Differential Chromatic Refraction

• Differential Chromatic Refraction (DCR) is the wavelength-dependent displacement of light caused by atmospheric dispersion, altering the apparent position and PSF based on a source's SED and observing geometry.
• In astronomical imaging, DCR manifests as color-dependent centroid shifts and PSF elongation, leading to biases in flux measurements and astrometric errors if not properly corrected.
• Correction strategies range from empirical astrometric calibration to forward-modeling of PSF-level chromatic effects, crucial for improving measurements in weak-lensing, supernova cosmology, and fiber spectroscopy.

Searching arXiv for recent and foundational papers on Differential Chromatic Refraction. Differential Chromatic Refraction (DCR) is the wavelength-dependent displacement of an astronomical source’s apparent position caused by atmospheric refraction. Because the refractive index of air decreases with wavelength, shorter-wavelength light is bent more strongly than longer-wavelength light, so the apparent position, centroid, and effective point-spread function (PSF) of a source depend on its spectral energy distribution (SED), the bandpass, and the observing geometry. In broadband imaging, DCR appears as a color-dependent centroid shift toward the zenith and as a wavelength-dependent elongation of the PSF along the parallactic direction; in spectroscopy and high-contrast imaging the same underlying phenomenon is often discussed as atmospheric dispersion or differential atmospheric refraction (DAR) (Lee et al., 2023).

1. Physical basis and terminology

DCR arises because Earth’s atmosphere is a dispersive medium: its refractive index $n(\lambda)$ varies with wavelength, so light at shorter wavelengths is bent more strongly than light at longer wavelengths. The first-order description used across the literature is

$R(\lambda, z) \approx [n(\lambda)-1]\tan z,$

with differential refraction between two wavelengths given by

$$\Delta R(\lambda_1,\lambda_2,z)\approx [n(\lambda_1)-n(\lambda_2)]\tan z.$$

This dependence on $\tan z$ makes the effect grow with zenith distance, or equivalently with airmass, and the chromatic term is largest in bluer passbands (Lee et al., 2023).

The observed displacement is directed along the zenith direction, usually parameterized on the sky by the parallactic angle. In imaging work this means that DCR is not an arbitrary astrometric perturbation: it is highly anisotropic, aligned with the altitude or parallactic direction, and its sign and amplitude are controlled by observing geometry and source color. For objects observed near the meridian, that projection can become predominantly declinational, which is why DCR in Pan-STARRS1 mean positions was treated primarily as a declination bias (White et al., 2022).

The same physics is described under slightly different names in different subfields. Weak-lensing and time-domain imaging papers generally use “Differential Chromatic Refraction,” while high-contrast imaging and instrumentation papers often use “Differential Atmospheric Refraction” or “atmospheric dispersion.” The underlying phenomenon is the same wavelength-dependent atmospheric refraction, but the emphasis changes with application: centroid shifts and PSF anisotropy in survey imaging, residual dispersion after an atmospheric dispersion corrector in coronagraphy, and wavelength-dependent flux losses in multi-object spectroscopy (Noel et al., 2024).

2. Mathematical description, broadband centroids, and geometry

For broadband observations, DCR is not set by a single monochromatic wavelength but by an SED-weighted refraction across the full passband. A physically transparent description integrates the source SED and total throughput over wavelength. In the DES supernova analysis, the color-dependent centroid shift relative to a reference spectrum is written as

$$\delta \approx \frac{\int S(\lambda)\,T(\lambda)\,R(\lambda,z)\,d\lambda}{\int S(\lambda)\,T(\lambda)\,d\lambda} - \mathrm{reference},$$

where $S(\lambda)$ is the source SED and $T(\lambda)$ is the system throughput (Lee et al., 2023). Closely related formulations appear in weak-lensing analyses through the photon-weighted first moment $\bar R$ and variance $V$ of the DCR kernel (Meyers et al., 2014).

This formalism makes explicit why DCR is source-dependent. Two objects observed in the same exposure, at the same airmass and through the same filter, can have different effective refractions because their SEDs weight the bandpass differently. The distinction is especially important for objects with time-variable spectra. Type Ia supernovae evolve in color and SED with phase, so the DCR-induced centroid and PSF shape evolve over the light curve; stellar flares become bluer during impulsive heating, shifting their effective wavelength and causing measurable zenith-ward centroid motion in a single band (Lee et al., 2023).

