WorksPhysics and Astronomy
DustER: Dust Extinction with Photometric Red Galaxies
Abstract
We present DustER, a new method for reconstructing dust reddening maps from photometric galaxy survey data. DustER uses red galaxies as "standard crayons". That is, it relies on the mean color of red-sequence galaxies being independent of their position in the sky. Given the galaxy's redshift, the difference between a galaxy's observed color and its expected color measures the reddening by dust along the galaxy's line of sight. When using photometric galaxy survey data, the method must rely on photometric redshifts, which, combined with selection effects, bias the recovered dust maps. DustER uses a novel approach to characterize and correct for these biases by leveraging the known statistical properties of the distribution of dust around the Milky Way. Testing on simulations, we find that the recovered dust maps are contaminated by large-scale structure. However, when we use the DustER map to deredden and select red-sequence galaxies, we find the resulting two-point function is not contaminated by dust.
Introduction
Interstellar dust in the Milky Way is concentrated in a thin layer along the Galactic disk. Because we observe from within this layer, light from any extragalactic source must travel through part of it before reaching us. At optical wavelengths, dust grains both absorb and scatter light, making sources appear fainter and redder than they are. Consequently, the photometry of extragalactic sources needs to be corrected for the reddening and attenuation accumulated along the line of sight. Performing these corrections requires two things: 1) the total dust column along each line of sight; and 2) the wavelength dependence of dust extinction, i.e., the reddening law. This paper addresses the first of these. For this purpose only the projected dust column matters: whereas studies of Galactic structure require the three-dimensional distribution of dust, extragalactic photometry is affected equally by dust anywhere along the line of sight.
There are several methods for determining the distribution of dust in the Galaxy. The first is through its emission in the infrared [e.g. 1, henceforth referred to simply as SFD]. However, these can be contaminated by large-scale structure (LSS) due to confusion with the Cosmic Infrared Background (CIB) [2]. The addition of spatial templates can be used to decorrelate the dust from the CIB emission at the cost of added variance on the map [3]. A second approach relies on tracers whose abundance correlates with dust, e.g. HI [4] and Polycyclic Aromatic Hydrocarbons (PAH) [5]. As with IR-based maps, PAH-based maps exhibit residual large-scale structure contamination from the diffuse PAH background, whereas HI maps appear to be free from large-scale structure contamination [2]. However, HI-based maps assume a fixed dust-to-gas ratio and miss dust associated with molecular gas, which restricts them to regions of low column density ( mmag) [4]. The third approach relies on measuring dust reddening directly, using sources of known intrinsic color: it measures the color excess of the sources themselves, which is the quantity needed to correct photometry, rather than inferring reddening from dust emission or from the abundance of a tracer through an assumed conversion. In this method the sources can be either stars [6, 7] or galaxies [8]. When using stars, if the distance to the calibration stars is known, then one can reconstruct three-dimensional maps of the Milky Way dust [9], a method that is particularly powerful in combination with Gaia data [10, 11, 12]. However, because of the disk structure of our Galaxy, the resolution of these maps decreases with increasing Galactic latitude, and the maps are limited to a distance of kpc [7]. Further, and somewhat surprisingly, stellar-based dust maps also suffer from contamination by large-scale structure, likely due to quasar/stellar confusion [2].
When the sources of known intrinsic color are galaxies, the method is referred to as the “standard crayon” method, first proposed by Peek & Graves [8] and improved by Bravo et al. [13]. Conceptually, in this method one identifies a population of extragalactic objects of known Spectral Energy Distribution (SED). By comparing the observed colors of these objects to their expected color, one can determine the amount of reddening due to dust along the line of sight to the object. Relative to IR or tracer-based methods, the standard crayon approach has the advantage of measuring dust reddening directly. Relative to the stellar-based methods, standard crayon maps benefit from uniform resolution and the guarantee that they are sensitive to the entirety of the dust column along each line of sight, since the sources lie well beyond the Galaxy. However, there is an important disadvantage: the standard crayon method is limited in resolution by the availability of extragalactic standard crayon sources. The standard crayon method also has another apparent disadvantage, namely, it requires the adoption of a reddening law a priori. However, every dust map requires a reddening law before it can be used to correct photometry: emission- and tracer-based maps are calibrated to , and converting into the extinction in any other band requires a reddening law. Consequently, this is not a significant drawback.
The main challenge of the standard crayon method is identifying the crayons. The original Peek & Graves analysis relied on early-type galaxies that were selected based on their spectroscopic features. The SED for these galaxies was then calibrated using spectra from galaxies along lines of sight with negligible dust column density. The most important drawback of the Peek & Graves approach is the reliance on spectroscopic data sets, which limits the resolution of the resulting dust maps, and prevents us from applying their algorithm on purely photometric data sets.
Bravo et al. proposed a generalization of the Peek & Graves methodology that simultaneously addresses both of the above drawbacks. They propose to bin galaxies in photometric redshift bins, assuming the latter are unbiased and have low scatter (). As in the Peek & Graves analysis, the true mean color of galaxies in a given photometric redshift bin is determined from galaxies at locations with negligible dust column density. They then measure the color offset relative to the predicted galaxy color, and use this offset to estimate the dust column along the line of sight to each galaxy. Averaging over many galaxies yields an estimate of the dust column in each sky pixel. In practice, however, photometric redshifts are generically neither Gaussian nor unbiased, presenting a challenge to this method. DustER addresses this challenge by relying on redMaGiC galaxies [14] which are selected precisely because of the quality of their photometric redshifts. Unlike previous approaches, which calibrate the crayons on lines of sight with negligible dust, DustER calibrates the red sequence on the raw photometry of the full survey footprint, so that no dust-free calibration region is required. This choice biases the initial reconstruction, which is also noisy. DustER therefore includes a Wiener filtering step and a statistical debiasing step, which reduce the noise and correct for these biases. We conclude by testing the resulting maps for large-scale structure contamination, finding that a residual contamination is present and that a linear template-based cleaning does not remove it. Nevertheless, when we use the DustER map to deredden and select red-sequence galaxies in the simulation, the clustering of the resulting sample agrees with that obtained using the true dust map well within full-sky cosmic variance. That is, in our simulations, dereddening with DustER does not measurably bias the clustering of red-sequence galaxies.
The layout of this paper is as follows. In § 2 we introduce notation and conventions. § 3 describes the Cardinal catalog and the procedure used to incorporate dust reddening and photometric noise. In § 4 we present the DustER algorithm. § 5 contains validation tests using the dust-modified Cardinal catalog, including controlled tests of how dust and survey depth affect red-galaxy selection, and an analysis of large-scale structure imprints and mitigation strategies. We conclude in § 6.
Dust Conventions
Because DustER infers the dust column from color excesses measured through broadband filters, it is important to be precise about what quantity a dust map represents. The Milky Way’s dust distribution is usually described using maps labeled as , where is usually described as the excess Johnson color caused by dust reddening. In this work we distinguish between the observed color excess and the underlying dust density field, which we denote as the extinction field . Concretely, we take to be proportional to the dust column density along the line of sight, with the proportionality constant fixed by the optically thin relation . Note that at high attenuation this linear relation breaks down for broadband filters due to the wavelength dependence of dust attenuation integrated over the bandpass (see equation (8)). As a result, the color excess is no longer proportional to the dust density , and the equality is violated.
Of course, it is generally the case that one is interested in the dust density rather than the actual color excess, so the latter has become synonymous with dust density. For example, the SFD maps are maps of scaled so that in the optically thin regime [1]. Although SFD makes the distinction clear, this notation turns the physically correct statement into the seemingly self-contradictory in regions of high column density. We therefore always denote the dust column by .
