percentage versus wavelength in nm) are the first calibration document an astrophotography Patreon should provide for each camera in the creator’s fleet — they directly determine which emission lines are accessible at usable efficiency.

Read noise is the electronic noise introduced during each analog-to-digital conversion readout event, measured in electrons RMS per pixel per frame. BSI sCMOS sensors achieve 1.0–1.5 e⁻ RMS at low-gain unity settings; conventional front-illuminated CMOS sensors measure 2–4 e⁻ RMS; cooled CCD sensors typically 2–4 e⁻ RMS. Read noise is the irreducible per-exposure noise floor: every sub-frame, regardless of duration, contributes this fixed noise amount. Its significance: if sky background noise × exposure duration ≫ read noise per exposure, adding more read noise per frame (by using shorter exposures and stacking more of them) costs SNR; the sky-limited condition eliminates this penalty.

Full well capacity (FWC) is the maximum number of electrons a pixel can hold before saturation — the charge capacity of the photodiode reservoir. FWC ranges from 25,000 e⁻ in cameras with small 3 µm pixel pitch to 100,000+ e⁻ in large 9 µm pixel cameras; saturation produces hard clipping in CMOS (pixel value locks at maximum ADU). Dynamic range = 20 × log₁₀(FWC / read_noise) in decibels: a camera with FWC = 50,000 e⁻ and read_noise = 2 e⁻ has DR = 20 × log₁₀(25,000) ≈ 88 dB — sufficient to record both the faint outer shell and the bright core of most nebulae without HDR blending. System gain in e⁻/ADU is the conversion factor between sensor electrons and image file ADU values: at unity gain (1 e⁻/ADU) each ADU represents one photoelectron; high gain settings (e.g., 0.12 e⁻/ADU) trade FWC range for detection sensitivity below one electron per ADU.

Pixel scale, Nyquist sampling, and telescope-camera pairing

Plate scale (PS) defines how many arcseconds of sky correspond to one pixel: PS = 206.265 × pixel_size_µm / focal_length_mm. The constant 206.265 converts radians to arcseconds (1 arcsecond = 4.848×10⁻⁶ radians; 1 / 4.848×10⁻⁶ ≈ 206,265 arcsec/radian; the extra factor of 1,000 accounts for millimeters versus micrometers in the units). Example: a camera with 3.76 µm pixels mounted on a 700 mm focal length apochromatic refractor: PS = 206.265 × 3.76 / 700 ≈ 1.11 arcsec/pixel. A 2× Barlow lens doubles the focal length to 1400 mm and halves the plate scale to 0.55 arcsec/pixel.

The Nyquist sampling theorem applied to telescope imaging requires the pixel scale be ≤ FWHM_seeing / 2.0 to properly sample the finest detail set by atmospheric turbulence. Typical suburban seeing FWHM 3.5–5 arcsec → ideal pixel scale 1.75–2.5 arcsec/pixel; rural dark-site seeing FWHM 2.0–3.0 arcsec → ideal pixel scale 1.0–1.5 arcsec/pixel. Under-sampling (pixel scale ≫ FWHM/2) aliases the PSF, producing blocky stars and losing resolution that the optics could theoretically deliver. Over-sampling (pixel scale << FWHM/2) dilutes stellar flux across many pixels unnecessarily: the star’s photons are spread over a 3×3 pixel area instead of a 2×2 area, reducing peak-pixel SNR by a factor of (3×3)/(2×2) = 2.25 without recovering any resolution beyond what seeing already set.

Drizzle algorithm, originally developed for the Hubble Space Telescope Wide Field Camera, recovers sub-Nyquist spatial resolution from multiple dithered sub-frames. Each sub-frame pixel is treated as a uniformly illuminated square ‘drop’ of flux; the drop is shrunk to a fraction of its original footprint (drop fraction 0.5–0.8 × pixel width) and then deposited with weighted fractional overlap onto a finer output grid (typically 2× finer pixel scale than the sub-frame pixels). Sub-frames must have random sub-pixel dither offsets relative to each other (from natural autoguide jitter of ±0.3–2 pixels, or intentional dithering by commanding the mount to shift position between subs): this dither provides the angular diversity that populates the fine output grid non-redundantly. Drizzle improves effective resolution by 15–30% compared to simple mean stacking when sufficient dithered frames are available (≥15–20 subs with ≥2 pixel dither).