The sky projection is determined by the parallactic angle. In the DES-SN5YR appendix, the parallactic-angle relations are

$$\sin(q)=\frac{-\cos(\phi)\sin(A)}{\cos(\delta)}, \qquad \cos(q)=\frac{\sin(\phi)-\sin(\delta)\sin(a)}{\cos(\delta)\cos(a)},$$

with $R(\lambda, z) \approx [n(\lambda)-1]\tan z,$0 the observer latitude, $R(\lambda, z) \approx [n(\lambda)-1]\tan z,$1 azimuth, $R(\lambda, z) \approx [n(\lambda)-1]\tan z,$2 declination, and $R(\lambda, z) \approx [n(\lambda)-1]\tan z,$3 altitude (Lee et al., 2023). In astrometric and flare applications, the displacement is commonly projected into equatorial coordinates through

$R(\lambda, z) \approx [n(\lambda)-1]\tan z,$4

or into a scalar zenith-directed component such as $R(\lambda, z) \approx [n(\lambda)-1]\tan z,$5, which isolates the atmospheric signal along the vertical circle (Lin et al., 2020).

Several papers also define an effective wavelength to compress the band-integrated SED dependence into a single chromatic coordinate. In survey photometry this can be written as

$R(\lambda, z) \approx [n(\lambda)-1]\tan z,$6

while flare studies adopt a log-weighted form for $R(\lambda, z) \approx [n(\lambda)-1]\tan z,$7 inside a filter (Lee et al., 2023). These constructions are not identical across papers, but they serve the same purpose: mapping a broadband SED to an effective chromatic refraction.

3. Imaging manifestations: astrometry, PSF morphology, and photometric bias

In imaging data, DCR manifests as a centroid shift of the source along the direction to the zenith and as a wavelength-dependent elongation of the PSF. In weak-lensing language, single-exposure DCR adds variance only along altitude, so the DCR-induced second-moment error is an altitude-aligned excess $R(\lambda, z) \approx [n(\lambda)-1]\tan z,$8 in $R(\lambda, z) \approx [n(\lambda)-1]\tan z,$9; multi-epoch centroid differences then produce additional second-moment errors in stacked images through color-dependent misregistration (Meyers et al., 2014). The principal consequence is that a stellar PSF is not the correct effective PSF for a galaxy or transient with a different SED.

This mismatch propagates directly into measurement bias. In DES Scene Modeling Photometry (SMP), the supernova flux estimator is written schematically as

$$\Delta R(\lambda_1,\lambda_2,z)\approx [n(\lambda_1)-n(\lambda_2)]\tan z.$$0

so even small centroid errors or width mismatches change the PSF weights $$\Delta R(\lambda_1,\lambda_2,z)\approx [n(\lambda_1)-n(\lambda_2)]\tan z.$$1 and bias the inferred flux (Lee et al., 2023). Simple tests with Gaussian and Moffat PSFs show that a centroid error of $$\Delta R(\lambda_1,\lambda_2,z)\approx [n(\lambda_1)-n(\lambda_2)]\tan z.$$2 of the PSF FWHM induces a $$\Delta R(\lambda_1,\lambda_2,z)\approx [n(\lambda_1)-n(\lambda_2)]\tan z.$$3–$$\Delta R(\lambda_1,\lambda_2,z)\approx [n(\lambda_1)-n(\lambda_2)]\tan z.$$4 mmag magnitude bias, and a PSF-width mismatch of $$\Delta R(\lambda_1,\lambda_2,z)\approx [n(\lambda_1)-n(\lambda_2)]\tan z.$$5 induces $$\Delta R(\lambda_1,\lambda_2,z)\approx [n(\lambda_1)-n(\lambda_2)]\tan z.$$6 mmag biases (Lee et al., 2023).

The DES-SN5YR analysis separated these effects into astrometric “COORD” corrections and PSF-shape “SHAPE” corrections. Ignoring the astrometric offsets and PSF shape changes causes average biases of $$\Delta R(\lambda_1,\lambda_2,z)\approx [n(\lambda_1)-n(\lambda_2)]\tan z.$$7 mmag and $$\Delta R(\lambda_1,\lambda_2,z)\approx [n(\lambda_1)-n(\lambda_2)]\tan z.$$8 mmag respectively, with standard deviations of $$\Delta R(\lambda_1,\lambda_2,z)\approx [n(\lambda_1)-n(\lambda_2)]\tan z.$$9 mmag and $\tan z$0 mmag across all DES observing bands (\textit{griz}) throughout all redshifts used in the analysis (Lee et al., 2023). The largest corrections appear in the bluest bands: for COORD, $\tan z$1 of $\tan z$2-band corrections exceed $\tan z$3 mag, while for SHAPE about $\tan z$4 of $\tan z$5 and $\tan z$6 corrections exceed $\tan z$7 mag (Lee et al., 2023).

