SNID-SAGE: A Modern Framework for Interactive Supernova Classification and Spectral Analysis

Abstract

We present SNID-SAGE (SuperNova IDentification-Spectral Analysis and Guided Exploration), a framework for supernova spectral classification with both a fully interactive graphical interface and a scriptable command-line pipeline for large-scale processing. The pipeline combines deterministic spectral preprocessing, FFT-based cross-correlation against a curated template library, ranking of candidate matches using a composite quality metric, and consolidation of redshift and classification solutions into a single result with associated quality and confidence estimates. SNID-SAGE includes an upgradeable template library (about 6000 spectra), interactive line identification with velocity measurements, and optional natural-language summaries of classification results. We evaluate SNID-SAGE using two complementary tests: (i) leave-one-out cross-validation, in which each template spectrum is matched against the remainder of the library; and (ii) large-scale application to WISeREP spectra with valid coverage across the 4000-7000 A interval, irrespective of spectral type, comprising approximately 46 000 spectra, with redshift validation against known host-galaxy measurements where available. The full validation results and the SNID-SAGE framework are publicly available, supporting integration into spectroscopic survey workflows.

Figures (10)

• Figure 1: Preprocessing pipeline illustrated for SN 2018bif. Top panel: Raw input spectrum (grey) and spectrum after logarithmic-wavelength rebinning (blue), with the fitted broadband continuum overlaid (dashed black). Bottom panel: Continuum-divided flattened spectrum ready for the cross-correlation stage.
• Figure 2: Illustration of cross-correlation components and match-quality diagnostics for SN 2018bif.Top panel: Normalised cross-correlation function between the observed spectrum and the best-matching template, shown as a function of redshift offset relative to the peak ($z - z_{\rm peak}$). The solid black curve shows the filtered correlation, while the dashed orange curve shows the antisymmetric component used in the classical Tonry--Davis $R$ statistic. The shaded band indicates the rms of the antisymmetric component. The peak height h and width w characterise the strength of the alignment and the sharpness of the redshift constraint. Middle panel: Continuum-flattened, apodised observed spectrum (blue) and the best-matching template (red) at the peak redshift, illustrating the degree of spectral agreement that drives the correlation. Bottom panel: Residual spectrum (observed minus template) at the peak redshift, with the shaded band indicating the standard deviation of the residuals. In SNID--SAGE, this residual scatter provides an explicit estimate of mismatch, complementing the peak width in the redshift uncertainty estimate and replacing reliance on the antisymmetric correlation component alone.
• Figure 3: Example of redshift-space clustering and cluster selection in SNID--SAGE. Visualisation of template matches for 2018bif shown as a function of redshift, template type, and $H\sigma\mathrm{LAP\text{-}CCC}$. Each point corresponds to a single template match; clusters form naturally as concentrations of matches at similar redshift within a given type. The highlighted cluster corresponds to the highest-ranked solution according to the cluster quality score $Q$.
• Figure 4: Leave-one-out residual distributions by transient type. Top: distribution of redshift residuals $\Delta z = z_{\mathrm{pred}} - z_{\mathrm{true}}$. Bottom: distribution of phase residuals $\Delta t = t_{\mathrm{pred}} - t_{\mathrm{true}}$. Vertical dashed lines mark zero residual. Redshift residuals are tightly centred for all major classes, while phase residuals show a stronger dependence on transient type, with broader distributions and extended tails for heterogeneous classes such as SLSNe and TDEs.
• Figure 5: Leave-one-out recovery rate as a function of phase for selected transient classes. Classification performance is highest near peak light and degrades at late phases, with stronger phase dependence and variability for heterogeneous and sparsely sampled classes such as SLSNe and TDEs.