Spatially Resolved SED Fitting

• Spatially resolved SED fitting is a method that derives physical parameters from localized regions within galaxies using multi-band photometry and enforced PSF consistency.
• The approach uses adaptive binning, precise image matching, and Bayesian inference to recover maps of stellar mass, star formation rates, dust attenuation, and other key properties.
• This technique advances our understanding of galaxy evolution by revealing spatial variations in star formation and mass assembly, influencing debates on outshining and quenching.

Searching arXiv for recent and foundational papers on spatially resolved SED fitting to ground the article in published work. Spatially resolved SED fitting is the inference of physical properties from multiband spectral energy distributions constructed for pixels, adaptively binned regions, apertures, or annuli within a galaxy rather than from a single integrated SED. In published implementations it is used to recover maps of stellar mass surface density $\Sigma_*$, star formation rate surface density $\Sigma_{\rm SFR}$, specific star formation rate $\mathrm{sSFR} \equiv \mathrm{SFR}/M_*$, dust attenuation, dust mass, and related quantities from UV–NIR or UV–FIR imaging, sometimes combined with integral-field spectroscopy, after enforcing a common PSF and aperture definition across bands (Abdurro'uf et al., 2019). The method is now applied from nearby spirals to galaxies at $z\sim 3$–4, and it has become central to debates about outshining, inside-out growth and quenching, dust heating, and the relation between resolved and unresolved stellar-mass estimates (Song et al., 2023).

1. Concept, observables, and spatial scales

The basic unit of analysis in spatially resolved SED fitting is not fixed. Some studies fit individual pixels directly, as in pixel-to-pixel SED fitting of local and high-redshift massive disks (Abdurro'uf et al., 2019). Others fit adaptively binned regions designed to satisfy signal-to-noise and SED-similarity criteria, as in piXedfit and several JWST analyses (Abdurro'uf et al., 2021). Narrow-band survey work has also used annular apertures to generate low-resolution pseudo-spectra, effectively treating imaging as an IFU-like dataset (Rodríguez-Martín et al., 15 Sep 2025).

The directly inferred quantities depend on wavelength coverage and model assumptions, but the most common products are $\Sigma_*$, $\Sigma_{\rm SFR}$, sSFR, attenuation parameters such as $A_V$, and, in panchromatic energy-balance work, dust luminosity and dust mass. In the CEERS resolved–unresolved comparison, resolved stellar mass is defined as

$M_{\ast}^{\rm res} = \sum_i M_{\ast,i},$

with the comparison metric

$$\Delta \log M_{\ast} = \log M_{\ast}^{\rm res} - \log M_{\ast}^{\rm unres}.$$

That formulation makes explicit that a “resolved” mass estimate is not a different physical quantity, but a different inference strategy applied to spatially partitioned photometry (Song et al., 2023).

The physical scale is set by the worst matched PSF and the adopted binning. Local GALEX+SDSS studies of massive spirals operate at approximately $1$–$\Sigma_{\rm SFR}$0 kpc resolution (Abdurro'uf et al., 2017). Euclid simulations span effective physical resolutions of approximately $\Sigma_{\rm SFR}$1 kpc for GALEX+LSST+Euclid, approximately $\Sigma_{\rm SFR}$2 kpc for LSST+Euclid, and approximately $\Sigma_{\rm SFR}$3 kpc for Euclid-only at $\Sigma_{\rm SFR}$4 (Collaboration et al., 19 Mar 2025). In miniJPAS/J-PAS, annular SED fitting can remain usable to approximately $\Sigma_{\rm SFR}$5 half-light radii with median $\Sigma_{\rm SFR}$6 in the blue bands, extending well beyond typical IFU coverage (Rodríguez-Martín et al., 15 Sep 2025).

