Kernel Regression Techniques for Enhancing Spitzer Photometric Precision
Abstract
The Infrared Array Camera (IRAC) on the Spitzer Space Telescope has been used to measure < 0.01% temporal variations in the fluxes of exoplanet systems. The IRAC PSF at both 3.6 and 4.5 μm is undersampled and thus the detector arrays show variations of as much as 8% in sensitivity as the center of the PSF moves across a pixel due to normal spacecraft motions. This is the largest source of correlated noise in IRAC photometry. We describe the latest progress towards an independent calibration of the intra-pixel gain that does not rely on the measurements to be calibrated. The technique begins with: (1) localizing the sub-pixel position of a point source using Spitzer’s Pointing Calibration and Reference Sensor (PCRS); and (2) harnessing a “training set” of many thousands of densely spaced photometric measurements of a non-variable star. Kernel regression, where the training data are nonlinearly combined based on a distance metric for each data point, leads to significant improvements in photometric precision over our previous gridded method. The distance metric we use was derived from a supervised learning algorithm to minimize regression error. We conclude that these results rival the precision obtained with self-calibration techniques, but do not risk the removal of astrophysical signals.
- Publication:
-
IAU General Assembly
- Pub Date:
- August 2015
- Bibcode:
- 2015IAUGA..2257999I