Spectroscopic Sky Subtraction and Data Optimization in Observation Exposures from the Sloan Digital Sky Survey Fiber Optic Telescope
Loading...
Date
2020-12
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
The Ohio State University
Abstract
The Ohio State University is a participating institution in the Sloan Digital Sky Survey (SDSS), whose mission is creating a 3 dimensional map of the universe, including spectra of astronomical objects. Key spectral information includes the strength of emission and absorption lines. The purpose of this research is to identify an improved method of subtracting the unwanted sky signal from the desired star signals using the data collected by SDSS-IV. The project also intends to identify the fewest number of calibration sky fibers that should be used in future SDSS-V observations in order to increase the amount of science data collected throughout the life of the telescope.
Completing this research began with obtaining access to the data, and collaboration with members of the SDSS team: Dr. Johnson at OSU, and Dr. Holtzman at the New Mexico State University. My analysis was guided by Dr. Kiryung Lee at OSU's ECE department. First, a mathematical model of the data characteristics was developed. Next, appropriate methods of sky subtraction were identified, to compare against the current method used by SDSS. This included principal component analysis (PCA), as well as sparse PCA. Evaluation of the methods was performed using synthetic data, followed by evaluation with measured data. Fits files which were generated during SDSS-IV were used to plot and align the fibers' spectra for subtraction. Then an evaluation process was developed to calculate subtraction errors, since the ground truth in this case was unknown. Finally, the minimum number of calibration fibers required to meet noise specifications was identified.
The results began with the synthetic case in which the model was sparse data set of n samples and p pixels in which the samples were assumed similar, with variations in amplitude and white gaussian noise. The SPCA signal representation yielded the lowest RMS error in the synthetic case. Next, the measured data was used and partitioned into training and testing samples. The results from measured data deviated significantly from the synthetic case and the averaging of the four nearest fibers yielded the lowest RMS error. This is the method currently used by SDSS, but which had not been evaluated for error analysis until these experiments. The study concluded that no significant information would be lost by modifying the current practices and reducing the number of calibration sky fibers used from 35 down to 15, which would increase the science data per-observation by 7.5%.
Description
Keywords
sky subtraction, denoising, principal component analysis, sloan digital sky survey, astronomy