2. Image preparation, segmentation, and spatial binning

A defining requirement is photometric consistency across bands. Published pipelines therefore perform PSF homogenization, astrometric alignment, reprojection to a common grid, and matched-aperture photometry before any fitting. The specific target PSF varies by dataset: F444W in CEERS (Song et al., 2023), W2 in piXedfit imaging+IFS workflows (Abdurro'uf et al., 2021), GALEX NUV in local GALEX/SDSS analyses (Abdurro'uf et al., 2017), and F480M in the Abell 2744 JWST/HST ram-pressure-stripping study (Benotto et al., 7 Nov 2025).

Segmentation is likewise integral. CEERS uses per-galaxy cutouts defined as $\Sigma_{\rm SFR}$7 the segmentation map to preserve aperture consistency in the resolved–unresolved comparison, with only pixels satisfying $\Sigma_{\rm SFR}$8 in all six JWST filters retained and a minimum of $\Sigma_{\rm SFR}$9 selected pixels per galaxy imposed to avoid unusably large mass uncertainties (Song et al., 2023). In piXedfit, segmentation maps from SExtractor or merged multi-band masks define the galaxy region, and the package provides dedicated modules for background subtraction, PSF matching, spatial resampling, and binning (Abdurro'uf et al., 2021).

Spatial binning is not merely a computational convenience; it sets the trade-off between spatial fidelity and parameter stability. In the massive-disk analysis over the last $\mathrm{sSFR} \equiv \mathrm{SFR}/M_*$0 Gyr, neighboring pixels with similar SED shapes are adaptively binned to enforce $\mathrm{sSFR} \equiv \mathrm{SFR}/M_*$1 in all filters for each spatially resolved SED (Abdurro'uf et al., 2019). The local resolved-SFMS study similarly requires bins to be spatially contiguous, similar in SED shape, and to reach $\mathrm{sSFR} \equiv \mathrm{SFR}/M_*$2 in all seven bands (Abdurro'uf et al., 2017). In the Abell 2744 JWST/HST analysis, weighted Voronoi tessellations are constructed with target $\mathrm{sSFR} \equiv \mathrm{SFR}/M_*$3 in F200W and a minimum bin area of $\mathrm{sSFR} \equiv \mathrm{SFR}/M_*$4 pixels, explicitly to mitigate residual PSF mismatch (Benotto et al., 7 Nov 2025). Py2DJPAS instead defines elliptical annuli whose inner and outer semi-major axes are integer multiples of the worst-PSF FWHM, with local background subtraction becoming critical for faint apertures (Rodríguez-Martín et al., 15 Sep 2025).

These implementations support a general methodological point: spatially resolved SED fitting is as dependent on image-domain rigor as on the fitting engine itself. A plausible implication is that apparently small choices in PSF handling, masking, or bin geometry can propagate directly into color gradients and hence into inferred mass-to-light ratio gradients.

3. Forward models, priors, and inference strategies

Most resolved SED-fitting studies use stellar population synthesis libraries such as BC03, FSPS, or PEGASE, combined with an IMF choice and a parameterized SFH. Exponentially declining histories,

$\mathrm{sSFR} \equiv \mathrm{SFR}/M_*$5

are used in pixel-to-pixel BC03 analyses of massive disks (Abdurro'uf et al., 2019) and in the local resolved-SFMS study (Abdurro'uf et al., 2017). Delayed-$\mathrm{sSFR} \equiv \mathrm{SFR}/M_*$6 histories,

$\mathrm{sSFR} \equiv \mathrm{SFR}/M_*$7

appear in JWST quiescent-galaxy work and CEERS-like mass-fitting pipelines (Song et al., 2023). piXedfit also supports double power-law SFHs, and Lightning adopts a non-parametric piecewise-constant SFH in user-defined age bins (Abdurro'uf et al., 2021, Doore et al., 2023).