Weak-lensing analyses reach the same conclusion at the level of shear systematics. Uncorrected DCR and wavelength-dependent seeing induce PSF moment errors that map into multiplicative and additive shear biases, and the paper “Impact of Atmospheric Chromatic Effects on Weak Lensing Measurements” shows that uncorrected atmospheric chromatic effects can bias shear beyond DES and LSST statistical limits (Meyers et al., 2014). A central result of that work is that catalog-level corrections are insufficient because a “chromatic model bias” appears when a galaxy convolved with its galactic effective PSF is fitted with a stellar effective PSF.

4. Empirical and forward-model correction strategies

A major methodological division in the literature is between empirical astrometric calibration and explicit chromatic forward modeling. In Pan-STARRS1, DCR in mean positions was corrected empirically by iteratively subtracting color- and declination-dependent PS1/Gaia EDR3 declination residuals. For about $\tan z$8 million point-like reference objects cross-matched to Gaia EDR3, Gaia EDR3 provided a $\tan z$9 improvement over Gaia DR2, and DCR corrections provided an additional $$\delta \approx \frac{\int S(\lambda)\,T(\lambda)\,R(\lambda,z)\,d\lambda}{\int S(\lambda)\,T(\lambda)\,d\lambda} - \mathrm{reference},$$0 improvement; for blue objects with $$\delta \approx \frac{\int S(\lambda)\,T(\lambda)\,R(\lambda,z)\,d\lambda}{\int S(\lambda)\,T(\lambda)\,d\lambda} - \mathrm{reference},$$1 observed $$\delta \approx \frac{\int S(\lambda)\,T(\lambda)\,R(\lambda,z)\,d\lambda}{\int S(\lambda)\,T(\lambda)\,d\lambda} - \mathrm{reference},$$2 away from the zenith, the astrometric improvement reached $$\delta \approx \frac{\int S(\lambda)\,T(\lambda)\,R(\lambda,z)\,d\lambda}{\int S(\lambda)\,T(\lambda)\,d\lambda} - \mathrm{reference},$$3 (White et al., 2022).

Gaia-based frame-level calibration has also been used for targeted astrometry. The Yunnan Observatory study modeled the zenith-direction DCR term as

$$\delta \approx \frac{\int S(\lambda)\,T(\lambda)\,R(\lambda,z)\,d\lambda}{\int S(\lambda)\,T(\lambda)\,d\lambda} - \mathrm{reference},$$4

with $$\delta \approx \frac{\int S(\lambda)\,T(\lambda)\,R(\lambda,z)\,d\lambda}{\int S(\lambda)\,T(\lambda)\,d\lambda} - \mathrm{reference},$$5, and calibrated filter-specific DCR coefficients using Gaia DR2 stars (Lin et al., 2020). For observations taken through a Null filter, the median of the mean observed-minus-computed residual for well-exposed stars decreased from $$\delta \approx \frac{\int S(\lambda)\,T(\lambda)\,R(\lambda,z)\,d\lambda}{\int S(\lambda)\,T(\lambda)\,d\lambda} - \mathrm{reference},$$6 mas to $$\delta \approx \frac{\int S(\lambda)\,T(\lambda)\,R(\lambda,z)\,d\lambda}{\int S(\lambda)\,T(\lambda)\,d\lambda} - \mathrm{reference},$$7 mas after DCR correction (Lin et al., 2020). A related two-filter astrometric method later used the contrast in filter-dependent $$\delta \approx \frac{\int S(\lambda)\,T(\lambda)\,R(\lambda,z)\,d\lambda}{\int S(\lambda)\,T(\lambda)\,d\lambda} - \mathrm{reference},$$8 coefficients to infer an object’s Gaia-matched color index directly from DCR, yielding a mean BP-RP of $$\delta \approx \frac{\int S(\lambda)\,T(\lambda)\,R(\lambda,z)\,d\lambda}{\int S(\lambda)\,T(\lambda)\,d\lambda} - \mathrm{reference},$$9 magnitude for Himalia from $S(\lambda)$0 CCD frames (Guo et al., 2023).

Forward-model approaches treat DCR as part of the PSF rather than as a catalog residual. In DES-SN5YR, COORD corrections were derived from measured astrometric offsets of calibration stars versus stellar $S(\lambda)$1 color in each exposure, while SHAPE corrections were computed with GalSim using Moffat PSFs, atmospheric dispersion, and wavelength-dependent seeing scaling as $S(\lambda)$2 and $S(\lambda)$3 (Lee et al., 2023). The resulting lookup tables were evaluated by trilinear interpolation at the observed airmass, stellar FWHM, and supernova $S(\lambda)$4 color (Lee et al., 2023).

