Better Input, Better Output: Improving Photometric Redshifts by enhancing training data and optimizing observations
Abstract
Photometric redshift methods are widely used in astronomy to estimate the distances to galaxies from broadband photometry. Accurate photometric redshifts are necessary for high precision cosmological measurements in surveys such as DES and LSST. This work presents three different approaches, based upon probabilistic techniques, to enhance existing photometric estimation techniques. The first method uses Information Gain and Entropy to calculate the effectiveness of filter sets in terms of their photometric redshift accuracy. Using code written to implement this technique we construct an optimal filter set of 6 filters designed for photometric redshifts and evaluate how the properties of these filters impact their photometric redshift performance. We further determine how to create a single complementary filter to the LSST filters that would most improve photometric redshift performance. The second method uses Gaussian Processes and Principal Component Analysis to map simulated Spectral Energy Distributions to the observed colors of galaxies. We show that this approach can reduce the bias and uncertainties in current photometric redshift estimators. Finally, we once again use Gaussian Processes but this time augment existing training data to improve redshift estimation. In this talk I will explain each new method in detail and show results demonstrating that the new techniques can complement and improve currently existing photometric redshift algorithms.
- Publication:
-
American Astronomical Society Meeting Abstracts #233
- Pub Date:
- January 2019
- Bibcode:
- 2019AAS...23340902K