To illustrate the difference between and , consider a source with flux , and let be the transmission fraction given a bandpass . We will assume the source is attenuated by dust with a reddening law , where gives the shape of the wavelength dependence of the extinction curve ,
with the reference wavelength chosen by convention. The constant determines the attenuation strength of the dust, and is proportional to the dust column density. Given this setup, the source’s observed flux is attenuated by a factor
and therefore its attenuated magnitude is larger than its unattenuated magnitude by an additive constant
Given equation (3), we can calculate , and hence , for any source, dust attenuation strength , and reddening law . In the limit of low attenuation, , which leads to
where
Adopting in equation (2) and an O’Donnell [15] reddening law, we find and , and therefore
We define the extinction field
so that in the limit of low extinction. It is this quantity that is mapped out in traditional dust maps. Given , we can then readily compute via equation (3). The origin of the non-linearity between and becomes clear when equation (3) is expanded to second order in :
where is defined as in equation (5) with replaced by . The leading-order correction is set by the variance of the extinction curve across the band, computed with the same weighting as equation (5), and all corrections vanish if is constant across the band. At fixed central wavelength, this variance grows with the width of the bandpass: approximating as linear across the band gives , where and are the mean and the variance in wavelength across the band. Narrow bands therefore remain linear in up to high attenuation, whereas very broad bands are strongly non-linear.
Figure 1 shows the color excess as a function of the extinction field . Also shown for reference is the line. We see the two curves match at low extinction, but differ at high extinction. This is why it is necessary to differentiate between and , since is not always linear in . In this work, we are concerned with mapping out the extinction field .
Figure 1. Color excess as a function of the extinction field , computed for the Johnson and bandpasses with in equation (2) and an O’Donnell reddening law [15]. The dashed line corresponds to the optically thin limit in which . At higher extinction, the linear relation breaks down due to the wavelength-dependent dust attenuation integrated over the bandpass, highlighting the importance of distinguishing these two quantities.
Simulations
DustER is an algorithm for recovering the distribution of Milky-Way dust using the colors of photometrically-selected red galaxies. We test DustER using Cardinal, a simulated photometric data set designed to match DES Y3 data [16]. This synthetic data set is useful not only for testing DustER, but also for illustrating each of the steps in the algorithm as it is described. Readers unconcerned with how the simulation data is generated need to know only two things: 1) our simulated data set takes the form of a photometric galaxy catalog in which every galaxy has been attenuated by dust using either the SFD or Planck GNILC map (see § 3.2), and then “observed” in a way that mimics the DES Y3 data; and 2) the galaxies in the galaxy catalog are clustered in a way that mimics DES Y3 data. If the reader is willing to believe these two properties of our synthetic data set, they can skip to § 4 for a description of the DustER algorithm. We use simulation tests to assess the robustness of our results to systematic uncertainties.
The Cardinal Simulation
We test the DustER algorithm using synthetic data sets generated from the Cardinal simulation [16]. Cardinal populates a lightcone simulation with galaxies of -band magnitude over the redshift range . The lightcone is constructed using three N-body simulations with sizes , , and , and mass resolution of , , and . The N-body simulations are generated using a CDM cosmology with , , , , , , and three massless neutrino species with . Galaxies are assigned to dark-matter particles by giving them galaxy positions, velocities, and absolute magnitudes based on their Lagrangian overdensities. The relation between galaxy properties and the overdensities is derived from a subhalo abundance matching model (SHAM) constrained by the galaxy-galaxy correlations and group-galaxy correlations measured in SDSS. The resulting model approximately reproduces the luminosity functions, star formation histories, and observed luminosity- and color-dependent clustering of galaxies. Additionally, Cardinal includes orphan subhalos as constrained by the galaxy-galaxy and group-galaxy correlations. Cardinal then uses a conditional abundance matching algorithm to assign every galaxy an SED from a Sloan Digital Sky Survey (SDSS) galaxy, which then is used to calculate the galaxy’s magnitudes. For a more detailed description of Cardinal we refer the reader to [16].
Adding Dust
The magnitudes of the galaxies in the Cardinal galaxy catalog are dust-free. As light from distant sources propagates through the dust clouds of the Milky Way, some fraction of the light will be absorbed or scattered. We refer to the magnitude of a galaxy observed through dust at infinite signal-to-noise as the true dusty magnitude of the galaxy, . This is related to the true magnitude of the galaxy via
where is the extinction field, and is a band-dependent extinction coefficient.1 Note that because we assume linearity, our analysis must be limited to low-extinction regions of the sky. For the purposes of our fiducial simulation, we will assume the extinction field is given by the famous Schlegel, Finkbeiner, and Davis dust map [1], referred to simply as the SFD map. To test whether our results depend on this choice, we also generate a second catalog in which the extinction field is given by the Planck GNILC dust map [3] (henceforth P16-GNILC); the two catalogs are otherwise identical. Unless stated otherwise, all results refer to the fiducial SFD-reddened catalog. The P16-GNILC catalog is used only in § 5.1 to test the robustness of DustER to the input dust map.
The true dusty magnitudes are then used to generate observed magnitudes for each of the galaxies in Cardinal using the prescription described in [17, 18]. Briefly, let be the effective exposure time and be the number of photons that impact the telescope in this time. The expectation value of the number of photons is given by
where is the zero-point magnitude. We assume the survey depth in each band is characterized in terms of the limiting magnitude .2 From our Poisson assumption, it follows that the expectation value for the number of noise photons associated with a galaxy of magnitude is
The observed flux of a galaxy can therefore be written as
where denotes a random draw from a Poisson distribution with mean . Hence, the observed magnitudes and magnitude errors in the presence of dust are given by
and
where . Unfortunately, the background subtraction can lead to negative fluxes, and therefore ill-defined magnitudes. Consequently, we use arcsinh magnitudes (sometimes referred to as luptitudes) as proposed in [19]. With this convention, one has
and
where is the flux associated with a galaxy with and is defined as a softening parameter which determines the flux level at which linear behavior sets in. In Table 1 we provide the numerical values of the extinction coefficients and softening parameters used in this work. Finally, we generate synthetic luptitudes and luptitude errors for every CARDINAL galaxy as described above. The resulting galaxy catalog is the input to DUSTER.
| Band | ||
| 3.186 | ||
| 2.140 | ||
| 1.569 | ||
| 1.196 |
Table 1. Extinction coefficients and softening parameters with their respective photometric bands used in this work. The extinction coefficients are the fiducial DES DR1 values [20]. These extinction coefficients are appropriate for when dust density is measured in its standard units of .
The DustER Algorithm
The fundamental insight/assumption of the DustER algorithm is that, at fixed redshift, all red-sequence galaxies follow a tight color–magnitude relation with small intrinsic scatter. Since dust extinction is stronger in bluer bands, dust extinction reddens the colors of galaxies.
If one is able to calibrate the average color of red-sequence galaxies as a function of redshift, then one can use coherent fluctuations in the color of these galaxies to map out fluctuations in the dust density of the Milky Way. Consequently, the first step towards mapping dust fluctuations is selecting samples of red-sequence galaxies of known redshift and colors.
Calibration of the Color–Redshift Relation
We calibrate the color–redshift relation of red-sequence galaxies using the redMaPPer cluster finding algorithm [21]. redMaPPer runs in two stages. The first is a calibration stage in which the color–redshift relation of red galaxies is empirically self-calibrated. The second consists of a cluster finding stage in which the calibrated model is used to identify galaxy clusters and measure their richness. Our results rely exclusively on the output of the first stage, which we briefly describe below.
redMaPPer begins by identifying bright spectroscopic galaxies, which we refer to as “seeds”. It then identifies the seed galaxies that reside within overdensities of similarly colored galaxies. The color of the seed galaxy is then used in redMaPPer to identify likely cluster members around the seed. Finally, redMaPPer simultaneously fits all high-likelihood cluster members around seeds to derive the redshift-dependent color model of red-sequence galaxies as a function of redshift. This model describes the red sequence at each redshift through its intercept, its slope (the tilt of the red sequence in the color–magnitude diagram), and its intrinsic scatter [21]. With this calibrated color model in hand, the procedure is iterated three times, resulting in the final color–redshift relation used to select red galaxies as described in § 4.2.
We emphasize that we calibrate our red-sequence model using raw (i.e. not dereddened) galaxy magnitudes. This is done to ensure the DustER dust maps are not informed by the reddening values from existing dust maps. While this choice implies our recovered dust maps are biased at first, we demonstrate below that we can remove this bias as part of our algorithm.