Dust treatment divides the literature into two broad branches. One branch fits UV–NIR photometry with attenuation laws such as Calzetti or Charlot & Fall, deriving $\mathrm{sSFR} \equiv \mathrm{SFR}/M_*$8 and $\mathrm{sSFR} \equiv \mathrm{SFR}/M_*$9 without explicitly modeling dust re-emission (Abdurro'uf et al., 2019). The other imposes energy balance, with absorbed stellar light tied to dust emission,

$z\sim 3$0

as in Magphys, CIGALE, piXedfit, and Lightning (Smith et al., 2018, Abdurro'uf et al., 2021, Doore et al., 2023). In such frameworks, dust luminosity, dust mass, and attenuation are solved self-consistently with the stellar component, and some pipelines further include AGN torus emission or X-ray components (Doore et al., 2023).

Inference engines are similarly diverse. Pixel-to-pixel BC03 fitting in massive disks uses Bayesian inference with a Student’s $z\sim 3$1 likelihood rather than a Gaussian, explicitly to down-weight but not exclude models with relatively large $z\sim 3$2 (Abdurro'uf et al., 2019). piXedfit supports both MCMC and Random Dense Sampling of Parameter Space (RDSPS), with the latter reported to give equally good fitting results as the MCMC while being much faster (Abdurro'uf et al., 2021). Lightning offers affine-invariant MCMC, adaptive Metropolis–Hastings MCMC, and MPFIT, extending resolved SED fitting from X-ray to submillimeter wavelengths (Doore et al., 2023).

Priors are now recognized as active ingredients rather than neutral bookkeeping. Euclid proof-of-concept tests show that classical flat priors bias age, metallicity, and dust attenuation estimates, motivating additional mass–age–$z\sim 3$3 priors; these significantly improve mass-weighted age and metallicity recovery, but may play an excessive role compared to the data when no UV data are available (Collaboration et al., 19 Mar 2025). This directly connects resolved fitting to a broader Bayesian issue: the posterior can be data-limited in some wavelength regimes even when the formal fit is numerically stable.

4. Resolution limits, outshining, and other methodological controversies

A central controversy has concerned whether unresolved photometry intrinsically underestimates stellar mass because of outshining. Pixel-by-pixel fitting in the XDF reported that unresolved masses can be systematically underestimated by factors of up to $z\sim 3$4, with the discrepancy becoming rapid above $z\sim 3$5 and unresolved masses at $z\sim 3$6 falling to one half to one fifth of the resolved value (Sorba et al., 2018). By contrast, CEERS HST+JWST analyses found no significant disparity between resolved and unresolved stellar mass estimates once the reddest fitted band probes rest-frame wavelengths beyond $z\sim 3$7 \AA, with a median $z\sim 3$8 dex and $z\sim 3$9-$\Sigma_*$0 uncertainty approximately $\Sigma_*$1 dex (Song et al., 2023). The later result does not deny outshining; it reframes the earlier conflict as largely a rest-frame NIR coverage problem. Without rest-frame NIR data, masses can be overestimated by up to approximately $\Sigma_*$2 dex because age and dust attenuation are both overestimated, inflating the mass-to-light ratio (Song et al., 2023).

A second controversy concerns the validity of energy balance on small spatial scales. Controlled Magphys tests on a simulated isolated disc galaxy show that energy balance becomes increasingly incorrect in sub-kpc pixels because dust heating is often dominated by stars outside the pixel (Smith et al., 2018). In that experiment, Magphys yields statistically acceptable fits to more than $\Sigma_*$3 per cent of the pixels within the $\Sigma_*$4-band effective radius but only $\Sigma_*$5 to $\Sigma_*$6 per cent of pixels within $\Sigma_*$7 kpc, and SFR/sSFR estimates exhibit large scatter below approximately $\Sigma_*$8 kpc (Smith et al., 2018). This is a physical, not merely numerical, limitation: within-pixel absorbed starlight is not guaranteed to equal within-pixel dust emission.

