2021ApJ...921...95Z

We present a large uniform forward-modeling analysis for 55 late-T (T7-T9) dwarfs, using low-resolution (R ≃ 50-250) near-infrared (1.0-2.5 µm) spectra and cloudless Sonora-Bobcat model atmospheres. We derive the objects' effective temperatures, surface gravities, metallicities, radii, masses, and bolometric luminosities using our newly developed Bayesian framework, and use the resulting population properties to test the model atmospheres. We find (1) our objects' fitted metallicities are 0.3-0.4 dex lower than those of nearby stars; (2) their ages derived from spectroscopic parameters are implausibly young (10 Myr-0.4 Gyr); (3) their fitted effective temperatures show a similar spread to empirical temperature scales at a given spectral type but are ∼50-200 K hotter for >=T8 dwarfs; and (4) their spectroscopically inferred masses are unphysically small (mostly 1-8 M_Jup). These suggest the Sonora-Bobcat assumptions of cloudless and chemical-equilibrium atmospheres do not adequately reproduce late-T dwarf spectra. We also find a gravity and metallicity dependence of effective temperature as a function of spectral type. Combining the resulting parameter posteriors of our sample, we quantify the degeneracy between the fitted surface gravity and metallicity such that an increase in Z combined with a 3.4x increase in logg results in a spectrum that has similar fitted parameters. We note the systematic difference between the late-T dwarf spectra and Sonora-Bobcat models is on average ≃2%-4% of the objects' peak J-band fluxes over the 1.0-2.5 µm range, implying modeling systematics will exceed measurement uncertainties when analyzing data with J-band S/N >= 50. Using our large, high-quality sample, we examine the spectral-fitting residuals as a function of wavelength and atmospheric properties to discern how to improve the model assumptions. Our work constitutes the largest analysis of brown dwarf spectra using multimetallicity models and the most systematic examination of ultracool model atmospheres to date.