Dark current, bias frames, and flat frames: the calibration stack

Dark current arises from thermally generated electron-hole pairs within the silicon depletion region: thermal energy promotes electrons from the valence band across the band gap without any photon input, producing signal indistinguishable from photoelectrons. Dark current rate is measured in e⁻/pixel/second at a specified sensor temperature. Dark current follows Arrhenius temperature dependence: dark current doubles approximately every 5.5–7°C increase in sensor temperature (the doubling temperature varies by manufacturing lot; empirical measurement at two temperatures determines the specific doubling temperature). Consequence: cooling the sensor from 20°C ambient to −15°C — a 35°C drop — reduces dark current by approximately 2^(35/6) ≈ 32×. A sensor measuring 0.5 e⁻/pixel/s at 20°C generates only 0.016 e⁻/pixel/s at −15°C. Thermoelectric (Peltier) cooling to −10°C to −20°C is standard in dedicated astronomical cameras.

Calibration frames: (1) Dark frames: exposures taken with the sensor covered, at the same temperature and exposure duration as the lights; dark current signal + read noise + fixed pattern noise; stacking 20–50 darks as a median produces a master dark with negligible stacking noise; subtract from each light frame. (2) Bias frames: zero-duration exposures capturing the ADC offset (fixed value added to all pixel reads to keep them positive) plus readout fixed pattern noise; 50–100 bias frames median-stacked form a master bias. (3) Flat frames: short exposures of a uniformly illuminated target (dawn/dusk sky at blue hour, diffuse light panel, illuminated tracing paper over aperture) calibrate pixel-to-pixel sensitivity variation (quantum efficiency variation, vignetting from optical system, and dust shadows from particles on the sensor or field flattener); normalize master flat by its median value, then divide each light frame by the normalized master flat to remove multiplicative non-uniformity. Sky flats at dawn require rapid sequential acquisition as sky brightness changes.

Sky-limited SNR formula and sub-frame exposure strategy

The fundamental signal-to-noise ratio formula for a single astrophotography sub-frame:

SNR = (S × t) / √(S×t + B×t + D×t + R²)

where S = object signal rate (e⁻/pixel/s from the target object), B = sky background rate (e⁻/pixel/s from light pollution and airglow), D = dark current rate (e⁻/pixel/s), R = read noise (e⁻ RMS), and t = sub-frame exposure duration in seconds. Sky-limited condition: when B×t ≫ R², the read noise term is negligible and SNR ∝ √t; entering sky-limited operation requires sub-frame duration t ≥ R² / B. Sky background rates by light pollution level: Bortle 9 (inner city, SQM 17–18 mag/arcsec²): B ≈ 100–300 e⁻/pixel/s; Bortle 6 (semi-urban, SQM 19–20): B ≈ 20–60 e⁻/pixel/s; Bortle 4 (rural, SQM 20.5–21.5): B ≈ 5–15 e⁻/pixel/s; Bortle 2 (truly dark site, SQM 21.5–22): B ≈ 0.5–2 e⁻/pixel/s.

Minimum sub-frame duration for BSI sCMOS (R = 1.5 e⁻ RMS) at Bortle 9 (B = 200 e⁻/pixel/s): t_min = 1.5² / 200 = 0.011 s — essentially any sub-frame duration is sky-limited; 30-second subs are therefore efficient. At Bortle 4 (B = 8 e⁻/pixel/s): t_min = 1.5² / 8 = 0.28 s — still very permissive; 3-minute subs are sky-limited. For narrowband imaging with a 5 nm Hα filter reducing sky background by 1,000× (effective B ≈ 0.008 e⁻/pixel/s from Bortle 4 sky): t_min = 1.5² / 0.008 = 281 s ≈ 4.7 min; 15–20 minute sub-frames are standard for narrowband from any site. Total SNR after N sub-frames stacked: SNR_total = SNR_per_sub × √N; total integration time T = N×t. Doubling T always improves SNR by √2 ≈ 41% in the sky-limited regime, regardless of how T is split across sub-frames.

Stacking algorithms: Kappa-Sigma clipping and Winsorized Sigma

Image stacking combines N registered sub-frames to improve SNR and remove transient artifacts. Simple mean stacking gives SNR improvement of √N but is sensitive to outliers: a single satellite trail or cosmic ray in any sub-frame at a given pixel position corrupts the mean indefinitely. Median stacking: sort the N pixel values at each position and return the central value; outliers exceeding the median by any amount are automatically excluded; SNR improvement from median stacking ≈ 0.798√N (about 80% of mean efficiency); requires odd N for clean median definition.

Kappa-Sigma clipping: iterative algorithm that (1) computes the mean μ and standard deviation σ across the stack at each pixel position; (2) marks pixels with values > μ + κσ or < μ − κσ as outliers (satellite trails appear as extreme high outliers; κ = 2.0–3.0; more aggressive κ removes more outliers but may clip legitimate signal from faint nebula emission); (3) recomputes μ and σ from remaining non-outlier pixels only; (4) repeats steps 2–3 for N_iterations (typically 3–5); (5) takes the mean of non-outlier pixels as the output value. Kappa-Sigma requires a minimum of ∼8 sub-frames per position for reliable σ estimation. Winsorized Sigma Clipping: instead of removing outliers from the calculation, replaces them with the clipping boundary value (μ ± κσ) before computing the final mean; preserves sample count for more stable σ estimation with thin stacks (<8 frames); particularly valuable when imaging a target with a heavily satellite-impacted corridor where many sub-frames have outlier values at the same pixel.

