piXedfit: Resolved SED Fitting for Galaxies

• piXedfit is a Python package that performs pixelized SED fitting by processing multiband imaging and IFS data to extract spatially resolved physical properties of galaxies.
• Its modular architecture streamlines tasks from image preparation, PSF matching, and pixel/bin extraction to Bayesian SED fitting and star-formation history recovery.
• Applied to both local and high-redshift galaxies, piXedfit has been validated to yield consistent physical parameters and scalable processing for large galaxy samples.

piXedfit—pixelized spectral energy distribution fitting—is a Python package designed to extract spatially resolved physical properties of galaxies from multiband imaging data alone or in combination with integral-field spectroscopy (IFS) data. Introduced in "Introducing piXedfit -- a Spectral Energy Distribution Fitting Code Designed for Resolved Sources" (Abdurro'uf et al., 2021), it is a self-contained package with six modules that can handle all tasks in the spatially resolved SED fitting, including image preparation, spatial matching, pixel/bin extraction, Bayesian SED fitting, star-formation-history recovery, and parallel execution. The software has been applied to panchromatic analyses of nearby spiral galaxies on spatial scales of $\sim 1$–$2$ kpc out to at least $3$ effective radii and to JWST/HST studies of massive quiescent galaxies at $2Abdurro'uf et al., 2021, Prasal et al., 4 May 2026, Haryana et al., 26 Aug 2025).

1. Scope, data modalities, and scientific domain

piXedfit is designed for resolved sources rather than integrated galaxy photometry. Its stated purpose is to provide tools for analyzing spatially resolved properties of galaxies using multiband imaging data alone or in combination with IFS data, and it is optimized to handle large samples (hundreds to thousands of galaxies), automatically performing image preparation, PSF/astrometric matching, pixel/bin SED extraction, Bayesian SED fitting, star-formation-history recovery, and parallel execution (Abdurro'uf et al., 2021).

The package operates across heterogeneous imaging sets. The input can be science frames and variance (or weight) images in any combination of GALEX, SDSS, 2MASS, WISE, Spitzer, Herschel, HST, and related facilities, and later applications used combinations such as GALEX, SDSS, 2MASS, WISE, Spitzer/IRAC + MIPS, Herschel/PACS + SPIRE, HST/ACS, HST/WFC3, and JWST/NIRCam (Abdurro'uf et al., 2021, Abdurro'uf et al., 2022, Prasal et al., 4 May 2026, Haryana et al., 26 Aug 2025).

A common misconception is that piXedfit is exclusively a photometric pixel fitter. In the original software description, one module explicitly combines the multiband image cube with an IFS data cube and outputs a spectrophotometric cube in which each pixel contains a spectrum plus photometric fluxes, and the code can spatially match the imaging and IFS data (Abdurro'uf et al., 2021). This suggests that the package was conceived as a resolved SED framework spanning both purely imaging-based and spectrophotometric analyses.

2. Modular architecture

The package is organized into six core modules. In the original description, these modules were presented as covering the full resolved-SED workflow, from image handling to posterior analysis (Abdurro'uf et al., 2021).

Module	Role
piXedfit_images	data handling & image processing
piXedfit_spectrophotometric	spatial matching
piXedfit_bin	pixel/binning
piXedfit_model	model SED generation
piXedfit_fitting	Bayesian SED fitting + SFH recovery
piXedfit_analysis	visualization & output interpretation

piXedfit_images performs background estimation with Photutils’ Background2D with sigma-clipping, PSF matching using precomputed convolution kernels or analytical PSFs, projection and resampling with reproject_exact to a common WCS and pixel scale, segmentation with SExtractor, and outputs a multiextension FITS cube of fluxes and errors per band. piXedfit_spectrophotometric converts an IFS cube into a sequence of narrowband images, smooths them to match the image PSF, reprojects them to the imaging grid, applies a smooth multiplicative Legendre polynomial of order 3 for flux-offset correction, and outputs a spectrophotometric cube (Abdurro'uf et al., 2021).

piXedfit_model is built on FSPS + python-fsps, and later workflow summaries describe preprocessing through piXedfit.image process, fitting through piXedfit.fitting, and post-processing through piXedfit.analysis, with outputs including posterior PDFs for physical parameters, best-fit model SEDs, and reconstructed maps such as $\Sigma_*$, $\Sigma_{\rm SFR}$, and $\Sigma_{\rm dust}$ (Abdurro'uf et al., 2022). This suggests continuity between the original six-module design and later application-level interfaces.

