Toward Robust Atmospheric Retrieval on Cloudy L Dwarfs: the Impact of Thermal and Abundance Profile Assumptions

• Published in: The Astrophysical journal Vol. 947; no. 1; pp. 6 - 22
• Main Authors: Rowland, Melanie J., Morley, Caroline V., Line, Michael R.
• Format: Journal Article
• Language: English
• Published: Philadelphia The American Astronomical Society 01-04-2023; IOP Publishing
• Subjects: Abundance; Astrophysics; Atmospheric composition; Atmospheric models; Atmospheric structure; Brown dwarfs; Clouds; Infrared spectra; L dwarfs; Near infrared radiation; Opacity; Parameterization; Planetary atmospheres; Stellar atmospheres; Synthetic data
• ISSN: 0004-637X; 1538-4357
• DOI: 10.3847/1538-4357/acbb07

Constraining L dwarf properties from their spectra is challenging. Near-infrared (NIR) spectra probe a limited range of pressures, while many species condense within their photospheres. Condensation creates two complexities: gas-phase species “rain out” (decreasing in abundances by many orders of magnitude) and clouds form. We designed tests using synthetic data to determine the best approach for retrieving L dwarf spectra, isolating the challenges in the absence of cloud opacity. We conducted atmospheric retrievals on synthetic cloud-free L dwarf spectra derived from the Sonora Bobcat models at SpeX resolution using a variety of thermal and chemical abundance profile parameterizations. For objects hotter than L5 (
T
eff
∼ 1700 K), the limited pressure layers probed in the NIR are mostly convective; parameterized pressure–temperature (
PT
) profiles bias results and free, unsmoothed profiles should be used. Only when many layers both above and below the radiative-convective boundary are probed can parameterized profiles provide accurate results. Furthermore, a nonuniform abundance profile for FeH is needed to accurately retrieve bulk properties of early-to-mid L dwarfs. Nonuniform prescriptions for other gases in NIR retrievals may also be warranted near the L/T transition (CH
4
) and early Y dwarfs (Na and K). We demonstrate the utility of using realistic, self-consistent models to benchmark retrievals and suggest how they can be used in the future.