Rejection thresholds and iteration count are the patron-calibration-level parameters: the optimal κ depends on the ratio of satellite trail signal to the actual nebula signal at the affected pixel positions; a creator who documents the satellite rejection statistics (percentage of sub-frames rejected at each pixel, histogram of rejection rates across the frame) provides patrons with a reproducible recipe, not just a finished image. Banding noise from electronic interference (horizontal or vertical striping in the image at specific spatial frequencies) requires separate treatment: vertical/horizontal FFT masking in the frequency domain, or a band-pass filter on the problematic axis applied before stacking.

Narrowband emission line imaging: Hα, OIII, SII, and the Hubble palette

Emission nebulae emit light at discrete atomic transition wavelengths corresponding to specific electronic state changes in ionized atoms. Hydrogen alpha (Hα) at 656.3 nm is the Balmer series n = 3 → n = 2 transition, releasing a photon of energy E = hc/λ = (6.626×10⁻³⁴ × 3×10⁸) / 656.3×10⁻⁹ ≈ 3.03×10⁻¹⁹ J = 1.89 eV. Hα is the dominant emission of all HII regions — star-forming clouds where ionizing UV radiation from O and B stars (T > 30,000 K) photoionizes hydrogen atoms; recombining electrons emit Hα as they cascade downward. Hα is also emitted in planetary nebulae (hydrogen shells expelled by red giant stars), supernova remnant shock fronts where the blast wave compresses and heats interstellar medium, and stellar chromospheres.

[OIII] at 500.7 nm (with the weaker companion at 495.9 nm in a 3:1 intensity ratio) is a ‘forbidden line’ from doubly ionized oxygen (O²⁺). Forbidden lines arise from electronic transitions that violate the electric dipole selection rules of quantum mechanics; these transitions have extremely long spontaneous emission lifetimes (seconds to minutes in isolation). At any terrestrial density, such metastable states are quenched by collisions long before spontaneous emission can occur; only at nebular densities of 10–10,000 particles/cm³ (compared to Earth atmosphere’s 2.7×10¹⁹ particles/cm³) do O²⁺ ions survive long enough to emit [OIII]. Excitation is primarily collisional: high-energy free electrons impart energy to O²⁺ ions; the ion emits the forbidden photon as energy is released. [OIII] traces high-ionization zones nearest the central hot star.

[SII] at 671.6 nm and 673.1 nm is the [S II] doublet from singly ionized sulfur (S⁺), also a forbidden line. [SII] is produced in lower-ionization, denser, and cooler regions: the outer envelopes of HII regions where the ionizing UV flux has been attenuated; supernova remnant forward shock regions where gas is compressed and heated behind the shock front to T ≈ 10,000–100,000 K. The [SII] / Hα intensity ratio is a standard diagnostic: [SII]/Hα > 0.4 indicates supernova remnant shock excitation rather than photoionization from stellar UV.

Narrowband filters (bandpass 3–7 nm FWHM) transmit only the emission line wavelength and reject 99.9% of background sky continuum: sodium street lamp emission at 589 nm, mercury vapor lines at 546/577/579 nm, and the broad LED road lighting spectrum are all blocked. Effective sky background rate through a 5 nm Hα filter from Bortle 8 skies: roughly 1/500 of the broadband sky background rate. The Hubble Space Telescope palette (SHO — also called ‘Hubble palette’) maps [SII]→Red, Hα→Green, [OIII]→Blue in false color. This palette was used for the iconic 1995 HST images of the Eagle Nebula (M16) Pillars of Creation. In SHO renditions: HII regions with dominant Hα emission appear green/gold (Hα in green channel); outer OIII shells appear blue-cyan; SII-bright shock fronts and supernova remnant filaments appear red-orange; mixing creates the golden-green pillar hue characteristic of HST emission nebula imagery. Bicolor narrowband (Hα + OIII only, without SII) assigns Hα→Red and OIII→Blue, with Hα+OIII blended →Green; this is more achievable with a 2-panel narrowband setup.

Autoguiding and periodic error: PHD2 and the worm gear

Equatorial mounts track the sky by rotating a worm gear at the sidereal rate: one full worm rotation per worm period (typically 6–12 minutes) turns the RA axis by one tooth of the worm wheel, corresponding to 24 teeth × 10 min = 240 minutes = 4 hours per worm wheel revolution, for a 144-tooth worm wheel. Geometric imperfections in the worm and worm wheel machining produce periodic error (PE): cyclic variation in the tracking speed at the worm period. Budget mount PE: ±20–60 arcsec peak-to-peak; premium mount PE: ±2–8 arcsec; PE causes stars to trail sinusoidally in long sub-frames without correction.

