HSLA

While HASP provides visit-level and program-level coadds of 1-D spectra for both COS and STIS, the HSLA combines data from multiple observing programs into a single spectrum for each target in the MAST archive. It then assigns a target classification to each object.

For a given target, the software first creates separate spectra for each combination of instrument, grating, and (for COS) lifetime position. The COS spectra from multiple lifetime positions are combined into separate spectra for each grating.  Finally, the COS and STIS spectra are spliced together (not co-added) to span the full available wavelength range for each target.

The software then performs an automatic hierarchical object classification based on classifications from SIMBAD, NED and Phase II information. This enables searches based on object classification as well as by target name: see the full list of HSLA classification tiers.

The project provides a publicly available Python script and Jupyter Notebooks to enable custom coaddition by the community.

History:  Between 2016 and 2018, the original HSLA provided combined spectra for all publicly-available COS data (Peeples et al. 2017), but that effort required considerable human intervention to create the data products. It did not include new data obtained after 2018, nor were the coadded products updated with the latest calibration products. The new HSLA is fully automated and updates the spectrum of each target whenever it is re-observed or there are updates to the calibration.

Example HSLA coadd spectra are presented below.

The HASP and HSLA project releases are accompanied by a set of user-friendly Jupyter Notebooks that enable users to perform custom coadditions for specific use cases not supported by the automatic coadds. These Notebooks were developed to aid in setting up and running the coadd script and creating custom coadd products.

• Setup
• Introduction to the HSLA Data Products and Tools
• Combining COS Data from Multiple Lifetime Positions and Central Wavelengths  

Please see the Coadd Tutorial, Flux Scaling Tutorial, and Data Diagnostics notebooks on the HASP homepage.

To ensure the flux and wavelength accuracy of our data products, we performed a series of tests comparing each coadded spectrum with its component x1d datasets.  We used the analysis software developed for HASP and described in the HASP ISR (Debes et al. 2024).  To test the flux accuracy, each x1d spectrum was filtered to remove pixels with serious data-quality flags, binned into 20 wavelength bins, and compared with the similarly prepared coadded spectrum.  We considered only wavelength bins with S/N > 20 and only x1d spectra with at least five such wavelength bins.  For each wavelength bin, we computed the residual r = (F_x1d - F_coadd) / F_coadd.  We found that both the mean and standard deviation of the residuals are less than 0.05 for more than 96% of x1d spectra with S/N sufficient to make the comparison.  The wavelength accuracy was assessed by cross-correlating the prepared (unbinned) coadded and x1d spectra; we considered only the high S/N spectra described above.  Nearly 98% of spectra yield wavelength offsets of less than 1 pixel.  Both the flux and wavelength tests substantially exceed the benchmark success rate of 75%.

To characterize the flux-calibration accuracy of the HSLA data products, we examined the residuals between coadded spectra of standard WDs and their synthetic models. We focused on three standards, G191-B2B, GD71, and WD0308-565, which together span all COS and STIS modes.  For each standard star and observing mode, we checked the flux-calibration accuracy by comparing the observed spectra in the HSLA “cspec” files with high-fidelity synthetic model spectra from the CALSPEC database.  Across all the targets and modes, most of the residual means are less than 1%, and the residual standard deviations have a median of 2.3%.  Based on these results we conclude that the HSLA cspec spectra retain the absolute flux-calibration accuracy of 5% of individual observations and have a relative accuracy of 2.3%, only slightly lower than for individual observations.  Figure 3 shows the HSLA quick-look spectrum (stored in the “aspec.fits” file) of GD71 overlaid on the corresponding CALSPEC model.

To test the accuracy of our target associations (identifying multiple observations of the same target) and target classifications (star, galaxy, etc.), we conducted a survey of 100 randomly selected HSLA targets. We found that the accuracy of the target associations and coordinates was 98%, while the accuracy of the classifications and naming was 92%.  Most classification and naming failures occurred when a structure within a spatially resolved galaxy was the target and SIMBAD lists multiple unique objects within 2" of that target.  Users should exercise caution when using HSLA products for objects for which multiple unique SIMBAD or NED objects lie within 2".