Differential Reddening and Extinction Law Analyses of Galactic Open Clusters

This paper presents a homogeneous, member-based characterization of extinction properties in Galactic open clusters (OCs), including mean reddening, differential reddening, and color excess ratios (CERs). Using Gaia-era open cluster member samples combined with Gaia DR3 and 2MASS photometry and intrinsic colors estimated from SHBoost stellar parameters, the authors derive color excesses for member stars and provide cluster-scale extinction measurements under a uniform analysis framework.

The study adopts the membership catalog of 1481 OCs from T. Cantat-Gaudin & F. Anders (2020), applying an additional membership-probability threshold of 80% to reduce systematic biases. Optical photometry (GBP and GRP) is taken from Gaia DR3, while near-infrared KS photometry is from the 2MASS Point Source Catalog. Quality cuts include KS uncertainties 1.4 removed. Stellar parameters (Teff, log g, [M/H]) are obtained from the SHBoost catalog, with additional quality cuts on parameter uncertainties and a restriction to Teff between 4000 and 8000 K and log g between 1 and 5.

For each member star, color excesses E(GBP − GRP) and E(GBP − KS) are calculated as observed color minus intrinsic color, where intrinsic colors are predicted using an XGBoost model trained on over one million low reddening stars (H. Zhao et al. 2024), applied to SHBoost stellar parameters. Uncertainties in color excesses include propagated stellar parameter uncertainties, representative intrinsic color uncertainties (0.014 mag for (GBP − GRP)0 and 0.050 mag for (GBP − KS)0), and photometric uncertainties. This procedure identifies 1225 OCs with measurable reddening.

The mean reddening of each cluster is estimated using a maximum-likelihood model that accounts for individual color excess uncertainties and an additional dispersion term. Differential reddening is quantified using the median absolute deviation (MAD) of member star color excesses, with statistical uncertainties estimated via bootstrap resampling. The CER for each cluster is determined as the slope of a weighted linear fit of E(GBP − KS) against E(GBP − GRP), constrained to pass through the origin. The zero-intercept constraint is validated by finding a median fitted intercept of b = 0.026 mag, smaller than the typical uncertainty σb = 0.068 mag, with 78% of clusters having intercepts consistent with zero within 2σ.

Systematic uncertainties are evaluated by repeating the analysis with alternative stellar parameter sets (SHBoost vs. GSP-Phot) and by shifting Teff by ±100 K. The total systematic uncertainty combines these contributions in quadrature. Tests excluding Gaia-identified non-single star sources (169 stars, 0.36% of the sample) show negligible changes in reddening, differential reddening, and CER, with median signed differences consistent with zero and changes smaller than statistical uncertainties for more than 99% of clusters.

For reliable cluster statistics, only clusters with color excess measurements for more than ten member stars (Nused ≥ 10) are retained, leaving 729 OCs for the color excess analysis. Their mean E(GBP − GRP) and E(GBP − KS) values are 0.471 and 1.049 mag, respectively. Reddening ranges from ≲ 0.2 mag to ≳ 1.5–2 mag and is spatially non-uniform across the Galactic plane.

External comparisons show broad consistency with independent estimates. Compared with X. Wei et al. (2025), who derived cluster reddening from Gaia CMD isochrone fitting, the median difference is 0.062 mag with a normalized median absolute deviation scatter of 0.114 mag, and a linear fit gives a slope consistent with unity. Compared with the AV values of T. Cantat-Gaudin et al. (2020), the fitted slope is 2.073 ± 0.022, smaller than the expected value of 2.394 ± 0.018 for a standard RV = 3.1 extinction curve, which the authors interpret mainly as a systematic offset between methods rather than evidence for a strongly non-standard mean extinction law. Comparison with the three-dimensional dust map of X. Zhang & G. M. Green (2025) for clusters within 5 kpc shows positive correlation, with residual dispersion increasing toward larger E(GBP − GRP), attributed to differences in angular resolution and sampling scale.

Differential reddening increases systematically with mean reddening. A Huber robust linear fit gives MAD(EBP−RP) = (0.090 ± 0.021) × E(GBP − GRP) + (0.034 ± 0.011). Differential reddening also decreases with distance from the Galactic plane, approaching 0.047 mag at Z ≳ 200 pc. The youngest clusters tend to show larger differential reddening than older clusters, consistent with their association with dust-rich environments, though this trend is complicated by the preferential location of young clusters near the Galactic plane.

The impact of differential reddening on CMDs is assessed by measuring CMD-width before and after star-by-star dereddening. Among 435 clusters with reliable CMD-width measurements, 369 (85%) show a decrease in CMD-width after correction. Clusters with stronger differential reddening tend to exhibit larger CMD-width reductions. No single threshold in cluster angular size (r50) or age reliably predicts the amount of CMD-width reduction. Cluster color excess maps reveal small-scale extinction structures that are only weakly reflected in lower-resolution dust maps.

For the CER analysis, additional quality cuts require CER uncertainty 10, and R2 > 0.9, leaving 709 OCs. Restricting to b ≤ 10° leaves 600 OCs for spatial analysis. The CER distribution has a median value of 2.22, with 16th and 84th percentiles of 2.12 and 2.35. This is compatible with a fiducial diffuse ISM extinction law with RV = 3.1, which gives a median CER of 2.22 with a 16th–84th percentile interval of 2.15–2.27. However, the mean CER alone cannot uniquely determine RV or distinguish it from nearby extinction curves (RV = 2.9–3.2 produce overlapping CER ranges).