Red Galaxy Selection
Our algorithm to select red galaxies is a simplified version of the redMaGiC algorithm [14]. We first fit every galaxy with the redMaPPer red-sequence model to determine a photometric redshift estimate for each galaxy. This is done by minimizing
where is the color vector of a galaxy, is the redMaPPer color model, is the covariance characterizing the intrinsic width of the red sequence, and is the covariance matrix describing the photometric errors in the galaxy colors. After assigning each galaxy a photometric redshift , we select galaxies with and , where
The value is set to at to match redMaPPer. The evolution of is computed using the Bruzual & Charlot [22] stellar population synthesis code as implemented in the EzGal Python package.3 Note that as with the red-sequence calibration, our selection of red galaxies is performed using raw (i.e. not dereddened) photometry.
The Raw DustER Map
We now use the spatial modulation of color residuals of red-sequence galaxies to construct what we refer to as the “raw” DustER map. We can choose to arbitrarily normalize the dust density so that it has a mean of one. It follows then that where denotes the average across all pixels in the survey. This normalization is useful for decoupling the physical dust density from the dust law used to relate dust column density to dust attenuation.
With this reparameterization in hand, the change in a galaxy magnitude is given by
where is the band-dependent extinction coefficient introduced in § 3.2. Consequently, the change in the galaxy color due to dust is
Using our selection of red galaxies, we model the dust density over a sky pixel as constant, and we measure the color of every galaxy in the pixel. To ensure that the dust estimate per pixel is statistically reliable, we require at least 50 red-sequence galaxies within each pixel. This constraint sets the angular resolution of the map, corresponding to a HEALPix resolution of (pixel area ), which balances spatial resolution against per-pixel noise. Pixels that do not meet this threshold are excluded from the analysis. We model the observed color of red-sequence galaxies in sky pixels as having Gaussian random noise of variance where is the sum in quadrature of the observed error in the color and the intrinsic spread of the red-sequence. The color expectation value is
To compute the predicted color , we evaluate the predicted color for a red-sequence galaxy at the photometric redshift of the galaxy, as calibrated using redMaPPer. Note that since the red-sequence calibration and photometric redshifts are themselves impacted by reddening, the recovered dust maps are necessarily biased.
With these assumptions, the log-likelihood of our model is
where the sum runs over colors , , and . For simplicity, we neglect the covariance between colors induced by shared photometric bands. Defining
and
we find that the maximum likelihood point is
and . We refer to as the raw DustER map. The left panel in Figure 2 shows the dust density from the raw DustER map against the true dust density in the simulation. As expected, the raw dust map is both noisy and biased, failings which we will address in the next two sections.

Figure 2. Left: Raw DustER map as a function of the true dust density in simulations (in our case, SFD). The raw DustER map is both noisy and biased. Because the red-sequence calibration is performed on raw photometry over the full survey footprint, which is not free of dust, the mean reddening of the footprint is absorbed into the color model. As a result, is centered near zero and takes both positive and negative values. This offset is removed in the debiasing step (§ 4.5), where the dust density is required to be non-negative through the one-point prior of Figure 3. The red curve is a fit to the expectation value of the raw map as a function of the input dust density as per equation (23). The black dashed line corresponds to the line. Right: Filtered DustER map as a function of the true dust density in simulations. Wiener filtering of the raw map reduces the noise of the recovered map, but the large systematic biases in the dust density remain.
Wiener Filtering
Before we address the bias in the raw DustER map, we will address another undesirable property of the maps: they are very noisy. Fortunately, the noise is not likely to be correlated from pixel to pixel: in the ideal case, it would be driven purely by the random realizations of the color draws in the galaxies from different pixels. For this reason, we expect the power-spectrum of the raw DustER map to exhibit white noise, which we can suppress using standard de-noising techniques.
The power spectrum of the raw dust map is the sum of a signal contribution and a constant white noise term . The Wiener filtered map is defined in terms of its spherical harmonic coefficients via
where is the raw DustER map. To apply the Wiener filter we must estimate and . After inspecting the power spectrum of the raw map, we model the power spectrum as the sum of a power-law and a constant noise term , so that
where , , and are model parameters. To find these parameters, we compute from the mean-subtracted raw map and fit equation (22) to individual multipoles with by unweighted least squares in logarithmic space, i.e., by minimizing the cost function
Fitting in logarithmic space prevents the fit from being dominated by the low- multipoles, where the power is largest. We use the best-fit values of , , and directly, ignoring their statistical uncertainties. With our fits in hand, we apply the Wiener filter in harmonic space, and then transform it back to real space to arrive at our de-noised dust map. We will refer to the map as the filtered map. The right panel in Figure 2 shows the filtered DustER map in our simulation as a function of the input dust density. We can see the filtered map is significantly less noisy than the raw dust map (left panel), but the bias remains.
Map Debiasing
We model the bias in the filtered map as a power-law plus a constant,
where the parameters , , and characterize the systematic bias offset between the filtered map and the input map. We will further assume that the probability distribution is Gaussian, with an unknown variance :
The scatter is itself treated as a model parameter.
Our challenge is to recover the parameters , , , and that characterize the true dust map and the filtered map without utilizing pixel-level information from any external dust map for calibration. To do so, we rely on the statistical properties of the dust maps. Specifically, we characterize the one-point function of a dust map using the SFD and P16-GNILC dust maps. We find that after rescaling these maps to enforce a unit mean, the resulting one-point distributions are nearly identical (Figure 3), and can be accurately described via
Figure 3. A comparison of the one-point function of the SFD [1 blue] and P16-GNILC [3 orange] dust maps, and the best-fit model to SFD (dashed), adopted as the prior in DustER.
where
and , , and are unknown parameters. These parameters are constrained by the requirement that , which allows us to relate to and . We find the final result is best parameterized in terms of and , with
The equations above define and in terms of and , allowing us to fully specify the probability distribution in terms of these two variables. We model the values of in the SFD and P16-GNILC dust maps as independent random draws from the above distribution, and recover nearly identical values for the parameters and . Throughout this work, we adopt the best-fit parameters from SFD. DustER assumes the one-point function is known, and we ignore the uncertainty in the recovered parameters and . Note that while the statistical properties of the SFD dust map do inform the statistical properties of the DustER map, the dust density recovered by DustER at any given location in the sky is independent of the dust density reported by SFD or P16-GNILC at that same location.
With a known probability distribution in hand, we are now in a position to predict the probability distribution function of the filtered map . Specifically, we have that
The corresponding likelihood is
We sample this likelihood function to arrive at the posterior distribution for , , , and . We then compute the mean and variance of the filtered map given , marginalizing over the parameters , , , and . The first and second order moment of are given by
and
where is the point in the chain , and is the number of points in the chain. The variance is . With the assumption of Gaussianity, equations (29) and (30) can be used to model the probability distribution
In practice, we are interested in the converse probability distribution, . Using Bayes theorem, we can evaluate the expectation value of given , which is given by
We calculate this expectation value on a pixel by pixel basis to arrive at the final DustER map. Likewise, the second moment is given by
which we can use to determine the error in the DustER map on a pixel by pixel basis via
We emphasize that these are pixel-by-pixel marginalized uncertainties, and that in its present state, DustER is unable to generate correlated posteriors that can fully characterize the spatial structure of the statistical uncertainties. We postpone the development of these capabilities to future work.
To follow standard conventions, we can now convert our final dust map into the extinction field by multiplying by mmag, evaluated using some reference map (in our case, SFD) as averaged over the DustER footprint. The map uncertainties are similarly scaled by this factor.
Simulation Tests
Statistical Performance
We apply the DustER algorithm described above to our simulated data set. Figure 4 compares the input SFD dust map used to redden all the galaxies in our simulation (top panel) with the DustER map obtained using equation (32) (bottom panel). Visually, the two maps are in excellent agreement. This agreement is better quantified in Figure 5. The left panel plots the DustER density in the -axis against the input (SFD) dust map in the -axis. We see that the DustER values scatter around the line (black dashed line), demonstrating that the DustER map is an unbiased estimate of the true map.

Figure 4. Comparison between the input SFD dust map used to redden the simulated galaxies in Cardinal (top panel), to the dust map constructed using the DustER algorithm (bottom panel). Both maps are in units of mmag. The color scale of both maps is the same.
Figure 5. Left: DustER density as a function of the input (SFD) dust density in the Cardinal simulation. The black dashed lines corresponds to the line. Right: A histogram of the difference between the DustER output and the input dust map (SFD), in units of the estimated uncertainty in DustER. The two histograms are for pixels with mmag (blue) and mmag (orange). The red solid curve is a Gaussian of zero mean and unit variance.