3. Image homogenization, segmentation, and bin construction

A defining operational step in piXedfit is spatial homogenization prior to SED fitting. In the original implementation, all images are degraded to the lowest (largest) PSF, reprojected to a common WCS and pixel scale with flux conservation, and segmented into the target region (Abdurro'uf et al., 2021). In the nearby-galaxy panchromatic workflow, all images are convolved and reprojected (“PSF-matched”) to the resolution and pixel grid of the worst-resolution band, SPIRE 350 $\mu$m, with common-resolution PSF FWHM $\approx 24.9''$ and a common $2$0–$2$1 grid; in the JWST cosmic-noon application, all NIRCam broad- and medium-band mosaics and HST images are resampled and reprojected onto a common $2$2 grid using the F444W image as the reference PSF (Abdurro'uf et al., 2022, Prasal et al., 4 May 2026).

Segmentation is dataset-dependent. The general package description states that SExtractor on each band produces a merged segmentation map and ellipse cropping defines the target’s region (Abdurro'uf et al., 2021). In the nearby-galaxy study, a master segmentation map is generated with SExtractor to define the galaxy footprint, while in the cosmic-noon JWST study a three-filter (F115W+F150W+F200W) segmentation map is generated via SEP with minarea=40 pixels, threshold=2σ, deblend_nthresh=40, and deblend_cont=0.001, and foreground interlopers are manually deblended and masked where necessary (Abdurro'uf et al., 2022, Prasal et al., 4 May 2026).

The binning strategy is not a simple signal-to-noise coaddition. The original piXedfit_bin algorithm picks the brightest unbinned pixel in a reference band, grows a circular bin with minimum diameter $2$3 PSF FWHM until S/N thresholds in all bands are met, and accepts only neighboring pixels whose SEDs satisfy

$2$4

with

$2$5

The output is a FITS map of bin indices plus binned fluxes and errors (Abdurro'uf et al., 2021).

Later implementations preserve the same principle while adapting the operational thresholds. In the nearby-galaxy study, adjacent pixels are grouped until every band has S/N $2$6. In the JWST cosmic-noon study, the modified Voronoi algorithm of Cappellari & Copin (2003) is used as implemented in piXedfit_bin, requiring target_snr = 5 in all filters redward of rest 4000 Å, Dmin_bin = 8 pixels, del_r = 2 pixels, and redc_chi2_limit = 3, with pixels merged only if their SED shapes are similar, enforcing a reduced-$2$7; typical bin sizes span $2$8–$2$9 pixels, corresponding to $3$0–$3$1 kpc at $3$2 (Abdurro'uf et al., 2022, Prasal et al., 4 May 2026).

4. Physical modeling and Bayesian inference

The SED engine in piXedfit is based on FSPS via the python-fsps wrapper, with stellar population synthesis using Padova isochrones, the MILES stellar library, and a Chabrier (2003) IMF. Nebular emission is available via CLOUDY-based models, dust attenuation can be modeled with either the Calzetti (2000) law or the two-component Charlot & Fall (2000) prescription, dust emission is included through energy balance with Draine & Li (2007) templates, and AGN torus emission can be modeled with CLUMPY templates (Abdurro'uf et al., 2021, Abdurro'uf et al., 2022).

The package supports several parametric SFH forms. The original paper lists $3$3-model, delayed-$3$4, log-normal, Gaussian, and double-power-law, with the double-power-law written as

$3$5

In the cosmic-noon quiescent-galaxy application, each bin is modeled with a delayed-$3$6 star formation history,

$3$7

or equivalently in the piXedfit parameterization,

$3$8

with free parameters $3$9 and $2Abdurro'uf et al., 2021, Prasal et al., 4 May 2026).

The fitting formalism is Bayesian. In the original description,

$2

with flat unless user-supplied priors, and two likelihood options: Gaussian,

$2

and Student’s $2

$2

where

$2

Posterior sampling can use either MCMC with emcee or random densely-sampling of parameter space (RDSPS) (Abdurro'uf et al., 2021).

RDSPS is described in the later $2

$2

with Gaussian likelihood

$2

and parameter vector $2Haryana et al., 26 Aug 2025).

