BayeSED-GALAXIES II. Bayesian full spectrum analysis of galaxies and application in the CSST wide-field slitless spectroscopy survey

Abstract

The China Space Station Telescope (CSST) will conduct wide-field multiband photometric imaging and slitless spectroscopic surveys, advancing cosmology and galaxy evolution studies. Achieving CSST's cosmological goals requires precise redshifts ($σ_{\rm NMAD}\lesssim 0.002-0.005$) from low-resolution ($R\sim200$) and potentially blended slitless spectra. We present BayeSED3, extended for Bayesian full-spectrum analysis, including nebular emission modeling (via \textsc{Cloudy}) and a Bayesian treatment of the model scaling factor, improving reliability over optimization methods for low SNR spectra. Validated on realistic mock data generated with the CESS emulator (median SNR=1.65, including instrumental and self-blending effects), our method achieves excellent redshift precision with three-band (GU+GV+GI) spectroscopy: $σ_{\rm NMAD}=0.0008$ ($\sim$80% success) for star-forming and $σ_{\rm NMAD}=0.0015$ ($\sim$50% success) for quiescent galaxies. Stellar mass ($σ_{\rm NMAD}\approx0.015$ dex for SF, $\approx0.016$ dex for quiescent) and SFR ($σ_{\rm NMAD}\approx0.05$ dex for SF, especially at SNR>1) are reliably recovered. Self-blending increases scatter by $\gtrsim30%$, but combining spectroscopy with CSST's seven-band photometry significantly improves accuracy, especially for quiescent galaxies and data-limited cases. Single-band spectroscopy plus photometry yields reasonable redshifts: GU+photometry is limited, GI+photometry gives >60% (SF) and >40% (quiescent) success at $σ_{\rm NMAD}\lesssim0.002$, GV+photometry gives >35% (SF) and $\sim$40% (quiescent) at similar precision. The Bayesian framework offers a powerful method for accurate galaxy characterization, enhancing CSST's scientific outcomes despite the challenges of slitless spectroscopy.

Figures (16)

• Figure 1: Distributions of DESI magnitudes (top row) and redshift (bottom left) for the test sample (orange) compared to the parent sample (blue), demonstrating the representativeness of our test sample. The bottom middle and right panels show SNR distributions (probability density and cumulative distribution) of CSST slitless spectroscopy in the GU, GV, and GI bands. Median SNR values are 0.44, 1.31, and 2.35 for GU, GV, and GI respectively, with median mean SNR across all bands of 1.65. The GI band shows significantly better SNR characteristics, with $<$10% of measurements below $SNR=1$, compared to $\sim$40% for GV and $\sim$90% for GU.
• Figure 2: Residual analysis and color space validation for galaxies in the DESI DR9 photometric catalog. The five panels show magnitude residuals (observed minus model-predicted) for DESI optical bands (depths: $g$=24.5, $r$=23.9, $z$=22.9; top row) and WISE infrared bands (depths: W1=20.4, W2=19.5; bottom row). Each panel displays rest-frame wavelength on the x-axis versus magnitude residuals on the y-axis, with the density of points shown on a logarithmic color scale. Red dashed lines indicate perfect agreement (zero residuals). The tight distribution around zero residuals demonstrates the accuracy of our SED fitting across all bands and wavelengths. The sixth panel shows a UMAP embedding of standardized multi-band magnitudes with observed galaxies (blue) and model predictions (orange) overplotted. The accuracy of a simple domain classifier acc$\approx 0.50$ indicate that the models and observations are well-mixed throughout the color space, demonstrating that our model library adequately spans the observational manifold. These results validate our model's ability to reproduce the observed photometry and generate reliable spectral templates for CSST simulations.
• Figure 3: Bayesian full spectrum analysis of CSST slitless spectra for a quiescent (left) and star-forming (right) galaxy. Top row: BayeSED3 results comparing -NNLM (Bayesian sampling) and +NNLM (optimization) approaches. For the quiescent galaxy (SNR=1.12, $z_{\rm true}=0.424$), -NNLM achieves accurate redshift ($z_{\rm median}=0.423$) while +NNLM yields a different redshift ($z_{\rm median}=0.411$). For the star-forming galaxy (SNR=1.86, $z_{\rm true}=0.438$), -NNLM perfectly recovers the redshift ($z_{\rm median}=0.438$) while +NNLM catastrophically fails ($z_{\rm median}=1.530$). Middle row: BAGPIPES achieves competitive quality but requires much longer runtime than BayeSED3. Bottom row: Corner plots comparing posterior distributions from BayeSED3 variations (CEH=1/-NNLM, CEH=0/-NNLM, CEH=1/+NNLM) and BAGPIPES for all six free parameters. Generally, the Bayesian sampling approach provides much more reliable posteriors for low-SNR slitless spectroscopy.
• Figure 4: Quality assessment of redshift estimation as a function of acceptance threshold ($z_{\rm mean}/\sigma_z$) for star-forming (top) and quiescent (bottom) galaxies, comparing NNLM optimization (dashed lines) and Bayesian sampling (solid lines) approaches for scaling determination. Left panels show quality metrics (completeness, purity, outlier fraction, success rate); right panels show precision metrics ($\sigma_{\rm NMAD}$, $|{\rm Bias}|$). The Bayesian sampling approach consistently outperforms NNLM optimization across all metrics for both galaxy types, with particularly dramatic improvements for quiescent galaxies. An optimal threshold of $z_{\rm mean}/\sigma_z > 20$ is adopted for subsequent analysis (see Section \ref{['sss:results_z_threshold']} for detailed discussion).
• Figure 5: Performance metrics for redshift estimation using CSST slitless spectroscopy for star-forming ($N=46609$; top) and quiescent ($N=53389$; bottom) galaxies. Left panels: SNR dependence; right panels: redshift dependence. Star-forming galaxies: Despite challenging low SNR, performance improves dramatically with increasing SNR. Overall: 92.87% completeness, 87.67% purity, 81.41% success rate, $\sigma_{\rm NMAD}=0.0007$. Performance is stable through $z\approx0.4$ but challenging at $z\approx0.4$-0.7. Quiescent galaxies: Clear SNR threshold at $SNR\sim2$ for reliable estimation. Overall: 76.38% completeness, 73.19% purity, 55.90% success rate, $\sigma_{\rm NMAD}=0.0011$. Performance degrades significantly at $z>0.5$. See Section \ref{['sss:results_z_performance']} for detailed discussion.