Single Transit Detection In Kepler With Machine Learning And Onboard Spacecraft Diagnostics

TL;DR

Detecting long-period exoplanets from Kepler data is challenging due to few transits per system. The authors implement a CNN ensemble that also ingests Kepler onboard diagnostics to classify single-transit events, locate transit centers, and recover orbital periods, achieving robust performance out to $800$ days. Applying the pipeline to the KOI 1271 system yields a new candidate KOI 1271.02 with $R_p = 5.32 \pm 0.20\,R_\oplus$ and mass constraints from TTV modeling, suggesting a resonant configuration with KOI 1271.01. The work demonstrates that using ancillary spacecraft data with ML improves sensitivity to long-period planets and informs estimates of the $\beta$-Earth occurrence rate, though additional transits are needed to tighten dynamical constraints. Together, these results pave the way for leveraging single-transit detections to expand the Kepler planet census and refine long-period planet demographics.

Abstract

Exoplanet discovery at long orbital periods requires reliably detecting individual transits without additional information about the system. Techniques like phase-folding of light curves and periodogram analysis of radial velocity data are more sensitive to planets with shorter orbital periods, leaving a dearth of planet discoveries at long periods. We present a novel technique using an ensemble of Convolutional Neural Networks incorporating the onboard spacecraft diagnostics of \emph{Kepler} to classify transits within a light curve. We create a pipeline to recover the location of individual transits, and the period of the orbiting planet, which maintains $>80\%$ transit recovery sensitivity out to an 800-day orbital period. Our neural network pipeline has the potential to discover additional planets in the \emph{Kepler} dataset, and crucially, within the $η$-Earth regime. We report our first candidate from this pipeline, KOI 1271.02. KOI 1271.01 is known to exhibit strong Transit Timing Variations (TTVs), and so we jointly model the TTVs and transits of both transiting planets to constrain the orbital configuration and planetary parameters and conclude with a series of potential parameters for KOI 1271.02, as there is not enough data currently to uniquely constrain the system. We conclude that KOI 1271.02 has a radius of 5.32 $\pm$ 0.20 $R_{\oplus}$ and a mass of $28.94^{0.23}_{-0.47}$ $M_{\oplus}$. Future constraints on the nature of KOI 1271.02 require measuring additional TTVs of KOI 1271.01 or observing a second transit of KOI 1271.02.

Figures (23)

• Figure 1: One of the 25 architectures within the ensemble using only the flux values of the host star. The convolutional layers are described with kernel size - # of filters, (# of layers), pooling layers are described with pool size - stride length, and the fully connected layers with # of neurons (# of layers). Shown above is an example input of a detrended light curve, containing 500 long-cadence data points.
• Figure 2: Histogram of the individual accuracies of the 25 networks in the flux-only ensemble. Also shown is the ensemble accuracy. The ensemble accuracy is greater than any one individual accuracy within the ensemble.
• Figure 3: Example of one of the 25 architectures in the specific engineering attribute ensembles used to determine the best-performing attributes. The columns follow the same convention as Figure \ref{['fig:NN_arch']}. Each column's input size is 500. After the dropout layer, the outputs for the two columns are concatenated and used as input into the fully connected layer.
• Figure 4: One of the 25 architectures in our final ensemble including the engineering attribute columns. The columns follow the same convention as Figure \ref{['fig:NN_arch']}. The name of the attribute for each column is listed at the top. Each column's input size is 500. After the dropout layer, the output for each column is concatenated and used as input into the fully connected layer.
• Figure 5: A comparison of two ensembles, one including the top 11 engineering attributes along with the flux (blue) and one including the top 5 engineering attributes along with the flux (red). The histograms are the individual network accuracies within the ensemble, and the vertical lines are the respective ensemble accuracy. The ensemble accuracy for both ensembles is well above any individual accuracy within the ensemble. We also see over a 3% increase in the ensemble accuracy compared to using only the flux values (see Figure \ref{['fig:flux_accuracies']}.