For weak lensing, the recommended strategy is more stringent: apply PSF-level perturbative corrections in an ordered sequence. The atmosphere-first description in “Impact of Atmospheric Chromatic Effects on Weak Lensing Measurements” corrects a stellar PSF to a fiducial monochromatic PSF, interpolates that PSF, then reapplies galaxy-specific chromatic seeing and DCR perturbations using parameters predicted from six-band photometry with Extra Trees Regression (Meyers et al., 2014). The paper explicitly concludes that catalog-level corrections do not address chromatic model bias, whereas ordered PSF-level corrections do.

5. Applications beyond nuisance correction

Although DCR is commonly treated as a systematic effect, several recent studies use it as an information-carrying observable. The paper “Subband Image Reconstruction using Differential Chromatic Refraction” uses the airmass- and parallactic-angle dependence of broadband astrometric shifts to solve for multiple latent subband images through a generalized deconvolution procedure based on robust statistics (Lee et al., 2018). In that framework, DCR becomes a spectro-astrometric encoding of intra-band SED information rather than a distortion to be removed.

Time-domain astronomy has extended this idea further. “Every Datapoint Counts: Stellar Flares as a Case Study of Atmosphere Aided Studies of Transients in the LSST Era” proposed using DCR as a single-epoch color thermometer for stellar flares in LSST data (Clarke et al., 2024). For an M5 star plus flare, the paper models g-band centroid offsets of $S(\lambda)$5 at $S(\lambda)$6 and $S(\lambda)$7 at $S(\lambda)$8 for a $S(\lambda)$9 flare, and argues that flare temperatures at or above $T(\lambda)$0 can be constrained by a single g-band observation at airmass $T(\lambda)$1 given LSST’s minimum specified requirement on single-visit relative astrometric accuracy (Clarke et al., 2024).

That proposal was realized observationally in “First Temperature Profile of a Stellar Flare using Differential Chromatic Refraction,” which reports the first derivation of a stellar flare temperature profile from single-band photometry (Clarke et al., 25 Jul 2025). For a giant DECam $T(\lambda)$2-band flare on an M7 dwarf observed at $T(\lambda)$3, the paper measured zenith-ward centroid drift, modeled the flare SED as a blackbody plus line contributions, and—under its most physically motivated scenario—found a maximum temperature of $T(\lambda)$4 and a duration of about $T(\lambda)$5 minutes above $T(\lambda)$6 (Clarke et al., 25 Jul 2025). The same study emphasizes that ignoring line emission can overestimate temperature.

Supernova cosmology has also begun to use DCR constructively. “Astrometric Redshifts of Supernovae” models the DCR time series of LSST Deep Drilling Field SNe Ia and uses the resulting astrometric offsets to constrain redshift (Lee et al., 2024). For a conservative $T(\lambda)$7-mas systematic uncertainty floor, the paper finds that the astrometric redshift estimation is accurate at $T(\lambda)$8, and that combining DCR-based astrometric redshifts with host-galaxy photometric redshifts and supernova light-curve redshifts considerably improves the overall redshift estimates (Lee et al., 2024). This suggests that DCR is becoming a measurable transient observable, not only a calibration term.

6. Instrumentation, residual dispersion, and future survey requirements

Instrumental design determines whether DCR is exposed, tolerated, or actively corrected. Rubin/LSST and DECam do not use atmospheric dispersion correctors, so DCR remains present in survey imaging and must be modeled in astrometry, PSF construction, and scene-modeling photometry (Clarke et al., 2024). By contrast, high-contrast imagers and fibre spectrographs often attempt active correction, but the residuals remain scientifically important.

The Gemini Planet Imager study provides a direct example of incomplete correction. Its atmospheric dispersion corrector was designed to reduce residual DAR below $T(\lambda)$9 mas, but measured averages are $\bar R$0 mas in $\bar R$1 band and $\bar R$2 mas in $\bar R$3 band (Noel et al., 2024). The analysis attributes the $\bar R$4-band residual largely to a nearly constant undercorrection of about $\bar R$5 mas and the $\bar R$6-band residual largely to perpendicular dispersion induced by the ADC, which accounts for $\bar R$7 of the residual DAR in that band (Noel et al., 2024). The same paper also shows that the Roe-like refractivity model used by the instrument underestimates humidity’s impact on DAR and introduces a new humidity-inclusive approximation for $\bar R$8 from $\bar R$9 to $V$0m (Noel et al., 2024).