SFH flexibility is a third recurrent issue. For approximately $\Sigma_*$9 massive galaxies at $\Sigma_{\rm SFR}$0, integrated Prospector fits with flexible nonparametric SFHs agree well with global SFHs reconstructed by summing pixel-level $\Sigma_{\rm SFR}$1 models, with an average stellar-mass difference of approximately $\Sigma_{\rm SFR}$2 dex (Jain et al., 2023). Simpler integrated $\Sigma_{\rm SFR}$3 models, however, typically miss the oldest episode of star formation and underestimate stellar mass by approximately $\Sigma_{\rm SFR}$4 dex (Jain et al., 2023). This suggests that resolved fitting has not only revealed spatial structure, but also empirically motivated more flexible global SFH priors.

5. Scientific results enabled by resolved fitting

One of the most robust outcomes is the existence of a resolved star-formation main sequence. In $\Sigma_{\rm SFR}$5 local massive spirals, the mode-based relation between $\Sigma_{\rm SFR}$6 and $\Sigma_{\rm SFR}$7 is nearly linear over intermediate $\Sigma_{\rm SFR}$8, with $\Sigma_{\rm SFR}$9 and $A_V$0, while flattening appears above $A_V$1 (Abdurro'uf et al., 2017). In the broader nearby-spiral piXedfit sample, the ensemble resolved SFMS has $A_V$2, $A_V$3, and $A_V$4 dex, while the resolved Kennicutt–Schmidt and molecular-gas main-sequence relations are tighter, each with $A_V$5 dex (Abdurro'uf et al., 2022). These measurements consistently point to approximately constant sSFR across star-forming discs and systematic central suppression in high-$A_V$6 regions.

Resolved fitting has also clarified radial quenching. In massive disks at $A_V$7 and $A_V$8, high-redshift star-forming disks show approximately flat sSFR$A_V$9, whereas galaxies below the global SFMS already show stronger central suppression, and the local sample exhibits ubiquitous central sSFR suppression with comparatively constant outskirts (Abdurro'uf et al., 2019). An empirical radial SFH model using $M_{\ast}^{\rm res} = \sum_i M_{\ast,i},$0 as the quenching criterion implies quenching at $M_{\ast}^{\rm res} = \sum_i M_{\ast,i},$1 kpc within approximately $M_{\ast}^{\rm res} = \sum_i M_{\ast,i},$2 Myr from $M_{\ast}^{\rm res} = \sum_i M_{\ast,i},$3, while the outskirts require approximately $M_{\ast}^{\rm res} = \sum_i M_{\ast,i},$4 Gyr from $M_{\ast}^{\rm res} = \sum_i M_{\ast,i},$5 to cross the same threshold (Abdurro'uf et al., 2019). The same local GALEX/SDSS study found that barred galaxies have lower sSFR in a core region than non-barred galaxies, although the outside region remains similar and total sSFRs are comparable (Abdurro'uf et al., 2017).

Resolved fitting has further expanded from star-forming disks to quiescent systems. JWST/HST analyses of massive quiescent galaxies at $M_{\ast}^{\rm res} = \sum_i M_{\ast,i},$6 find that the half-mass radius grows from $M_{\ast}^{\rm res} = \sum_i M_{\ast,i},$7 kpc at $M_{\ast}^{\rm res} = \sum_i M_{\ast,i},$8 to $M_{\ast}^{\rm res} = \sum_i M_{\ast,i},$9 kpc at $$\Delta \log M_{\ast} = \log M_{\ast}^{\rm res} - \log M_{\ast}^{\rm unres}.$$0, a growth factor of approximately $$\Delta \log M_{\ast} = \log M_{\ast}^{\rm res} - \log M_{\ast}^{\rm unres}.$$1, while the central $$\Delta \log M_{\ast} = \log M_{\ast}^{\rm res} - \log M_{\ast}^{\rm unres}.$$2 kpc mass surface density remains roughly constant at $$\Delta \log M_{\ast} = \log M_{\ast}^{\rm res} - \log M_{\ast}^{\rm unres}.$$3–$$\Delta \log M_{\ast} = \log M_{\ast}^{\rm res} - \log M_{\ast}^{\rm unres}.$$4 (Haryana et al., 26 Aug 2025). The estimated SFRs are too low to explain the stellar-mass growth, implying an additional stellar-mass accumulation process such as mergers (Haryana et al., 26 Aug 2025).