The CER exhibits clear large-scale spatial variations across the Galactic disk. Clusters in the first and second Galactic quadrants tend to exhibit lower CER values than those in the third and fourth quadrants. This quadrant dependence is broadly consistent with the RV distribution inferred from the dust map of X. Zhang & G. M. Green (2025). Superimposed on this large-scale trend, koc shows local fluctuations with longitude, with directions having larger reddening gradients tending to show smaller koc. When low reddening clusters are excluded (E(GBP − GRP)oc > 0.25 mag), the amplitude of the longitude-dependent variation becomes weaker. Because cluster CERs are integrated along the line of sight up to the cluster distances, these variations are best interpreted as reflecting differences in the dominant dust environments sampled along different Galactic directions, rather than local extinction properties at individual cluster positions.

The main results are summarized as follows: (1) For 729 OCs, cluster reddening spans a broad range across the Galactic plane, from nearly zero to above 2 mag in E(GBP − GRP), and is broadly consistent with independent extinction measurements, supporting the robustness of the reddening measurements while preserving cluster-scale reddening structure that is smoothed out in coarse dust maps. (2) Individual reddening estimates improve CMD-based analyses of clusters affected by strong differential reddening, with star-by-star corrections recovering narrower and more coherent dereddened CMD sequences, and cluster color excess maps revealing small-scale extinction structures. The increase of differential reddening with mean reddening indicates that sight lines with larger dust columns tend to show stronger small-scale extinction inhomogeneity. (3) For the quality-selected CER subsample of 600 OCs near the Galactic plane, CER exhibits clear large-scale spatial variations across the Galactic disk, with systematically lower values in the first and second Galactic quadrants than in the third and fourth quadrants, indicating that the optical-to-near-infrared extinction behavior is not uniform across the Galactic disk, with implications for studies of the extinction law and precision distance measurements.

Improvements to AI Systems:

1. Uncertainty-Aware Stellar Parameter Regression

• Improve AI models that predict intrinsic stellar colors by training on the SHBoost parameter set with propagated uncertainties (e.g., Teff, log g, [M/H]) and incorporating representative intrinsic color uncertainties (0.014 mag for optical, 0.050 mag for optical-NIR).

• The improved system can output color excess estimates with calibrated confidence intervals, reducing systematic biases in reddening measurements for individual stars.

2. Robust Clustering with Differential Reddening Correction

• Enhance AI-based open cluster identification and member assignment by integrating the 80% membership probability threshold and the MAD-based differential reddening metric.

• The improved system can automatically flag clusters with high internal extinction inhomogeneity and apply star-by-star dereddening, yielding cleaner color-magnitude diagrams and more accurate cluster parameters (age, distance, metallicity).

3. Spatially Adaptive Extinction Law Modeling

• Develop AI models that predict color excess ratios (CER) as a function of Galactic position (quadrant, longitude, latitude) using the observed CER variations (median 2.22, range 2.12–2.35).

• The improved system can generate direction-dependent extinction laws, replacing a single RV=3.1 assumption, thereby improving distance estimates to stars and clusters in regions with non-standard dust properties (e.g., first/second quadrants vs. third/fourth).

4. Multi-Wavelength Data Fusion with Quality Cuts

• Train AI systems to automatically apply the paper’s quality filters (e.g., RUWE < 1.4, fractional parallax uncertainty < 20%, KS uncertainty < 0.05 mag) when combining Gaia DR3 and 2MASS photometry.

• The improved system can produce higher-fidelity training datasets for extinction mapping, reducing contamination from binaries, poor astrometry, and photometric outliers.

5. Bootstrap-Based Uncertainty Quantification for Small Samples

• Incorporate bootstrap resampling methods (as used for differential reddening uncertainties) into AI pipelines that estimate extinction from sparse stellar samples (Nused ≥ 10).

• The improved system can provide robust error bars for reddening and CER even when only a few member stars are available, enabling reliable comparisons across heterogeneous datasets.

6. Systematic Error Propagation Across Parameter Sets

• Build AI models that explicitly account for systematic differences between stellar parameter catalogs (SHBoost vs. GSP-Phot) and Teff shifts (±100 K) by training on multiple input versions.

• The improved system can output reddening estimates with both statistical and systematic uncertainty components, making it more reliable for precision astrophysics (e.g., distance ladder calibration).

7. Cross-Validation with Independent Dust Maps

• Enhance AI-based 3D dust mapping by training on the paper’s cluster-scale reddening measurements as ground truth, while comparing against lower-resolution maps (e.g., Zhang & Green 2025).

• The improved system can learn to recover small-scale extinction structures (missed by coarse maps) and quantify resolution-dependent biases, leading to higher-resolution Galactic dust models.

8. Predictive Modeling of Differential Reddening

• Use the observed linear relationship (MAD = 0.090 × E(GBP−GRP) + 0.034) and the decrease with Z to train AI models that predict differential reddening from mean reddening and Galactic height.

• The improved system can pre-correct photometric data for expected internal extinction variations in clusters, improving automated isochrone fitting and star formation history recovery.

9. Automated CMD Width Reduction Assessment

• Implement AI that measures color-magnitude diagram width before/after dereddening (as done for 435 clusters) to automatically evaluate the effectiveness of extinction corrections.

• The improved system can self-optimize its dereddening algorithm by minimizing CMD width, leading to better cluster parameter estimation in dusty regions.

10. Longitude-Dependent Extinction Prior for Bayesian Distance Estimation

• Integrate the observed CER variations with Galactic longitude into Bayesian models for stellar/cluster distances.

• The improved system can assign direction-specific extinction priors, reducing distance biases in the first/second quadrants where CER is systematically lower, and improving kinematic and dynamical studies of the Milky Way.