The scatter in the DustER map about the true map is shown in the right panel of Figure 5. A key feature of the DustER maps is that they come with statistical uncertainties. In the panel, we plot histograms of the difference between the DustER map and the input SFD map in units of the reported error . We show two histograms: one for mmag (blue), and one for mmag (orange). Also shown for reference is a Gaussian distribution of zero mean and unit variance. The reason we show two histograms split by dust density is because the condition that forces the errors in the DustER map to be non-Gaussian in low extinction regions, leading to the skewed distribution of residuals for mmag. (i.e., which is below average). In regions with a high dust density, the errors become approximately Gaussian.
Figure 6 shows the estimated uncertainty as a function of position on the sky. We see the error map is highly correlated with the dust map itself. This is better illustrated in the inset in the top-left of the plot, which shows the error in the DustER map as a function of the DustER dust density. Note the inset is actually a scatter plot: we do not impose a smooth relation between and its error. The sharp drop in statistical uncertainty as one

Figure 6. The uncertainty in the recovered dust density as estimated using the DustER algorithm. The noise map is strongly correlated with the input dust map. The inset in the upper left shows the error in the recovered dust density as a function of the recovered dust density.
| Input (reddening) map | Compared to | Mean bias (mmag) | RMS (mmag) | |
| SFD | SFD (truth) | 0.5 | 6.3 | 0.94 |
| P16-GNILC | P16-GNILC (truth) | 0.5 | 7.1 | 0.94 |
| P16-GNILC | SFD (prior only) | 0.5 | 8.3 | 0.89 |
caption: Table 2. Global performance of the DustER map relative to a reference map over the simulation footprint: the mean bias , the rms difference, and the Pearson correlation coefficient , computed over all valid pixels in each run. The input map is the map used to redden the simulated catalog, and the reference map is listed under “Compared to”. In all runs, the one-point prior and the normalization are derived from SFD. The last row compares the P16-GNILC-reddened map to SFD, the source of the one-point function prior.
:::
approaches is due to the physical condition that .
Table 2 summarizes the global performance of DustER in our fiducial simulation. The DustER map recovers the input SFD map with a mean bias of 0.5 mmag, an rms scatter of 6.3 mmag, and a Pearson correlation coefficient .
To characterize the agreement as a function of angular scale, we compute the cross-correlation coefficient
where and are the auto-power spectra of the DustER map and the true (input) dust map, respectively, and is their cross-power spectrum. All spectra are pseudo- estimates computed with NaMaster4[23]. Figure 7 shows that the DustER map is nearly perfectly correlated with the input map on large scales. The correlation decreases on smaller scales as the reconstruction becomes noise-dominated.
Figure 7. Cross-correlation coefficient between the DustER map and the input dust map used to redden the simulated catalog, either SFD (blue) or P16-GNILC (orange). Both runs use the same one-point prior for the one point function of . The recovered maps are nearly perfectly correlated with the input on large scales, and the correlation decreases on smaller scales, where noise dominates the reconstruction. The two curves are nearly indistinguishable, demonstrating that the mismatch between the SFD one-point function prior and the P16-GNILC one-point function has little impact on the quality of the reconstructed dust map. Values slightly above unity at low reflect the noise of the binned pseudo- estimates.
A natural concern is that DustER might simply reproduce its prior: the one-point function and the normalization are both derived from SFD, the same map used to redden the fiducial catalog. To test this, we apply DustER to the P16-GNILC-reddened catalog described in § 3.2, keeping the SFD-based prior and normalization unchanged. The results are shown in the second and third rows of Table 2. If we compare the output DustER map to the true map in the simulation, namely the P16-GNILC map, we obtain a mean bias of 0.5 mmag, an rms scatter of 7.1 mmag, and , comparable to the fiducial run. By contrast, the comparison of the DustER map to SFD is much noisier: rms of 8.3 mmag, . Evidently, even though DustER used an SFD-based prior, the DustER output is a much better tracer of the input P16-GNILC map. Critically, if we compare the correlation coefficient as a function of scale between the input map and DustER, the resulting correlation coefficient is nearly identical regardless of which dust map is used as the input to the simulations (see Figure 7). These results provide strong evidence that the choice of prior used in DustER has little impact on the recovered dust map.
Systematics Tests
The DustER map is obtained by using red-sequence galaxies as standard crayons. It is therefore natural to consider whether the large-scale distribution of the Universe contaminates the DustER map. We would like to address this question in a way that can be replicated with real data. To that end, we consider the cross-correlation between DustER and a sample of red-sequence galaxies selected as described in § 4.2. However, unlike in § 4.2, we will now select the galaxies after dereddening the raw photometry with the recovered DustER map. We further bin the red galaxies in three redshift bins given by , , . We will use these same redshift bins throughout the rest of this paper, so we will henceforth refer to them as bins 1, 2, and 3, respectively.
The left panel in Figure 8 shows the DustER–galaxy cross-correlation for each of the three redshift bins considered in this work. Statistical errors in the cross power spectrum are calculated following [24], and the corresponding are reported in Table 3. Bin 3 deviates from zero (, or ), while bins 1 and 2 are consistent with no cross-correlation. However, this alone is not sufficient evidence to unambiguously establish the presence of systematics in the DustER map, since the red-galaxies used for the test were selected using the DustER maps.
| -bin | DustER red galaxies | DustER true-magnitude cut |
| [0.20, 0.40) | 5.99 | 29.10 |
| [0.40, 0.55) | 13.92 | 26.08 |
| [0.55, 0.70) | 37.75 | 17.98 |
Table 3. values of the hypothesis that the dust–galaxy cross-power spectrum in Figure 8 is zero. The number of degrees of freedom is 12, so the range for is [4.31,23.64].
Figure 8. Left: The cross-power spectrum between the DustER map recovered from the simulations before (dashed line) and after (solid line) the LSS-subtraction using all three bins described in § 5.4 and a sample of red-sequence galaxies selected as described in § 4.2, with one exception: the galaxies are selected using magnitudes that were dereddened using the DustER map. The galaxies are further binned by redshift into three sub-samples, as labeled. Note the cross correlations of bins 1 and 3 have been shifted vertically to avoid overcrowding the plot. Right panel: As the left panel, but for a magnitude limited sample of galaxies selected using the true magnitudes in the simulations (see § 5.2).
To address this caveat, we cross-correlate the DustER map with a magnitude-limited sample comprising the brightest 5% of Cardinal galaxies in the band (), binned by spectroscopic redshift. The resulting cross correlations are shown in the right panel of Figure 8, and their are also reported in Table 3. The magnitude-limited sample shows excess correlation in bins 1 and 2 ( and , respectively).
These results establish the existence of correlations between the DustER map and the galaxy density fields. However, the physical origin of these correlations remains ambiguous: they could arise either from LSS contamination in the DustER map itself, or from systematic biases in red galaxy selection due to imperfect dust corrections, or both. We address this question in the following section.
The Impact of Dust on Red-Galaxy Selection
Before attempting to correct for the correlations detected in § 5.2, we must first establish their origin. Here we test whether red galaxy selection is systematically biased by dust or by survey depth variations. We consider seven different red galaxy samples, constructed as follows:
Sample A: We add photometric noise to Cardinal true magnitudes assuming a constant depth, and then select our usual red galaxy sample. By construction this sample contains no dust and no depth variation, so it is guaranteed to be uncorrelated with dust and serves as the control against which the remaining samples are compared.
Sample B: We add dust attenuation to Cardinal true magnitudes based on the SFD map at a resolution of . We then add photometric noise assuming constant depth, and deredden the resulting magnitudes using the same SFD map at . Finally, we use the resulting catalog to select red galaxies. Sample B is identical to Sample A except that dust has been added and then removed, and therefore
isolates whether dereddening magnitudes in the presence of photometric noise induces systematic density fluctuations in the galaxy density field.
Sample C: Sample C is exactly like Sample B, except that magnitude dereddening is done by first averaging the SFD dust map to a resolution of , and then using this low-resolution dust map to deredden galaxy magnitudes. The comparison between Samples B and C therefore isolates whether using a low resolution (but otherwise correct!) dust map can induce systematic density fluctuations in the galaxy density field.
