Abstract
Context. Blob detection is a common problem in astronomy. One example is instellar population modelling, where the distribution of stellar ages andmetallicities in a galaxy is inferred from observations. In this context, blobsmay correspond to stars born in-situ versus those accreted from satellites, andthe task of blob detection is to disentangle these components. A difficultyarises when the distributions come with significant uncertainties, as is thecase for stellar population recoveries inferred from modelling spectra ofunresolved stellar systems. There is currently no satisfactory method for blobdetection with uncertainties. Aims. We introduce a method for uncertainty-awareblob detection developed in the context of stellar population modelling ofintegrated-light spectra of stellar systems. Methods. We develop theory andcomputational tools for an uncertainty-aware version of the classicLaplacian-of-Gaussians method for blob detection, which we call ULoG. Thisidentifies significant blobs considering a variety of scales. As a prerequisiteto apply ULoG to stellar population modelling, we introduce a method forefficient computation of uncertainties for spectral modelling. This method isbased on the truncated Singular Value Decomposition and Markov Chain MonteCarlo sampling (SVD-MCMC). Results. We apply the methods to data of the starcluster M54. We show that the SVD-MCMC inferences match those from standardMCMC, but are a factor 5-10 faster to compute. We apply ULoG to the inferredM54 age/metallicity distributions, identifying between 2 or 3 significant,distinct populations amongst its stars.