On the Accuracy of High Redshift Galaxy Property Estimation

Spectral energy distribution (SED) fitting is now a cornerstone of galaxy research, enabling the recovery of properties across a wide range of galaxy configurations and redshifts. With the advent of increasingly powerful facilities able to probe deeper into the Universe, and surveys producing data on millions of galaxies, the need for fast and reliable SED fitting software is growing. Many SED fitting programs are now available using different theoretical models to build SEDs against which observations can be compared, and using different sampling techniques to find the model that best fits the observations. Testing SED fitting is problematic, results for local galaxies can be compared to results derived by other means, but higher redshifts must be tested by comparing different fitters’ results or by testing against simulations where the property values are known in advance. We have taken four cosmological zoom-in simulations of dusty star-forming galaxies from the FIRE project and compared the results derived by the fitters MAGPHYS and PROSPECTOR with the true values; we have also investigated possible biases in the results caused by the size of the true values, by the galaxy inclination, by the choice of the star formation history (SFH) model used to create the templates, and by any offset between stellar emission and that from dust. This latter is of particular interest as many fitters rely on an energy balance between the energy absorbed by dust and that emitted by dust in the observer’s line-of-sight. Recent works have suggested that energy-balance SED fitting may be of limited use for studying high-redshift galaxies for which the observed ultraviolet and farinfrared emission are offset. It has been proposed that such offsets could lead energy-balance codes to miscalculate the overall energetics, preventing them from recovering such galaxies’ true properties. To investigate this, we test how well the SED fitting code MAGPHYS can recover the stellar mass, star formation rate (SFR), specific SFR, dust mass and luminosity by fitting ≈ 5900 synthetic SEDs. Comparing our panchromatic results (using wavelengths 0.4–500 μm, and spanning 1 < z < 8) with fits based on either the starlight (λeff ≤ 2.2μm) or dust (≥ 100μm) alone, we highlight the power of considering the full range of multi-wavelength data alongside an energy balance criterion. Overall, we obtain acceptable fits for 83 per cent of the synthetic SEDs, though the success rate falls rapidly beyond z ≈ 4, in part due to the sparser sampling of the priors at earlier times since SFHs must be physically plausible (i.e. shorter than the age of the Universe). We use the ground truth from the simulations to show that when the quality of fit is acceptable, the fidelity of MAGPHYS estimates is independent of the degree of stellar and dust emission offset, with performance very similar to that previously reported for local galaxies. We then investigate how the recovery of galaxy star formation rates depends on their recent star formation history. We use MAGPHYS and PROSPECTOR on the same simulations and identify a previously unknown systematic error in the MAGPHYS results due to bursty star formation: the derived SFRs can differ from the truth by as much as 1 dex, at large statistical significance (> 5σ), depending on the details of their recent SFH. SFRs inferred using PROSPECTOR with nonparametric SFHs do not exhibit this trend. We show that using parametric SFHs (pSFHs) causes SFR uncertainties to be underestimated by a factor of up to 5×. Although this undoubtedly contributes to the significance of the systematic, it cannot explain the largest biases in the SFRs of the starbursting galaxies, which could be caused by details of the stochastic prior sampling or the burst implementation in the MAGPHYS libraries. We advise against using pSFHs and urge careful consideration of starbursts when SED modelling galaxies where the SFR may have changed significantly over the last ∼100 Myr, such as recently quenched galaxies or those experiencing a burst. This concern is especially relevant, e.g., when fitting JWST observations of very high-redshift galaxies. Finally, we re-run the PROSPECTOR on the simulations using four different SFH models, finding that PROSPECTOR can achieve successful SED fits at higher redshifts than MAGPHYS, with an overall success rate in excess of 93 per cent. We find three main causes for unsuccessful fits: simulations with zero dust-mass at redshifts where this would be unlikely; galaxies with atypically low stellar to dust mass ratios; PROSPECTOR significantly underestimating the visual attenuation. Where a fit has been successful, the derived properties are of similar accuracy to those successfully fitted by MAGPHYS. We also find that neither the offset between the stellar and dust emission, nor the galaxy inclination appear to bias the result in most cases. Contrary to previous works, we find that the four SFH models - three nonparametric and one parametric - produce similar results when averaged across all of our ≈ 5900 simulations. However, we note that the uncertainties are likely underestimated, particularly for the parametric model, and that for individual galaxies, different SFH models can lead to significantly different property recoveries. We re-enforce this finding by analysing the differences between each simulation for each pair of SFH models, revealing mean biases of low statistical significance, but with significant outliers indicating differing property recovery for different models.