Autoguiding corrects PE in real time using a separate guide camera monitoring a bright guide star. PHD2 (Push Here Dummy 2) captures guide camera exposures every 0.5–2 seconds; the guide star PSF (point spread function) is fitted with a 2D Gaussian function: I(x,y) = I × exp(−((x−x&sub0;)² + (y−y&sub0;)²) / (2σ²)), where x&sub0;,y&sub0; are the centroid coordinates; this Gaussian fit achieves centroid position precision of ∼FWHM/(SNR) in each axis (e.g., FWHM = 5 pixels, SNR = 20 → centroid precision ≈ 0.25 pixels ≈ 0.28 arcsec at 1.1 arcsec/pixel scale). Error signal = centroid position − reference position in arcsec; PHD2’s PI controller (proportional + integral) computes correction pulse duration in milliseconds: correction_ms = (P_gain × error) + (I_gain × integral_of_error); pulses are sent to the mount via ASCOM (Windows) or INDI (Linux) driver as guide pulse commands (PulseGuide or ST-4 interface). Guide RMS error in arcsec (total, combining RA and Dec axes in quadrature): good guiding achieves 0.5–1.0 arcsec total RMS; excellent guiding <0.5 arcsec total RMS; anything above 1.5×pixel_scale introduces noticeable star elongation in sub-frames >3 minutes. Differential flexure is systematic error not correctable by autoguiding: any relative movement between guide optical path and imaging optical path (e.g., guide scope bending under gravity as it rotates through different altitude angles) appears as a DC offset in guide corrections. Off-axis guider (OAG) samples the same primary focal plane as the imaging sensor, eliminating differential flexure at the cost of requiring a sufficiently bright guide star within the small OAG prism field of view.

Atmospheric dispersion correction: Risley prisms and ADC mechanics

Earth’s atmosphere refracts incident starlight toward the zenith because atmospheric density (and therefore refractive index) increases from the top of the atmosphere toward the ground. The atmospheric refractive index for air at sea level: n ≈ 1 + 2.88×10⁻⁴ (approximately); the dispersion of air across the visible spectrum means n is slightly higher at short wavelengths (blue/violet) than long wavelengths (red). Differential atmospheric dispersion = the wavelength-dependent difference in refraction, causing a stellar image to appear as a tiny vertical spectrum elongated toward the horizon: blue is refracted more strongly and appears slightly higher above the horizon than red. The magnitude of differential dispersion scales as tan(z), where z is the zenith angle (= 90° minus altitude above horizon): at z = 45°, differential dispersion across the visible spectrum (400–700 nm) is approximately 3–4 arcsec; at z = 60°, approximately 6–8 arcsec; at z = 75°, 15–20 arcsec or more. Critical for high-resolution planetary imaging (<0.3 arcsec/pixel systems) at planets below 40° altitude.

An ADC (Atmospheric Dispersion Corrector) uses a Risley prism pair: two identical wedge prisms mounted on concentric, independently rotatable rings. A single wedge prism refracts all wavelengths by a fixed angle proportional to its apex angle and the glass dispersion; when two identical prisms are mounted apex-to-apex (aligned, same direction), their dispersions add to produce maximum correction; when rotated 180° relative to each other (apex-to-base), their dispersions cancel and there is zero net refraction. Counter-rotating both prisms relative to each other between 0° (maximum correction) and 180° (zero correction) varies the applied dispersion continuously; rotating both prisms together (changing the azimuthal orientation of the combined correction) aligns the correction direction with the vertical (zenith direction) at the current pointing. Correct ADC operation: (1) align the ADC so its correction axis is parallel to the local vertical (horizon direction); (2) adjust the counter-rotation angle to match the current altitude of the target, verified by minimizing the wavelength-dependent PSF elongation observed with a defocused bright star in the guide camera at high magnification; (3) readjust as altitude changes during a session. For solar system imaging with monochrome cameras in RGB or narrowband, the ADC must be optimized for each narrowband wavelength if multiple filters are used in sequence.

iOS rates and the Apple Tax

Astrophotography creators distribute content across YouTube (full tutorials, imaging session walkthroughs, processing tutorials), Instagram (final image showcases), and Discord (patron community for image feedback and real-time session sharing). YouTube astrophotography and astronomy: 55–68% iOS — the equipment-research audience (comparing telescope specifications, mount periodic error specs, and camera QE curves) tends to use more desktop browsers than purely aesthetic photography audiences. Instagram astrophotography: 70–80% iOS. Discord-hosted content operates on a separate billing model outside Patreon’s iOS billing.

Frequently asked questions

Related: Patreon for astrophotography creators · Patreon for amateur astronomers · How the Apple Tax works · All explainers