[Paper Review] Pixel-z: Studying Substructure and Stellar Populations in Galaxies out to z~3 using Pixel Colors I. Systematics

N. Welikala, Andrew Hopkins+8 more|arXiv (Cornell University)|Dec 12, 2011

CCD and CMOS Imaging Sensors2 references3 citations

TL;DR

This paper evaluates systematic errors in the pixel- $z$ method for deriving spatially resolved stellar population properties (age, SFR, dust, metallicity) in high-redshift galaxies (z ~ 1–3) using multi-band Hubble imaging. It finds that model-dependent biases—especially between Maraston (2005) and Bruzual & Charlot (2003)/Charlot & Bruzual (2007) models—dominate systematic uncertainties when using optical-only colors, with differences exceeding 2σ in over 10% of pixels; near-IR data reduces these biases for z < 1 galaxies but has minimal impact for z > 1.

ABSTRACT

We perform a pixel-by-pixel analysis of 467 galaxies in the GOODS-VIMOS survey to study systematic effects in extracting properties of stellar populations (age, dust, metallicity and SFR) from pixel colors using the pixel-z method. The systematics studied include the effect of the input stellar population synthesis model, passband limitations and differences between individual SED fits to pixels and global SED-fitting to a galaxy's colors. We find that with optical-only colors, the systematic errors due to differences among the models are well constrained. The largest impact on the age and SFR e-folding time estimates in the pixels arises from differences between the Maraston models and the Bruzual&amp;Charlot models, when optical colors are used. This results in systematic differences larger than the 2σ uncertainties in over 10 percent of all pixels in the galaxy sample. The effect of restricting the available passbands is more severe. In 26 percent of pixels in the full sample, passband limitations result in systematic biases in the age estimates which are larger than the 2σ uncertainties. Systematic effects from model differences are reexamined using Near-IR colors for a subsample of 46 galaxies in the GOODS-NICMOS survey. For z &gt; 1, the observed optical/NIR colors span the rest frame UV-optical SED, and the use of different models does not significantly bias the estimates of the stellar population parameters compared to using optical-only colors. We then illustrate how pixel-z can be applied robustly to make detailed studies of substructure in high redshift galaxies such as (a) radial gradients of age, SFR, sSFR and dust and (b) the distribution of these properties within subcomponents such as spiral arms and clumps. Finally, we show preliminary results from the CANDELS survey illustrating how the new HST/WFC3 data can be exploited to probe substructure in z~1-3 galaxies.

Motivation & Objective

• To quantify systematic errors in pixel- $z$ when measuring spatially resolved stellar population properties in high-redshift galaxies (z ~ 1–3).
• To assess the impact of stellar population synthesis (SPS) model differences on age, SFR, dust, and metallicity estimates from pixel colors.
• To evaluate the effect of passband limitations (optical vs. optical+NIR) on systematic biases in pixel- $z$ measurements.
• To compare pixel-level SED fitting with global SED fitting to the same galaxy aperture and assess consistency.
• To demonstrate the robustness of pixel- $z$ for studying substructure such as spiral arms and clumps in z ~ 1–3 disk galaxies.

Proposed method

• Performs pixel-by-pixel SED fitting to 467 galaxies in the GOODS-VIMOS survey using multi-band photometry (bviz filters) to derive stellar population parameters per pixel.
• Compares results across three SPS models: Maraston (2005), Bruzual & Charlot (2003), and Charlot & Bruzual (2007) to quantify model-dependent systematic biases.
• Analyzes the impact of passband choice by comparing optical-only (bviz) results with those including near-IR (NICMOS) data for a subsample of 46 galaxies.
• Compares pixel- $z$ results (sum of individual pixel SED fits) with global SED fitting to the total flux within a fixed aperture to assess consistency.
• Uses radial and component-wise (e.g., clumps, spiral arms) analysis in example galaxies to test robustness of pixel- $z$ for substructure studies.
• Validates results against statistical consistency checks, including comparison of 2σ uncertainties and outlier detection in SFR estimates.

Experimental results

Research questions

• RQ1 How do differences between stellar population synthesis models affect the systematic uncertainties in pixel- $z$-derived stellar population parameters?
• RQ2 What is the impact of passband limitations (optical-only vs. optical+NIR) on systematic biases in age and SFR estimates from pixel- $z$?
• RQ3 How do pixel-level SED fits compare with global SED fitting to the same galaxy aperture in terms of SFR and other parameters?
• RQ4 Can pixel- $z$ robustly resolve radial gradients and component-specific properties (e.g., in clumps and spiral arms) in high-redshift disk galaxies?
• RQ5 How do systematic errors from model differences vary with redshift, particularly for z < 1 versus z > 1 galaxies?

Key findings

• Systematic uncertainties from SPS model differences exceed 2σ statistical uncertainties in over 10% of pixels when using only optical (bviz) passbands, with the largest discrepancies arising between Maraston (2005) and Bruzual & Charlot (2003)/Charlot & Bruzual (2007) models.
• Passband limitations cause systematic biases in age estimates larger than 2σ in 26% of pixels when using only optical colors, highlighting the importance of broad wavelength coverage.
• Adding near-IR passbands reduces model-dependent biases in age estimates for z < 1 galaxies, decreasing the fraction of pixels with <1σ differences from 0.89 (optical-only) to 0.81 (optical+NIR), but has minimal impact for z > 1 galaxies.
• For z > 1 galaxies, the rest-frame optical/NIR colors probe the rest-frame UV-optical SED, and model differences have less impact on parameter estimates than for z < 1.
• Pixel- $z$-derived SFRs generally agree with global SED fitting, but two outlier populations exist: galaxies with SFR < 1 M☉ yr⁻¹ show higher SFRs from pixel- $z$, while those with SFR ~10 M☉ yr⁻¹ show lower SFRs from pixel- $z$.
• Despite systematic uncertainties, pixel- $z$ is robust for measuring radial gradients and component-specific properties (e.g., in clumps and spiral arms), with model-dependent biases well-constrained in individual examples.