Color Classification of Extrasolar Giant Planets: Prospects and Cautions

TL;DR

This paper explores how to classify extrasolar giant planets using color photometry, analyzing a large grid of models to determine the accuracy of classifying planets by metallicity and cloud properties with upcoming observational facilities.

Contribution

It provides a detailed analysis of the potential and limitations of using multivariate classification algorithms on photometric data to categorize exoplanets based on atmospheric properties.

Findings

Planets can be classified by metallicity with over 90% accuracy if clouds are absent and at least 3 filters are used.

Cloud properties can be distinguished with over 90% accuracy regardless of prior cloud knowledge.

The statistical classification pipeline is publicly available for future research and mission planning.

Abstract

Atmospheric characterization of directly imaged planets has thus far been limited to ground-based observations of young, self-luminous, Jovian planets. Near-term space- and ground- based facilities like \emph{WFIRST} and ELTs will be able to directly image mature Jovian planets in reflected light, a critical step in support of future facilities that aim to directly image terrestrial planets in reflected light (e.g. HabEx, LUVOIR). These future facilities are considering the use of photometry to classify planets. Here, we investigate the intricacies of using colors to classify gas-giant planets by analyzing a grid of 9,120 theoretical reflected light spectra spread across different metallicities, pressure-temperature profiles, cloud properties, and phase angles. We determine how correlated these planet parameters are with the colors in the \emph{WFIRST} photometric bins and other photometric bins proposed in the literature. Then we outline under what conditions giant planet populations can be classified using several supervised multivariate classification algorithms. We find that giant planets imaged in reflected light can be classified by metallicity with an accuracy of $>$90\% if they are \emph{a prior} known to not have significant cloud coverage in the visible part of the atmosphere, and at least 3 filter observations are available. If the presence of clouds is not known \emph{a priori}, directly imaged planets can be more accurately classified by their cloud properties, as oppposed to metallicity or temperature. Furthermore, we are able to distinguish between cloudy and cloud-free populations with $>$90\% accuracy with 3 filter observations. Our statistical pipeline is available on GitHub and can be extended to optimize science yield of future mission concepts.