2:30–3:00 pm
ERC 401 Probabilistic Inference of Low-Surface-Brightness Galaxy Morphological Parameters Using Simulation-Based Inference In this work, we present a simulation-based inference framework for estimating posterior distributions of low-surface-brightness galaxy (LSBG) morphological parameters from simulated galaxy images. LSBGs are diffuse systems whose faint surface brightnesses make their structural properties difficult to measure reliably in wide-field imaging surveys, particularly in the presence of sky-background uncertainty and contaminating background sources. We generate DES-like simulated images of single-Sérsic LSBGs using PyImfit, with known position angle, ellipticity, Sérsic index, effective surface brightness, and effective radius, and train a normalizing-flow-based neural posterior estimator to infer these parameters directly from image data. We characterize the performance of this amortized inference framework using posterior predictive checks, empirical coverage curves, and rank-based calibration diagnostics. For isolated simulated galaxies, we show that the SBI posterior recovers the true input parameters, produces residuals consistent with the assumed noise model, and exhibits well-calibrated uncertainty estimates across the test set. We compare these results with PyImfit-based MCMC inference and find broadly consistent posterior constraints, while SBI enables substantially faster posterior sampling after training. We also explore the effect of compact background contaminants on posterior reliability. We show that a model trained only on isolated galaxies produces overconfident, undercovered posteriors when applied to contaminated images. In contrast, models trained on simulations with variable contaminant positions and fluxes achieve improved calibration across contaminated test sets. These results demonstrate the promise of SBI for scalable, uncertainty-aware LSBG morphology inference, while emphasizing that reliable posteriors require training simulations that capture the observational nuisance effects present in survey data. Advisor: Alex Drlica-Wagner
May
21