Dust and gas applications are equally important. In nearby spirals, CIGALE-based maps show that both $$\Delta \log M_{\ast} = \log M_{\ast}^{\rm res} - \log M_{\ast}^{\rm unres}.$$5 and $$\Delta \log M_{\ast} = \log M_{\ast}^{\rm res} - \log M_{\ast}^{\rm unres}.$$6 trace molecular and total gas surface density better than atomic gas, and that the attenuation toward young stars, $$\Delta \log M_{\ast} = \log M_{\ast}^{\rm res} - \log M_{\ast}^{\rm unres}.$$7, is in good agreement with $$\Delta \log M_{\ast} = \log M_{\ast}^{\rm res} - \log M_{\ast}^{\rm unres}.$$8 (Paspaliaris et al., 1 Sep 2025). In the ten-galaxy piXedfit sample, the $$\Delta \log M_{\ast} = \log M_{\ast}^{\rm res} - \log M_{\ast}^{\rm unres}.$$9–$1$0 and $1$1–$1$2 relations are the only major resolved scaling relations without noticeable galaxy-to-galaxy variation, with ensemble scatters of $1$3 and $1$4 dex, respectively (Abdurro'uf et al., 2022). Such results show that resolved SED fitting is no longer limited to stellar-population mapping; it is now a tool for connecting stars, dust, gas, and attenuation geometry on kpc scales.

6. Software frameworks and survey-era directions

Several dedicated frameworks now structure the field. piXedfit is explicitly designed for resolved sources, with six modules handling image processing, imaging+IFS matching, binning, model generation, Bayesian fitting, and analysis; it supports both MCMC and RDSPS and was validated on IllustrisTNG mock SEDs and on CALIFA/MaNGA galaxies (Abdurro'uf et al., 2021). Lightning extends resolved fitting to X-ray–submillimeter data with stellar, dust, and AGN models and multiple sampling algorithms, and has been demonstrated on M81, NGC 4631, NGC 628, and AGN test cases (Doore et al., 2023). Magphys and CIGALE remain important energy-balance engines, especially where full UV–FIR coverage is available (Smith et al., 2018, Song et al., 2023).

Survey-era work is shifting the emphasis from proof of concept to scalability. Euclid simulations indicate that stellar-mass surface densities can be recovered well using GALEX+LSST+Euclid, LSST+Euclid, or Euclid-only data cubes, largely irrespective of SPS model and prior variations, while age and metallicity benefit strongly from ancillary UV coverage and informative priors (Collaboration et al., 19 Mar 2025). Py2DJPAS shows that narrow-band imaging can support annular SED fitting with residuals below $1$5 per cent and no significant wavelength-dependent bias for regions with $1$6, with resolved analyses extending to $1$7 HLR (Rodríguez-Martín et al., 15 Sep 2025).

The emerging direction is therefore not a single “best” resolved-fitting recipe, but a set of regime-dependent practices. Where rest-frame NIR coverage is available, resolved and unresolved stellar masses can agree closely (Song et al., 2023). Where FIR data exist and physical scales are $1$8 kpc, energy-balance approaches recover dust-related quantities robustly (Smith et al., 2018). Where UV is absent, priors can stabilize age and metallicity but may also dominate the inference (Collaboration et al., 19 Mar 2025). This suggests that the future of spatially resolved SED fitting will be defined by explicit control of wavelength coverage, resolution, priors, and cross-validation against spectroscopy rather than by any single code or parameterization.