Sample D: We add photometric noise to Cardinal true magnitudes assuming variable depth, and then select our usual red galaxy sample. Sample D is identical to Sample A except for the depth treatment, and therefore isolates the effect of survey depth variations alone, in the complete absence of dust.
Sample E: Sample E is exactly like Sample B, except that photometric noise is added assuming variable depth. Because Sample E differs from the control in both dust and depth, it measures the combined effect of the two rather than isolating either one.
Sample F: Sample F is exactly like Sample C, except that photometric noise is added assuming variable depth; equivalently, it is Sample E with the dereddening map averaged to .
Sample G: Sample G is like Samples E and F, except that the low resolution map used to deredden galaxies is the DustER map itself. Unlike the SFD map used for the other samples, the DustER map carries its own reconstruction errors, so Sample G tests the practical consequence of dereddening a catalog with a DustER map.
We note that adding photometric noise to the simulations requires that we specify a depth map. Here, we rely on a depth map constructed using raw (not de-reddened) photometry which we decorrelated from the SFD input dust map. The reason is that the raw survey depth is a property of the survey, and is independent of the dust map. However, we found that the DES Y3 depth maps allow for the dust map to contribute to the recovered depth. When one generates photometry using depth maps that are contaminated with dust, then the resulting red-galaxy selection is contaminated by dust (see Appendix § 7). By using depth maps that are uncorrelated with dust, we are able to test whether DustER inherently leads to red-galaxy samples that are contaminated by dust.
To test for dust contamination of the galaxy density field, we cross-correlate each of the seven red galaxy samples with the input SFD dust map. The values of the zero cross-correlation hypothesis are summarized in Table 4. All seven samples are consistent with zero cross-correlation at the level, demonstrating that at the DES survey depth, and for the range of galaxy magnitudes we consider, the spatial distribution of red galaxies is not impacted by dust, survey depth variations, or dereddening with the DustER map. Appendix § 7 demonstrates that if the depth map meant to characterize the raw-depth photometry of the survey is incorrectly contaminated by dust, this does result in contamination of the large-scale structure probed by the red-sequence galaxies.
| -bin | SFD A | SFD B | SFD C | SFD D | SFD E | SFD F | SFD G |
| [0.20, 0.40) | 6.82 | 10.32 | 9.29 | 5.82 | 10.89 | 9.97 | 6.10 |
| [0.40, 0.55) | 9.07 | 9.34 | 8.73 | 8.50 | 9.14 | 8.71 | 11.75 |
| [0.55, 0.70) | 18.37 | 21.23 | 21.48 | 18.05 | 22.77 | 22.66 | 23.69 |
Table 4. values computed from the cross-power spectra between the input SFD dust map and each of the red galaxy samples. The number of degrees of freedom is 12, so the range for is [4.31,23.64]. All seven samples are consistent with zero at the level.
We also test whether the LSS contamination of the DustER map propagates into the clustering of red galaxies selected after dereddening with it. Figure 9 compares the angular power spectrum of Sample G to that of Sample E, which is selected identically but dereddened with the input SFD map at . For the two spectra agree within in all three redshift bins, and at all multipoles the ratio lies well within the full-sky cosmic-variance uncertainty of . The LSS contamination of the DustER map therefore does not measurably bias the clustering of the dereddened red-galaxy sample.
Figure 9. Top: Angular power spectra of red galaxies selected after dereddening with the input SFD map at (Sample E, solid) and with the DustER map (Sample G, dashed), for each redshift bin. Bottom: Ratio of the two spectra. The gray band shows the full-sky cosmic-variance uncertainty on . Because both samples trace the same underlying density field, the uncertainty on the ratio itself is smaller than this band; the band indicates the size of any bias relative to the statistical error of a clustering measurement.
Having ruled out galaxy selection biases as the source of the correlations observed in § 5.2, we conclude that the DustER map itself is contaminated by LSS, consistent with the excess correlation found with the magnitude-limited sample. In the following section, we attempt to mitigate this contamination.
| -bin | LSS Removal w/ Bins 1, 2, and 3 | LSS Removal w/ Bin 3 Only | ||
| Red galaxies | Magnitude cut | Red galaxies | Magnitude cut | |
| 19.24 | 63.78 | 13.63 | 31.08 | |
| 35.92 | 75.91 | 14.95 | 28.98 | |
| 63.03 | 30.47 | 7.37 | 6.45 |
caption: Table 5. values of the hypothesis that the dust–galaxy cross-power spectrum is zero after large-scale structure removal using either all three redshift bins (left), or just redshift bin 3 (right, see § 5.4 for details). The number of degrees of freedom is 12, so the range for is . Using all three bins over-corrects the map, increasing the cross-correlation in every bin; removing bin 3 only reduces the cross-correlation in bin 3, but not in bins 1 and 2.
:::
Systematics Amelioration
Having established the existence of LSS contamination in the DustER map, we test whether we can remove this contamination using an LSS-subtraction method based on the work by Chiang [25]. Let denote the density fluctuations of the “dirty” (i.e. LSS-contaminated) dust map. We assume the LSS contamination takes the form of an additive systematic related to the density fluctuations of the galaxy sample used to create the DustER map. This leads us to write the “cleaned” map as a linear combination of the dirty map and the galaxy density fields of interest. Here, we consider the red-galaxy samples for each of the three redshift bins considered in the previous section. In spherical harmonic space, the relation between these various maps is
where is the galaxy density field for field , and are constants which we will choose to remove the LSS signal from the DustER map.
To find the appropriate values of the constant , we consider the cross correlation between the cleaned DustER map and each of the three galaxy density fields. Multiplying equation 5.2 by , and averaging over , and demanding the clean map exhibit no cross correlation with the galaxy density field (with ) we arrive at
where is the cross-power spectrum between the dirty DustER map and galaxy bin , and is the cross-power spectrum between galaxy bin and galaxy bin . The above system corresponds to three equations for each multipole , which can be inverted to obtain the coefficients . In practice, however, we cannot evaluate the cross-power spectra at every individual because of low signal-to-noise and the presence of masking. Instead, we compute pseudo- cross-power spectra using the NaMaster Python package [23], which returns the cross-power at a set of effective multipoles . We solve for at each , and obtain at arbitrary by linearly interpolating the recovered values as a function of .
The results in Table 5 and Figure 8 show that the cleaning does not remove the contamination. When all three redshift bins are used, the of the cross-correlation increases in every bin, for both the red-galaxy and the magnitude-limited samples, indicating that the map has been over-corrected. Removing the contribution of bin 3 alone largely avoids this over-correction and substantially reduces the cross-correlation in bin 3, but the residual contamination in bins 1 and 2 is comparable to or higher than that in the uncleaned map. That is, the cleaning procedure is not effective at debiasing the DustER map.
Summary and Discussion
We have developed DustER, a photometric implementation of the standard crayon method to reconstruct Galactic dust reddening maps from red-sequence galaxies. The algorithm uses an empirically calibrated color–redshift relation to predict intrinsic galaxy colors as a function of redshift, and interprets coherent spatial color residuals as fluctuations in the dust extinction field. Because the red-sequence calibration and photometric redshifts are themselves affected by reddening, the initial reconstruction is both biased and noisy. DustER mitigates these effects through Wiener filtering and a statistical debiasing procedure that leverages the empirically determined one-point distribution of Galactic dust. The final dust map is obtained as the posterior expectation value of the extinction field, with pixel-level uncertainties derived from the posterior variance.
We validated DustER using simulated data based on the Cardinal catalog, in which dust attenuation is applied using the SFD dust map in our fiducial simulation. After debiasing, the recovered DustER map is statistically unbiased relative to the input dust field, with a mean bias of 0.5 mmag and an rms scatter of 6.3 mmag. The reported uncertainties behave as expected: they are approximately Gaussian in the high-extinction regime and become non-Gaussian at low extinction due to the physical constraint . The recovered map is nearly perfectly correlated with the input map on large scales, with the correlation decreasing on small scales where the reconstruction is noise-dominated. Because the one-point prior and normalization used by DustER are derived from SFD, we also reddened the simulation with the P16-GNILC dust map while keeping the SFD-based prior. The performance is nearly unchanged, and the recovered map agrees better with P16-GNILC than with SFD, demonstrating that DustER tracks the true dust field rather than reproducing its prior.
