Application of the Trend Filtering Algorithm for Photometric Time Series Data

TL;DR

This paper presents a modified implementation of the Trend Filtering Algorithm (TFA) for photometric time series data, aiming to improve trend filtering accuracy and provide a MATLAB software tool for ground-based surveys.

Contribution

The authors introduce modifications to TFA, including uncertainty incorporation and template selection via clustering, and provide a MATLAB implementation tested on synthetic and real data.

Findings

TFA reduces dispersion in light curves but can over-filter variables.

Modifications improve detection of sinusoidal variables.

No significant advantage for transit detection was observed.

Abstract

Detecting transient light curves (e.g., transiting planets) requires high precision data, and thus it is important to effectively filter systematic trends affecting ground based wide field surveys. We apply an implementation of the Trend Filtering Algorithm (TFA) (Kovacs et al. 2005) to the 2MASS calibration catalog and select Palomar Transient Factory (PTF) photometric time series data. TFA is successful at reducing the overall dispersion of light curves, however it may over filter intrinsic variables and increase "instantaneous" dispersion when a template set is not judiciously chosen. In an attempt to rectify these issues we modify the original literature TFA by including measurement uncertainties in its computation, including ancillary data correlated with noise, and algorithmically selecting a template set using clustering algorithms as suggested by various authors. This approach may be particularly useful for appropriately accounting for variable photometric precision surveys and/or combined data-sets. In summary, our contributions are to provide a MATLAB software implementation of TFA and a number of modifications tested on synthetics and real data, summarize the performance of TFA and various modifications on real ground based data sets (2MASS and PTF), and assess the efficacy of TFA and modifications using synthetic light curve tests consisting of transiting and sinusoidal variables. While the transiting variables test indicates that these modifications confer no advantage to transit detection, the sinusoidal variables test indicates potential improvements in detection accuracy.