Presentation #201.06 in the session Exoplanets Orbital Dynamics.
About the 60% of exoplanets have been detected through their transit signatures. This is one of the most effective detection methods and the simplest in terms of observation and data processing, however it requires high signal to noise ratio (SNR) for finding a real candidate. Since transits are shallow and low frequency signals it becomes difficult to find them in noisy lightcurves.
In this work we present a methodology to process noisy lightcurves to find transit signatures. In order to test our method, we observed 27 known transits at high cadences (20 fps) from the Observatorio Astronómico Nacional de San Pedro Mártir in Ensenada, B. C. Mexico, using the 0.84 m telescope and one of the 1.3 m telescopes of the TAOS II project (Transneptuninan Automated Occultation Survey) in different epochs. It is worth to mention that TAOS II will observe fields for ~2 hrs., it means that we do not expect to catch a full transit in a regular observation, so for this study we took only the ingress or the egress as events.
The methodology consists of three main steps: filtering, fitting and prediction. Firstly, we apply, to the lightcurves, a lowpass filter with cutoff ~10-4 Hz which is around the spectrum signature of transits; by filtering we improve SNRs by a factor of ~10. Then we use least squares to fit trapezoid functions to the filtered lightcurves, so we might obtain the depth and fall time. If the fitting is success, we finally apply Bayesian inference using Markov Chain Monte Carlo (MCMC) sampling, so we can compute values and their probabilities for the parameters of the transit modeling: period, fall time, conjunction time, orbital angle, planet radius, eccentricity and semimajor axis, additionally we used Chi square to validate our results. The aim is to compute a reliable prediction to plan future observations and follow up the candidate.
The main application of this algorithm is to analyze most of 50 thousand lightcurves of ~144,000 points each daily, to find transit candidates and predict the next event for following up purposes. Given that, the pipeline must be automated and fast. TAOS II will produce 50K high cadence lightcurves daily during five years, so it is worth to take advantage of this large data base.
We were able to find flux variations as low as 0.57% and fall times from 13.97 to 59.90 min, with Chi squares up to 0.996. From the 27 lightcurves 11 had photometry issues due to high temperature in detectors and other technical problems. This is the reason why only 16 curves passed the detection stage. In the prediction stage we were able to obtain ephemerids for the 16 events with errors from 20 min to 7 hrs.