A key result of our analysis is the detection of statistically significant cross-correlations between the reconstructed DustER map and galaxy density fields in our simulations. To identify their origin, we performed a series of controlled tests on simulated red-galaxy samples with and without dust, including samples selected from catalogs that were reddened and then dereddened, either with the input dust map or with the DustER map itself. In none of these samples do we detect a correlation between the red-galaxy density field and the input dust map: at the depth of DES, red-galaxy selection from dereddened photometry is not measurably affected by dust, by survey depth variations, or by dereddening with the DustER map. Consequently, the non-zero DustER–red galaxy cross-correlation indicates that the DustER map itself is contaminated by large-scale structure. This is corroborated by a magnitude-limited sample selected on true magnitudes, which is independent of dust and depth by construction, yet correlates with DustER at and in the first two redshift bins, respectively. Nevertheless, this contamination does not propagate measurably into the angular clustering of red galaxies selected after dereddening with DustER: their angular power spectrum agrees with that obtained using the true dust map well within full-sky cosmic variance.
We explored a linear LSS subtraction scheme and found that it does not remove the contamination: using all three redshift bins over-corrects the map, while using only the highest-redshift bin reduces the cross-correlation in that bin but not in the others. Thus, the proposed linear cleaning algorithm fails. In future work, we plan to develop a forward-modeling, field-level inference framework in which the dust and galaxy density fields are modeled jointly within a hierarchical Bayesian approach, so that their correlations can be explicitly parameterized and marginalized over rather than removed by template subtraction. We anticipate this approach will yield posterior samples of the dust map and its uncertainties, with the maps being free of large-scale-structure correlations.
We plan to apply DustER to real photometric survey data, including observations from the Dark Energy Survey and the Dark Energy Camera All Data Everywhere (DECADE) [26] program. These data will enable the construction of high-fidelity, purely photometric dust maps over large sky areas. The resulting reconstructions will provide an independent reddening-based perspective on Galactic extinction and will allow detailed comparisons with emission-based maps such as SFD and P16-GNILC.
AI Disclosure
The authors used Claude Opus 5.5 [27] to assist with editing the manuscript, checking internal consistency of the text, tables, and cross-references, and suggesting wording for sections. All scientific content, analysis, and results are the authors’ own, and the authors reviewed and take full responsibility for the final text.
Acknowledgments
IEB, ER, and SSB were funded in part by NSF Grant 2306667. ER was also supported by DOE grant DE-SC0009913. ESR’s work was performed in part under DOE Contract DE-AC02-76SF00515.
Dust Imprint in the Survey Depth Maps
We found the DES Y3 depth maps correlate with dust, even after correcting the depth maps so as to reflect the raw rather than dereddened photometry. Note this correlation must be spurious, and appears because of how the depth maps are built (see [28]). If we generate photometry using these maps in our simulations, we do find significant cross-correlation between red-sequence galaxies and the input dust map to the simulations, demonstrating LSS contamination by the dust map.
This effect is clearly visible in the variable-depth samples of § 5.3. Table 6 compares the of the SFD–red galaxy cross-power spectra obtained with the original and corrected depth maps. In particular, Sample D, which contains no dust attenuation at all, was correlated with SFD at the level when the original depth maps were used. This is only possible if the depth maps themselves are correlated with the dust.
| -bin | SFD D | SFD E | SFD F | SFD G |
| Original depth maps (dereddened photometry) | ||||
| [0.20, 0.40) | 17.09 | 30.94 | 28.98 | 16.20 |
| [0.40, 0.55) | 11.37 | 17.55 | 18.21 | 28.60 |
| [0.55, 0.70) | 25.71 | 31.04 | 29.74 | 34.26 |
| Corrected depth maps (observed photometry) | ||||
| [0.20, 0.40) | 5.82 | 10.89 | 9.97 | 6.10 |
| [0.40, 0.55) | 8.50 | 9.14 | 8.71 | 11.75 |
| [0.55, 0.70) | 18.05 | 22.77 | 22.66 | 23.69 |
Table 6. of the cross-power spectra between the input SFD map and the variable-depth red galaxy samples of § 5.3, using the original (dust-contaminated) and corrected depth maps. The number of degrees of freedom is 12, so the range for is [4.31, 23.64].
For these reasons, we perform the template subtraction method from § 5.4 to remove dust contamination from the depth and effective-time-exposure maps. It is these decorrelated maps that we used for our tests in the main body of the text. With these corrected depth maps, all variable-depth samples are consistent with zero at the level (Table 6, bottom rows; see also Table 4).
References
- [1]David J. Schlegel, Douglas P. Finkbeiner, and Marc Davis. Maps of Dust Infrared Emission for Use in Estimation of Reddening and Cosmic Microwave Background Radiation Foregrounds. ApJ, 500(2):525–553, June 1998. doi: 10.1086/305772.DOI
- [2]Yi-Kuan Chiang and Brice Ménard. Extragalactic Imprints in Galactic Dust Maps. ApJ, 870 (2):120, January 2019. doi: 10.3847/1538-4357/aaf4f6.DOI
- [3]Planck Collaboration, N. Aghanim, M. Ashdown, J. Aumont, C. Baccigalupi, M. Ballardini, A. J. Banday, R. B. Barreiro, N. Bartolo, S. Basak, K. Benabed, J.-P. Bernard, M. Bersanelli, P. Bielewicz, L. Bonavera, J. R. Bond, J. Borrill, F. R. Bouchet, F. Boulanger, C. Burigana, E. Calabrese, J.-F. Cardoso, J. Carron, H. C. Chiang, L. P. L. Colombo, B. Comis, F. Couchot, A. Coulais, B. P. Crill, A. Curto, F. Cuttaia, P. de Bernardis, G. de Zotti, J. Delabrouille, E. Di Valentino, C. Dickinson, J. M. Diego, O. Doré, M. Douspis, A. Ducout, X. Dupac, S. Dusini, F. Elsner, T. A. Enßlin, H. K. Eriksen, E. Falgarone, Y. Fantaye, F. Finelli, F. Forastieri, M. Frailis, A. A. Fraisse, E. Franceschi, A. Frolov, S. Galeotta, S. Galli, K. Ganga, R. T. Génova-Santos, M. Gerbino, T. Ghosh, Y. Giraud-Héraud, J. González-Nuevo, K. M. Górski, A. Gruppuso, J. E. Gudmundsson, F. K. Hansen, G. Helou, S. Henrot-Versillé, D. Herranz, E. Hivon, Z. Huang, A. H. Jaffe, W. C. Jones, E. Keihänen, R. Keskitalo, K. Kiviniemi, T. S. Kisner, N. Krachmalnicoff, M. Kunz, H. Kurki-Suonio, J.-M. Lamarre, M. Langer, A. Lasenby, M. Lattanzi, C. R. Lawrence, M. Le Jeune, F. Levrier, P. B. Lilje, M. Lilley, V. Lindholm, M. López-Caniego, Y.-Z. Ma, J. F. Macías-Pérez, G. Maggio, D. Maino, N. Mandolesi, A. Mangilli, M. Maris, P. G. Martin, E. Martínez-González, S. Matarrese, N. Mauri, J. D. McEwen, A. Melchiorri, A. Mennella, M. Migliaccio, M.-A. Miville-Deschênes, D. Molinari, A. Moneti, L. Montier, G. Morgante, A. Moss, P. Natoli, C. A. Oxborrow, L. Pagano, D. Paoletti, G. Patanchon, O. Perdereau, L. Perotto, V. Pettorino, F. Piacentini, S. Plaszczynski, L. Polastri, G. Polenta, J.-L. Puget, J. P. Rachen, B. Racine, M. Reinecke, M. Remazeilles, A. Renzi, G. Rocha, C. Rosset, M. Rossetti, G. Roudier, J. A. Rubiño-Martín, B. Ruiz-Granados, L. Salvati, M. Sandri, M. Savelainen, D. Scott, C. Sirignano, G. Sirri, J. D. Soler, L. D. Spencer, A.-S. Suur-Uski, J. A. Tauber, D. Tavagnacco, M. Tenti, L. Toffolatti, M. Tomasi, M. Tristram, T. Trombetti, J. Valiviita, F. Van Tent, P. Vielva, F. Villa, N. Vittorio, B. D. Wandelt, I. K. Wehus, A. Zacchei, and A. Zonca. Planck intermediate results. XLVIII. Disentangling Galactic dust emission and cosmic infrared background anisotropies. A&A, 596:A109, December 2016. doi: 10.1051/0004-6361/201629022.