5. Derived quantities, maps, and validation

piXedfit outputs posterior chains or posterior PDFs for physical parameters, best-fit model SEDs, corner plots of 1D/2D posteriors, SED plots with residuals, and SFH plots. In resolved applications it reconstructs pixel maps by redistributing bin-level quantities back to the image grid. In the nearby-galaxy workflow, a bin’s $0SFR are redistributed by weighting each pixel’s observed flux in F444W for $0profiles in concentric annuli of width $0Abdurro'uf et al., 2022, Prasal et al., 4 May 2026, Haryana et al., 26 Aug 2025).

The original software paper validated the fitting engine using mock SEDs of simulated galaxies from IllustrisTNG and empirical data from CALIFA and MaNGA. For photometry-only mock SEDs, reported recovery metrics were: $0dex, $0$\Sigma_*$0; dust $\Sigma_*$1 with $\Sigma_*$2 dex, $\Sigma_*$3 dex, $\Sigma_*$4; mass-weighted age with $\Sigma_*$5 dex and $\Sigma_*$6 dex; $\Sigma_*$7 with $\Sigma_*$8 dex, $\Sigma_*$9 dex, $\Sigma_{\rm SFR}$0; and SFR with $\Sigma_{\rm SFR}$1 dex, $\Sigma_{\rm SFR}$2 dex, $\Sigma_{\rm SFR}$3. SFH lookback times corresponding to 30%, 50%, 70%, and 90% of final mass were recovered with $\Sigma_{\rm SFR}$4–$\Sigma_{\rm SFR}$5 dex and $\Sigma_{\rm SFR}$6, although short-timescale bursts are washed out (Abdurro'uf et al., 2021).

For real galaxies from CALIFA and MaNGA, photometry-only fits predicted the spectral continuum with median $\Sigma_{\rm SFR}$7 dex for CALIFA and $\Sigma_{\rm SFR}$8 dex for MaNGA over 3700–7500 Å. Reported prediction statistics were offset $\Sigma_{\rm SFR}$9 dex and scatter $\Sigma_{\rm dust}$0 dex for $\Sigma_{\rm dust}$1, offset $\Sigma_{\rm dust}$2 dex and scatter $\Sigma_{\rm dust}$3 dex for H$\Sigma_{\rm dust}$4, and offset $\Sigma_{\rm dust}$5 dex and scatter $\Sigma_{\rm dust}$6 dex for H$\Sigma_{\rm dust}$7. The SFR derived by piXedfit is consistent with that derived from H$\Sigma_{\rm dust}$8 emission, and RDSPS with Student’s $\Sigma_{\rm dust}$9 likelihood was reported as $\mu$0 faster than MCMC on the same cores while giving equally good fitting results (Abdurro'uf et al., 2021).

Additional external validation appears in the nearby-galaxy studies. Paper I compared piXedfit SFRs against H$\mu$1+$\mu$2m and UV+IR prescriptions, showing that the energy-balance SED fits reduce systematic biases in quiescent regions, and reported stellar masses that agree to $\mu$3 dex with independent estimates. Paper II noted that the tight dust–gas relation, with scatter $\mu$4 dex, and the agreement of rKS, rMGMS, and rSFMS slopes and scatters with SAMI, CALIFA, ALMaQUEST, and PHANGS provide further external validation (Abdurro'uf et al., 2022).

6. Applications to nearby and distant galaxies

In "Dissecting Nearby Galaxies with piXedfit: I. Spatially Resolved Properties of Stars, Dust, and Gas as Revealed by Panchromatic SED Fitting" (Abdurro'uf et al., 2021), piXedfit was used on ten nearby spiral galaxies with more than 20 photometric bands ranging from far-ultraviolet to far-infrared. The software performed point spread function matching of images, pixel binning, and modeled the stellar light, dust attenuation, dust emission, and emission from a dusty torus heated by an active galactic nucleus simultaneously through the energy balance approach. The study presented the spatially resolved version of the IRX–$\mu$5 relation, finding that it is consistent with the relationship from the integrated photometry, and showed that old stellar populations contribute to the dust heating, which causes an overestimation of star formation rate derived from the total ultraviolet and infrared luminosities on kpc scales. Using archival high-resolution maps of atomic and molecular gas, it also examined radial variations in stellar mass, age, metallicity, SFR, dust mass, dust temperature, abundance of polycyclic aromatic hydrocarbon, gas, dust-to-stellar mass ratio, and dust-to-gas mass ratio, and observed a depletion of molecular gas mass fraction in the central region of the majority of the galaxies (Abdurro'uf et al., 2021).