In fibre spectroscopy, the dominant observable is wavelength-dependent flux loss rather than centroid bias alone. “Atmosphyre: Modelling Atmospheric Chromatic Dispersion for Multi-Object Spectrographs” models the chromatic displacement

$V$1

and its impact on fibre coupling (Stephan et al., 2024). The paper concludes that the guiding wavelength should typically be bluer than the observing-band mid-wavelength, around $V$2–$V$3 of the way through the band, and finds for MOSAIC that differential losses greater than $V$4 are unavoidable for $V$5 h observations after a local hour angle of $V$6 h or at declinations below $V$7 and above $V$8 (Stephan et al., 2024). It therefore argues that an atmospheric dispersion corrector would significantly reduce spectral distortions and increase survey speed.

Future survey guidance is consistent across these domains. The DES-SN5YR paper recommends deriving per-exposure, and even per-object, color estimates; incorporating atmospheric dispersion explicitly in PSF models; using improved atmospheric refractivity models with contemporaneous pressure, temperature, and humidity; and, where possible, using full SED information rather than a single color (Lee et al., 2023). Weak-lensing work similarly emphasizes exposure-specific $V$9 and $$\sin(q)=\frac{-\cos(\phi)\sin(A)}{\cos(\delta)}, \qquad \cos(q)=\frac{\sin(\phi)-\sin(\delta)\sin(a)}{\cos(\delta)\cos(a)},$$0, PSF-level chromatic corrections, and accurate throughput knowledge (Meyers et al., 2014). A plausible implication is that DCR will increasingly be treated as a first-class component of forward image formation in high-precision survey pipelines, especially for $$\sin(q)=\frac{-\cos(\phi)\sin(A)}{\cos(\delta)}, \qquad \cos(q)=\frac{\sin(\phi)-\sin(\delta)\sin(a)}{\cos(\delta)\cos(a)},$$1-band observations and high-airmass data, where its amplitude grows rapidly (Lee et al., 2023).

7. Limitations, misconceptions, and scope

A common misconception is that DCR is only an astrometric shift. The literature consistently shows that this is incomplete. In broadband data, DCR also alters second moments, induces PSF anisotropy aligned with the parallactic direction, and can create photometric bias when stellar PSFs are applied to objects with different SEDs (Meyers et al., 2014). In time-domain photometry, the resulting bias can remain at the millimagnitude level in DES yet still become important for future higher-precision surveys (Lee et al., 2023).

Another misconception is that a single broadband color fully characterizes DCR. Several papers use color proxies—such as $$\sin(q)=\frac{-\cos(\phi)\sin(A)}{\cos(\delta)}, \qquad \cos(q)=\frac{\sin(\phi)-\sin(\delta)\sin(a)}{\cos(\delta)\cos(a)},$$2 for supernovae or BP-RP for astrometric calibration—because they are practical and often effective, but the limitations are explicit. The DES supernova analysis notes that using $$\sin(q)=\frac{-\cos(\phi)\sin(A)}{\cos(\delta)}, \qquad \cos(q)=\frac{\sin(\phi)-\sin(\delta)\sin(a)}{\cos(\delta)\cos(a)},$$3 as a proxy for the full supernova SED ignores phase-dependent spectral features within the bandpass; host-galaxy light is assumed static and non-dispersed; interpolation is performed on precomputed grids rather than exact exposure conditions; and focal-plane optical variations such as lateral color are averaged over rather than modeled explicitly (Lee et al., 2023). These are approximations, not exact equivalences.

A further misconception is that DCR can always be removed at the catalog level after image measurement. Weak-lensing analyses show that this is false in the presence of chromatic model bias: if the wrong PSF is used during model fitting, later catalog corrections cannot reconstruct the measurement that would have been obtained with the correct chromatic PSF (Meyers et al., 2014). This is why PSF-level or scene-level chromatic modeling is repeatedly favored in high-precision applications.

The current literature therefore places DCR in a dual role. It is simultaneously a calibration complication and a measurable atmospheric encoding of source SED. In astrometry and photometry it must often be corrected to avoid bias; in flare physics, subband reconstruction, and supernova redshift inference it can be exploited as a signal (Clarke et al., 2024). This duality is central to the modern treatment of DCR: its scientific meaning depends less on the underlying atmospheric physics—which is fixed—than on whether the observing program regards chromatic refraction as contamination, calibration leverage, or an observable in its own right.