LineForest

Summary

LineForest (Saad et al. 2024) is a pipeline that measures the integrated line strengths (equivalent widths) of numerous key transitions in BOSS spectra using a neural network. A neural network approach was adopted because other methods attempted, such as Gaussian profile fitting, automated continuum determination, and automated width estimation, could not be generalized across all sources and only worked well for a subset of stars.

Detailed Description

The LineForest pipeline derives equivalent widths from MWM BOSS spectra based on a convolution neural net trained on by-hand equivalent width measurements from ~3500 BOSS spectra spanning a range of target classes from young stellar objects to OB stars to cataclysmic variables.

List of Line Indices

Table 1 from Saad et al. (2024) is reproduced below:

Line	Lambda (Å)	Delta Lambda (Å)	Precision (dex)	Recall (dex)	EqW sigma	AbW sigma	Notes
Halpha	6562.8	200	0.98	0.96	0.08	0.09
Hbeta	4861.3	200	0.98	0.94	0.09	0.1
Hgamma	4340.5	200	0.95	0.93	0.1	0.12
Hdelta	4101.7	200	0.94	0.94	0.1	0.11
Hepsilon	3970.1	200	0.95	0.94	0.08	0.1	[E]; Ca II H blend
H8	3889.064	200	0.91	0.66	0.08	0.06
H9	3835.391	200	0.91	0.71	0.11	0.1
H10	3797.904	200	0.82	0.77	0.12	0.08
H11	3770.637	200	0.9	0.72	0.1	0.08
H12	3750.158	50	0.85	0.79	0.14	0.09
H13	3734.369	50	0.77	0.65	0.13	0.09
H14	3721.945	50	0.87	0.68	0.09	0.07
H15	3711.977	50	0.87	0.81	0.12	0.15
H16	3703.859	50	0.88	0.7
H17	3697.157	50	0.56	0.38	0.11	0.2
Pa7	10049.4889	200	0.9	0.84	0.13	0.14
Pa8	9546.0808	200	0.93	0.9	0.12	0.11
Pa9	9229.1200	200	0.92	0.85	0.11	0.12
Pa10	9014.909	200	0.83	0.72	0.16	0.17
Pa11	8862.782	200	0.86	0.78	0.08	0.12
Pa12	8750.472	200	0.86	0.92	0.11	0.11
Pa13	8665.019	200	0.73	1.0	0.07	0.08	[E]; Ca II 8662.140 Å blend
Pa14	8598.392	200	0.75	0.9	0.11	0.08
Pa15	8545.383	200	0.83	0.83			[E]; Ca II 8542.089 Å blend
Pa16	8502.483	200	0.62	1.0			[E]; Ca II 8498.018 Å blend
Pa17	8467.254	200	0.73	0.8
Ca II	8662.140	50	0.99	0.96	0.06	0.07	[L]; Pa 13 blend
Ca II	8542.089	50	0.98	0.98	0.06	0.07	[L]; Pa 15 blend
Ca II	8498.018	50	0.97	0.94	0.07	0.07	[L]; Pa 16 blend
Ca II K	3933.6614	200	0.94	0.92	0.08	0.08
Ca II H	3968.4673	200	0.96	0.92	0.07	0.08	[L]; Hepsilon blend
He I	6678.151	50	0.92	0.83	0.13	0.08	[E]; Fe I 6678 Å blend
He I	5875.621	50	0.71	0.55	0.13	0.1	[E]
He I	5015.678	50	0.43	0.51	0.15	0.17	[E]; Fe II 5018 Aring blend
He I	4471.479	50	0.81	0.62	0.14	0.11	[E]
He II	4685.7	50	0.08	0.38			[E]
N II	6583.450	50	0.74	0.66	0.19	0.11
N II	6548.050	50	0.65	0.53	0.21	0.12
S II	6716.440	50	0.81	0.74	0.12	0.08
S II	6730.816	50	0.69	0.59	0.16	0.1
Fe II	5018.434	50	0.76	0.85	0.15	0.16
Fe II	5169.030	50	0.81	0.81	0.09	0.08
Fe II	5197.577	50	0.57	1.0
Fe II	6432.680	50	0.71	0.5
O I	5577.339	50	0.8	0.53	0.12	0.06
O I	6300.304	50	0.47	0.56	0.19	0.11
O I	6363.777	50	0.2	0.25
O II	3727.42	50	0.82	0.53	0.23	0.17	[E]; Fe I 3727 Å blend
O III	4958.911	50	0.5	0.33			[E]
O III	5006.843	50	0.78	0.78	0.16	0.06	[E]
O III	4363.85	50	1.0	0.33			[E]
Li I	6707.760	50	0.88	0.64	0.12	0.09	[L]; Fe I 6707 Å blend

Validation and Uncertainties

The development of LineForest was motivated by YSO studies. In Saad et al. (2024), the indices are shown to correlate well with YSO properties.

However, how well LineForest works for other types of stars is less well vetted. For example, although CVs were included in the training set, comparison of CVs identified by visual inspection with those identified by Hα emission show poor overlap.

Data Products

LineForest expects rest-frame BOSS spectra sampled onto the common wavelength array. 

LineForest runs on the `mwmStar` (combined) BOSS spectra and the `mwmVisit` (individual visit) BOSS spectra.

LineForest produces only summary data products: the `astraAllStarLineForest` product contains results based on combined `mwmStar` BOSS spectra, and the `astraAllVisitLineForest` product contains results based on `mwmVIsit` BOSS spectra.

Flags

There are no flags for the LineForest measurements.