- [4]Daniel Lenz, Brandon S. Hensley, and Olivier Doré. A New, Large-scale Map of Interstellar Reddening Derived from H I Emission. ApJ, 846(1):38, September 2017. doi: 10.3847/1538-4357/aa84af.DOI
- [5]Aaron M. Meisner and Douglas P. Finkbeiner. A Full-sky, High-resolution Atlas of Galactic 12 μm Dust Emission with WISE. ApJ, 781(1):5, January 2014. doi: 10.1088/0004-637X/781/1/5.DOI
- [6]Michael Berry, Željko Ivezić, Branimir Sesar, Mario Jurić, Edward F. Schlafly, Jillian Bellovary, Douglas Finkbeiner, Dijana Vrbanec, Timothy C. Beers, Keira J. Brooks, Donald P. Schneider, Robert R. Gibson, Amy Kimball, Lynne Jones, Peter Yoachim, Simon Krughoff, Andrew J. Connolly, Sarah Loebman, Nicholas A. Bond, David Schlegel, Julianne Dalcanton, Brian Yanny, Steven R. Majewski, Gillian R. Knapp, James E. Gunn, J. Allyn Smith, Masataka Fukugita, Steve Kent, John Barentine, Jurek Krzesinski, and Dan Long. The Milky Way Tomography with Sloan Digital Sky Survey. IV. Dissecting Dust. ApJ, 757(2):166, October 2012. doi: 10.1088/0004-637X/757/2/166.DOI
- [7]E. F. Schlafly, G. Green, D. P. Finkbeiner, M. Jurić, H.-W. Rix, N. F. Martin, W. S. Burgett, K. C. Chambers, P. W. Draper, K. W. Hodapp, N. Kaiser, R.-P. Kudritzki, E. A. Magnier, N. Metcalfe, J. S. Morgan, P. A. Price, C. W. Stubbs, J. L. Tonry, R. J. Wainscoat, and C. Waters. A Map of Dust Reddening to 4.5 kpc from Pan-STARRS1. ApJ, 789(1):15, July 2014. doi: 10.1088/0004-637X/789/1/15.DOI
- [8]J. E. G. Peek and Genevieve J. Graves. A Correction to the Standard Galactic Reddening Map: Passive Galaxies as Standard Crayons. ApJ, 719(1):415–424, August 2010. doi: 10.1088/0004-637X/719/1/415.DOI
- [9]Gregory M. Green, Edward F. Schlafly, Douglas P. Finkbeiner, Hans-Walter Rix, Nicolas Martin, William Burgett, Peter W. Draper, Heather Flewelling, Klaus Hodapp, Nicholas Kaiser, Rolf Peter Kudritzki, Eugene Magnier, Nigel Metcalfe, Paul Price, John Tonry, and Richard Wainscoat. A Three-dimensional Map of Milky Way Dust. ApJ, 810(1):25, September 2015. doi: 10.1088/0004-637X/810/1/25.DOI
- [10]R. H. Leike and T. A. Enßlin. Charting nearby dust clouds using Gaia data only. A&A, 631: A32, November 2019. doi: 10.1051/0004-6361/201935093.DOI
- [11]Gregory M. Green, Edward Schlafly, Catherine Zucker, Joshua S. Speagle, and Douglas Finkbeiner. A 3D Dust Map Based on Gaia, Pan-STARRS 1, and 2MASS. ApJ, 887(1):93, December 2019. doi: 10.3847/1538-4357/ab5362.DOI
- [12]H.-L. Guo, B.-Q. Chen, H.-B. Yuan, Y. Huang, D.-Z. Liu, Y. Yang, X.-Y. Li, W.-X. Sun, and X.-W. Liu. Three-dimensional Distribution of the Interstellar Dust in the Milky Way. ApJ, 906 (1):47, January 2021. doi: 10.3847/1538-4357/abc68a.DOI
- [13]Matías Bravo, Eric Gawiser, Nelson D. Padilla, Joseph DeRose, Risa H. Wechsler, and LSST Dark Energy Science Collaboration. Simultaneous Estimation of Large-scale Structure and Milky Way Dust Extinction from Galaxy Surveys. ApJ, 921(2):108, November 2021. doi: 10.3847/1538-4357/ac1fea.DOI
- [14]E. Rozo, E. S. Rykoff, A. Abate, C. Bonnett, C. Davis, B. Hoyle, B. Leistedt, H. V. Peiris, R. H. Wechsler, T. Abbott, F. B. Abdalla, M. Banerji, A. H. Bauer, A. Benoit-Lévy, G. M. Bernstein, E. Bertin, D. Brooks, E. Buckley-Geer, D. L. Burke, D. Capozzi, A. Carnero Rosell, D. Carollo, M. Carrasco Kind, J. Carretero, F. J. Castander, M. J. Childress, C. E. Cunha, C. B. D’Andrea, T. Davis, D. L. DePoy, S. Desai, H. T. Diehl, J. P. Dietrich, P. Doel, T. F. Fieiler, A. E. Evrard, A. Fausti Neto, B. Flaugher, P. Fosalba, J. Frieman, E. Gaztanaga, D. W. Gerdes, K. Glazebrook, D. Gruen, R. A. Gruendl, K. Honscheid, D. J. James, M. Jarvis, A. G. Kim, K. Kuehn, N. Kuropatkin, O. Lahav, C. Lidman, M. Lima, M. A. G. Maia, M. March, P. Martini, P. Melchior, C. J. Miller, R. Miquel, J. J. Mohr, R. C. Nichol, B. Nord, C. R. O’Neill, R. Ogando, A. A. Plazas, A. K. Romer, A. Roodman, M. Sako, E. Sanchez, B. Santiago, M. Schubnell, I. Sevilla-Noarbe, R. C. Smith, M. Soares-Santos, F. Sobreira, E. Suchyta, M. E. C. Swanson, J. Thaler, D. Thomas, S. Uddin, V. Vikram, A. R. Walker, W. Wester, Y. Zhang, and L. N. da Costa. redMaGiC: selecting luminous red galaxies from the DES Science Verification data. MNRAS, 461(2):1431–1450, September 2016. doi: 10.1093/mnras/stw1281.DOI
- [15]James E. O’Donnell. R v-dependent Optical and Near-Ultraviolet Extinction. ApJ, 422:158, February 1994. doi: 10.1086/173713.
