Detecting Ca II Absorption Lines with a Fe II assisted Dual Neural Network

TL;DR

This work addresses the rarity and weak features of Ca II absorbers by deploying a dual CNN framework that first detects Ca II and then cross-validates with Fe II quintuplet lines. Synthetic training data, a carefully curated Ca II test set, and the Faked Quintuplet Method enable high-precision detections across ~108k–110k SDSS DR16 quasar spectra, yielding 1,646 Ca II absorbers (including 1,121 new) and 95 2DAs. Comprehensive analyses—completeness, composite spectra, curve-of-growth, column densities, reddening, and abundance patterns—show strong vs. weak Ca II absorbers trace different gas environments, with Ca II present in ~1.5% of Mg II systems. The work significantly enlarges the Ca II catalog, enables robust ISM/dust studies, and offers a scalable approach for forthcoming large spectroscopic surveys such as DESI.

Abstract

Ca II absorbers, characterized by dusty and metal-rich environments, provide unique insights into the interstellar medium of galaxies. However, their rarity and weak absorption features have hindered comprehensive studies. In this work, we present a novel dual CNN approach to detect Ca II absorption systems, analyzing over 100,000 quasar spectra from the Sloan Digital Sky Survey (SDSS) Data Release 16. Our primary CNN identifies Ca II features, while a secondary CNN cross-verifies these detections using five Fe II absorption lines. This approach yielded 1,646 Ca II absorption systems, including 525 previously known absorbers and 1,121 new discoveries, nearly tripling the size of any previously reported catalog. Among our Ca II absorbers, 95 are found to show the 2175Å dust feature (2DA), corresponding to 22% of strong absorbers, 7% of weak absorbers, and $\sim$12% of the overall Ca II population at $0.8 < z_{\text{abs}} < 1.4$. Across the full redshift range of $0.36 < z_{\text{abs}} < 1.4$, $\sim$1.5% of Mg II absorbers host Ca II.

Figures (29)

• Figure 1: Examples of quasar spectra from the Ca II test set. The red dotted lines mark the expected positions of the Ca II $\lambda\lambda$ 3934, 3969 absorption lines. The top two spectra are positive samples containing Ca II absorption features, taken from the catalog provided by 2022MNRAS.517.4902X. The bottom two spectra are false samples randomly selected from the Mg II absorption catalog provided by 2019MNRAS.487..801Z, and do not contain Ca II absorption.
• Figure 2: Example of a Ca II absorber from 2014MNRAS.444.1747S that lacks visible Mg II absorption. The left panel shows the Ca II doublet, while the right panel displays the Mg II region, with red dashed lines indicating the expected line positions based on the absorption redshift.
• Figure 3: The distribution of the EW for the catalog by 2022MNRAS.517.4902X compared to the distribution of the measured EW of the our artificial training data. The training samples have slightly lower equivalent widths to improve our CNNs sensitivity.
• Figure 4: Samples of artificial spectra that are generated. The red dotted lines indicate the locations at which Ca II absorption was injected.The top plot shows a negative artificial spectrum, and the bottom plot shows a positive one. Both of these come from the same DR12 quasar spectra, with the only difference being the injected lines. Note that the top plot shows the spectra after 10 pixels around each line location have been replaced with white noise. This negative spectra is among the 70% of total spectra that have neither line injected. The other 30% would either have only the first line injected, or the second line injected.
• Figure 5: A demonstration of the process of cropping and concatenating two windows such that the five Fe II lines are joined and unnecessary data is removed to create a faked quintuplet for CNN training and Fe II line detection.