In "Dissecting Nearby Galaxies with piXedfit: II. Spatially Resolved Scaling Relations Among Stars, Dust, and Gas" (Abdurro'uf et al., 2022), the resolved maps derived with piXedfit were used to investigate $\mu$6–$\mu$7–$\mu$8 and dust scaling relations. While relations using all sub-galactic regions were reasonably tight, with $\mu$9 dex, most scaling relations exhibited galaxy-to-galaxy variations in normalization and shape. Two relations, $\approx 24.9''$0–$\approx 24.9''$1 and $\approx 24.9''$2–$\approx 24.9''$3, did not show noticeable galaxy-to-galaxy variations among the sample galaxies. The paper further reported significant correlations among the normalization of the $\approx 24.9''$4–$\approx 24.9''$5–$\approx 24.9''$6 relations, suggesting that galaxies with higher levels of resolved $\approx 24.9''$7 tend to have higher levels of resolved star formation efficiency and specific star formation rate, and concluded that both global and local factors contribute to governing the star formation process in galaxies (Abdurro'uf et al., 2022).

piXedfit has also been extended to high-redshift JWST analyses. In the cosmic-noon quiescent-galaxy study, spatially resolved SED modeling with piXedfit showed that $\approx 24.9''$8 of galaxies exhibit positive radial sSFR gradients, providing direct evidence for inside-out quenching, with the mean sSFR increasing by $\approx 24.9''$9 dex from $2$00 to $2$01; formation time $2$02 profiles indicated that inner regions formed $2$03 Gyr earlier, on average, than outer regions, and quenching timescale profiles showed that cores were quenched more rapidly than the outskirts (Prasal et al., 4 May 2026). In the broader $2$04 study of massive quiescent galaxies, spatially resolved SED fitting with piXedfit found that at $2$05, the half-mass radius is about 5.4 times smaller than at $2$06, with growth driven by stellar mass buildup in the outskirts while the central regions remain largely unchanged (Haryana et al., 26 Aug 2025).

7. Interpretive issues, limitations, and methodological implications

Several methodological cautions recur across the piXedfit literature. First, the package’s binning is constrained by both S/N and local SED shape rather than by S/N alone, and later JWST implementations explicitly imposed reduced-$2$07 thresholds during bin growth (Abdurro'uf et al., 2021, Prasal et al., 4 May 2026). This suggests that spatial resolution in the final maps is inseparable from assumptions about spectral similarity within each bin.

Second, resolved SFR inference can be sensitive to the physical mechanism producing dust luminosity. In the nearby-galaxy panchromatic analysis, old stellar populations contribute to the dust heating, which causes an overestimation of SFR derived from the total ultraviolet and infrared luminosities on kpc scales (Abdurro'uf et al., 2021). A plausible implication is that the value of piXedfit’s energy-balance framework is not merely numerical convenience but the explicit joint modeling of stellar light, attenuation, and dust emission.

Third, the validation results delimit the time resolution of recoverable SFHs. The original paper reports that short-timescale bursts are washed out in SFH recovery, and that adding mock spectra tightens constraints on metallicity and SFH parameters relative to photometry-only fits (Abdurro'uf et al., 2021). This indicates that piXedfit can recover physically informative resolved histories, but that the precision of those histories depends on both the data modality and the adopted SFH parameterization.

Finally, the package is not confined to a single astrophysical regime. It has been used on nearby spiral galaxies with FUV–FIR coverage, on CALIFA and MaNGA galaxies with matched imaging and IFS, and on JWST/HST observations of massive quiescent galaxies at cosmic noon and beyond (Abdurro'uf et al., 2021, Abdurro'uf et al., 2022, Prasal et al., 4 May 2026, Haryana et al., 26 Aug 2025). This breadth does not eliminate dataset-specific choices—such as the reference PSF, the segmentation procedure, the attenuation law, or the SFH family—but it demonstrates that piXedfit functions as an end-to-end resolved-SED framework rather than a single fixed fitting recipe.