- [16]Chun-Hao To, Joseph DeRose, Risa H. Wechsler, Eli Rykoff, Hao-Yi Wu, Susmita Adhikari, Elisabeth Krause, Eduardo Rozo, and David H. Weinberg. Buzzard to Cardinal: Improved Mock Catalogs for Large Galaxy Surveys. ApJ, 961(1):59, January 2024. doi: 10.3847/1538-4357/ad0e61.DOI
- [17]Joseph DeRose, Risa H. Wechsler, Matthew R. Becker, Michael T. Busha, Eli S. Rykoff, Niall MacCrann, Brandon Erickson, August E. Evrard, Andrey Kravtsov, Daniel Gruen, Sahar Allam, Santiago Avila, Sarah Bridle, David Brooks, Elizabeth Buckley-Geer, Aurelio Carnero Rosell, Matias Carrasco Kind, Jorge Carretero, Francisco J. Castander, Ross Cawthon, Martin Crocce, Luiz N. da Costa, Christopher Davis, Juan De Vicente, Jörg P. Dietrich, Peter Doel, Alex Drlica-Wagner, Pablo Fosalba, Josh Frieman, Juan Garcia-Bellido, Gaston Gutierrez, Will G. Hartley, Devon L. Hollowood, Ben Hoyle, David J. James, Elisabeth Krause, Kyler Kuehn, Nikolay Kuropatkin, Marcos Lima, Marco A. G. Maia, Felipe Menanteau, Christopher J. Miller, Ramon Miquel, Ricardo L. C. Ogando, Andrés Plazas Malagón, A. Kathy Romer, Eusebio Sanchez, Rafe Schindler, Santiago Serrano, Ignacio Sevilla-Noarbe, Mathew Smith, Eric Suchyta, Molly E. C. Swanson, Gregory Tarle, and Vinu Vikram. The Buzzard Flock: Dark Energy Survey Synthetic Sky Catalogs. arXiv e-prints, art. arXiv:1901.02401, January 2019. doi: 10.48550/arXiv.1901.02401.DOI
- [18]Risa H. Wechsler, Joseph DeRose, Michael T. Busha, Matthew R. Becker, Eli Rykoff, and August Evrard. ADDGALS: Simulated Sky Catalogs for Wide Field Galaxy Surveys. ApJ, 931(2):145, June 2022. doi: 10.3847/1538-4357/ac5b0a.DOI
- [19]Robert H. Lupton, James E. Gunn, and Alexander S. Szalay. A Modified Magnitude System that Produces Well-Behaved Magnitudes, Colors, and Errors Even for Low Signal-to-Noise Ratio Measurements. AJ, 118(3):1406–1410, September 1999. doi: 10.1086/301004.DOI
- [20]T. M. C. Abbott, F. B. Abdalla, S. Allam, A. Amara, J. Annis, J. Asorey, S. Avila, O. Ballester, M. Banerji, W. Barkhouse, L. Baruah, M. Baumer, K. Bechtol, M. R. Becker, A. Benoit-Lévy, G. M. Bernstein, E. Bertin, J. Blazek, S. Bocquet, D. Brooks, D. Brout, E. Buckley-Geer, D. L. Burke, V. Busti, R. Campisano, L. Cardiel-Sas, A. Carnero Rosell, M. Carrasco Kind, J. Carretero, F. J. Castander, R. Cawthon, C. Chang, X. Chen, C. Coneslice, G. Costa, M. Crocce, C. E. Cunha, C. B. D’Andrea, L. N. da Costa, R. Das, G. Daues, T. M. Davis, C. Davis, J. De Vicente, D. L. DePoy, J. DeRose, S. Desai, H. T. Diehl, J. P. Dietrich, S. Doddelson, P. Doel, A. Drlica-Wagner, T. F. Fiedler, A. E. Elliott, A. E. Evrard, A. Farahi, A. Fausti Neto, E. Fernandez, D. A. Finley, B. Flaugher, R. J. Foley, P. Fosalba, D. N. Friedel, J. Frieman, J. García-Bellido, E. Gaztanaga, D. W. Gerdes, T. Giannantonio, M. S. S. Gill, K. Glazebrook, D. A. Goldstein, M. Gower, D. Gruen, R. A. Gruendl, J. Gschwend, R. R. Gupta, G. Gutierrez, S. Hamilton, W. G. Hartley, S. R. Hinton, J. M. Hislop, D. Hollowood, K. Honscheid, B. Hoyle, D. Huterer, B. Jain, D. J. James, T. Jeltema, M. W. G. Johnson, M. D. Johnson, T. Kacprzak, S. Kent, G. Khullar, M. Klein, A. Kovacs, A. M. G. Kozioł, E. Krause, A. Kremin, R. Kron, K. Kuehn, S. Kuhlmann, N. Kuropatkin, O. Lahav, J. Lasker, T. S. Li, R. T. Li, A. R. Liddle, M. Lima, H. Lin, P. López-Reyes, N. MacCrann, M. A. G. Maia, J. D. Maloney, M. Manera, M. March, J. Marriner, J. L. Marshall, P. Martini, T. McClintock, T. McKay, R. G. McMahon, P. Melchior, F. Menanteau, C. J. Miller, R. Miquel, J. J. Mohr, E. Morganson, J. Mould, E. Neilsen, R. C. Nichol, F. Nogueira, B. Nord, P. Nugent, L. Nunes, R. L. C. Ogando, L. Old, A. B. Pace, A. Palmese, F. Paz-Chinchón, H. V. Peiris, W. J. Percival, D. Petravick, A. A. Plazas, J. Poh, C. Pond, A. Porredon, A. Pujol, A. Refregier, K. Reil, P. M. Ricker, R. P. Rollins, A. K. Romer, A. Roodman, P. Rooney, A. J. Ross, E. S. Rykoff, M. Sako, M. L. Sanchez, E. Sanchez, B. Santiago, A. Saro, V. Scarpine, D. Scolnic, S. Serrano, I. Sevilla-Noarbe, E. Sheldon, N. Shipp, M. L. Silveira, M. Smith, R. C. Smith, J. A. Smith, M. Soares-Santos, F. Sobreira, J. Song, A. Stebbins, E. Suchyta, M. Sullivan, M. E. C. Swanson, G. Tarle, F. Thaler, D. Thomas, R. C. Thomas, M. A. Troxel, D. L. Tucker, V. Vikram, A. K. Vivas, A. R. Walker, R. H. Wechsler, J. Weller, W. Wester, R. C. Wolf, H. Wu, B. Yanny, A. Zenteno, Y. Zhang, J. Zuntz, DES Collaboration, S. Juneau, M. Fitzpatrick, and R. Nikutta. The Dark Energy Survey: Data Release 1. ApJS, 239(2):18, December 2018. doi: 10.3847/1538-4357/aae9f0.DOI
- [21]E. S. Rykoff, E. Rozo, M. T. Busha, C. E. Cunha, A. Finoguenov, A. Evrard, J. Hao, B. P. Koester, A. Leauthaud, B. Nord, M. Pierre, R. Reddick, T. Sadibekova, E. S. Sheldon, and R. H. Wechsler. redMaPPer. I. Algorithm and SDSS DR8 Catalog. ApJ, 785(2):104, April 2014. doi: 10.1088/0004-637X/785/2/104.DOI
- [22]G. Bruzual and S. Charlot. Stellar population synthesis at the resolution of 2003. MNRAS, 344(4):1000–1028, October 2003. doi: 10.1046/j.1365-8711.2003.06897.x.DOI
- [23]David Alonso, Javier Sanchez, Anže Slosar, and LSST Dark Energy Science Collaboration. A unified pseudo-Cℓ framework. MNRAS, 484(3):4127–4151, April 2019. doi: 10.1093/mnras/stz093.DOI
- [24]Elisabeth Krause and Tim Eifler. cosmolike - cosmological likelihood analyses for photometric galaxy surveys. MNRAS, 470(2):2100–2112, September 2017. doi: 10.1093/mnras/stx1261.DOI
- [25]Yi-Kuan Chiang. Corrected SFD: A More Accurate Galactic Dust Map with Minimal Extragalactic Contamination. ApJ, 958(2):118, December 2023. doi: 10.3847/1538-4357/acf4a1.DOI
- [26]D. Anbajagane, Z. Zhang, C. Chang, C. Y. Tan, M. Adamow, L. F. Secco, M. R. Becker, P. S. Ferguson, A. Drlica-Wagner, R. A. Gruendl, K. Herron, A. Tong, M. A. Troxel, D. Sanchez-Cid, I. Sevilla-Noarbe, N. Chicoine, R. Teixeira, A. Alarcon, D. Suson, A. N. Alsina, A. Amon, C. R. Bom, J. A. Carballo-Bello, W. Cerny, A. Choi, Y. Choi, C. Doux, K. Eckert, M. Gatti, D. Gruen, M. Jarvis, D. J. James, N. Kuropatkin, C. E. Martínez-Vázquez, P. Massana, S. Mau, J. McCullough, G. E. Medina, B. Mutlu-Pakdil, M. Navabi, N. E. D. Noël, A. B. Pace, J. Prat, M. Raveri, A. H. Riley, E. S. Rykoff, J. D. Sakowska, D. J. Sand, L. Santana-Silva, T. Shin, M. Soares-Santos, G. S. Stringfellow, A. K. Vivas, and M. Yamamoto. The DECADE cosmic shear project I: A new weak lensing shape catalog of 107 million galaxies. The Open Journal of Astrophysics, 8:46158, October 2025. doi: 10.3233/001c.146158.DOI
- [27]Anthropic. Claude opus 5.5. https://claude.ai, 2026. Large language model.
- [28]E. S. Rykoff, E. Rozo, and R. Keisler. Assessing Galaxy Limiting Magnitudes in Large Optical Surveys. arXiv e-prints, art. arXiv:1509.00870, September 2015. doi: 10.48550/arXiv.1509.00870.DOI





