STag II: Classification of Serendipitous Supernovae Observed by Galaxy Redshift Surveys

TL;DR

STag II enhances serendipitous supernova classification in large galaxy redshift surveys by training on DESI-like data that include host-galaxy contamination, introducing equivalent-width features, and applying an rlap-based trust cutoff and redshift checks. It redefines spectral tags, expands the input feature set, and scales the classifier to more realistic data, achieving high accuracy on simulated spectra (Type Ia and Ib ~99%, Ic ~88%, II ~100%). When tested on eight DESI observations matched to Transient Name Server (TNS), STag II correctly classified two cases after applying the rlap cutoff, while others remained inconclusive due to sparse SN light or redshift mismatches, highlighting the method’s robustness and current limitations. The work demonstrates STag II’s potential to support DESI’s transients pipeline and paves the way for future improvements in phase handling, sub-type expansion, and integration into value-added DESI catalogs.

Abstract

With the number of supernovae observed expected to drastically increase thanks to large-scale surveys like the Dark Energy Spectroscopic Instrument (DESI), it is necessary that the tools we use to classify these objects keep up with this increase. We previously created Supernova Tagging and Classification (STag) to address this problem by employing machine learning techniques alongside logistic regression in order to assign 'tags' to spectra based on spectral features. STag II is a continuation of this work, which now makes use of model supernova spectra combined with real DESI spectra in order to train STag to better deal with realistic data. We also make use of the rlap score as a trustworthiness cut, making for a more robust and accurate supernova classifier than before.

Figures (8)

• Figure 1: Histogram showing the distribution of flux ratios for the simulated DESI spectra. Due to the logarithmic conversion between magnitudes and fluxes, there is a tendency to generate lower flux ratios.
• Figure 2: Comparison of the original $\beta$ values for the Si ii$\lambda$6355 tag (originally labelled as Si ii$\lambda$6150 in Davison2022), which can be seen to also possibly include the absorption feature associated with He i$\lambda$5876, and the $\beta$ values for the Si ii$\lambda$6355 tag used in STag II, which now only contains the relevant feature to avoid potential contamination from other features.
• Figure 3: The cross-correlation function of a real DESI spectrum with a Type Ia template spectrum. This is a functional recreation of similar plots found in Blondin2007 and Muthukrishna2019, found in Figures 5 and 3 respectively.
• Figure 4: The rlap score for the spectra of SN 2021qtc when pre-processed at different redshifts. The rlap score drops quickly outside $\sim$1% of the true redshift value, meaning only when the redshift is accurate will the rlap score be able to pass the cutoff point of 6, indicated by the horizontal black dashed line.
• Figure 5: The normalised confusion matrix of the DESI simulated spectra set aside for testing STag II, which compares the estimated classification to the true class (as determined by which SN model was used). The numbers in brackets below the normalised fraction that were correctly classified correspond to the number of spectra that were classified as that class by STag II.