Accurate Machine Learning Atmospheric Retrieval via a Neural Network Surrogate Model for Radiative Transfer
Michael D. Himes, Joseph Harrington, Adam D. Cobb, Atilim Gunes Baydin, Frank Soboczenski, Molly D. O'Beirne, Simone Zorzan, David C. Wright, Zacchaeus Scheffer… 展开作者
Michael D. Himes, Joseph Harrington, Adam D. Cobb, Atilim Gunes Baydin, Frank Soboczenski, Molly D. O'Beirne, Simone Zorzan, David C. Wright, Zacchaeus Scheffer, Shawn D. Domagal-Goldman, Giada N. Arney
备注：
16 pages, 4 figures, submitted to PSJ 3/4/2020, revised 1/22/2021, accepted 2/4/2021, published 4/25/2022. Updated to match the published manuscript. Himes et al. 2022, PSJ, 3, 91
Atmospheric retrieval determines the properties of an atmosphere based on its measured spectrum. The low signal-to-noise ratio of exoplanet observations require a Bayesian approach to determine posterior probability distributions of each model parameter, given observed spectra. This inference is computationally expensive, as it requires many executions of a costly radiative transfer (RT) simulation for each set of sampled model parameters. Machine learning (ML) has recently been shown to provide a significant reduction in runtime for retrievals, mainly by training inverse ML models that predict parameter distributions, given observed spectra, albeit with reduced posterior accuracy. Here we present a novel approach to retrieval by training a forward ML surrogate model that predicts spectra given model parameters, providing a fast approximate RT simulation that can be used in a conventional Bayesian retrieval framework without significant loss of accuracy. We demonstrate our method on the emission spectrum of HD 189733 b and find good agreement with a traditional retrieval from the Bayesian Atmospheric Radiative Transfer (BART) code (Bhattacharyya coefficients of 0.9843--0.9972, with a mean of 0.9925, between 1D marginalized posteriors). This accuracy comes while still offering significant speed enhancements over traditional RT, albeit not as much as ML methods with lower posterior accuracy. Our method is ~9x faster per parallel chain than BART when run on an AMD EPYC 7402P central processing unit (CPU). Neural-network computation using an NVIDIA Titan Xp graphics processing unit is 90--180x faster per chain than BART on that CPU.
This and companion papers by Harrington et al. and Blecic et al. present the Bayesian Atmospheric Radiative Transfer (BART) code, an open-source, open-development package to characterize extrasolar-planet atmospheres. BART combines a thermochemical equilibrium abundances (TEA), a radiative-transfer (Transit), and a Bayesian statistical (MC3) module to constrain atmospheric temperatures and molecular abundances for given spectroscopic observations. Here, we describe the Transit radiative-transfer package, an efficient line-by-line radiative-transfer C code for one-dimensional atmospheres, devel…
This and companion papers by Harrington et al. 2021, submitted and Cubillos et al. 2021, submitted describe an open-source retrieval framework, Bayesian Atmospheric Radiative Transfer (BART), available to the community under the reproducible-research license via https://github.com/exosports/BART . BART is a radiative-transfer code (transit, https://github.com/exosports/transit , Rojo 2009, 2009ASPC..420..321R), initialized by the Thermochemical Equilibrium Abundances (TEA, https://github.com/dzesmin/TEA , Blecic et al. 2016, arXiv:1505.06392) code, and driven through the parameter phase space …
We present the open-source Bayesian Atmospheric Radiative Transfer (BART) retrieval package, which produces estimates and uncertainties for an atmosphere's thermal profile and chemical abundances from observations. Several BART components are also stand-alone packages, including the parallel Multi-Core Markov chain Monte Carlo (MC3), which implements several Bayesian samplers; a line-by-line radiative-transfer model, transit; a code that calculates Thermochemical Equilibrium Abundances, TEA; and a test suite for verifying radiative-transfer and retrieval codes, BARTTest. The codes are in Pytho…
The Radiative Transfer Model (RTM) and the retrieval algorithm, incorporated in the SCIATRAN 2.2 software package developed at the Institute of Remote Sensing/Institute of Enviromental Physics of Bremen University (Germany), allows to simulate, among other things, radiance/irradiance spectra in the 2400-24 000 Å range. In this work we present applications of RTM to two case studies. In the first case the RTM was used to simulate direct solar irradiance spectra, with different water vapor amounts, for the study of the water vapor content in the atmosphere above Sierra Nevada Observatory. Simula…
Radiative transfer models explain and predict interaction between solar radiation and the different elements present in the atmosphere, which are responsible for energy attenuation. In Colombia there have been neither measurements nor studies of atmospheric components such as gases and aerosols that can cause turbidity and pollution. Therefore satellite images cannot be corrected radiometrically in a proper way. When a suitable atmospheric correction is carried out, loss of information is avoided, which may be useful for discriminating image land cover. In this work a computational model was u…
Transmission spectroscopy is one of the premier methods used to probe the temperature, composition, and cloud properties of exoplanet atmospheres. Recent studies have demonstrated that the multidimensional nature of exoplanet atmospheres -- due to non-uniformities across the day-night transition and between the morning and evening terminators -- can strongly influence transmission spectra. However, the computational demands of 3D radiative transfer techniques have precluded their usage within atmospheric retrievals. Here we introduce TRIDENT, a new 3D radiative transfer model which rapidly com…
A semi-analytical surrogate model for giant planet evolution: bypassing ordinary differential equation solvers with localised thermodynamics, softplus asymptotes, and B-spline photometry
LIRA, Observatoire de Paris, Université PSL, Sorbonne Université, Université Paris Cité, CY Cergy Paris Université, CNRS（巴黎天文台，PSL大学，索邦大学，巴黎西岱大学，塞吉-蓬图瓦兹大学，法国国家科学研究中心）；Space Telescope Science Institute（太空望远镜科学研究所）；Department of Physics & Astronomy, University of Exeter（埃克塞特大学物理与天文学系）；Observatoire de la Côte d’Azur, Université Côte d’Azur（蔚蓝海岸天文台，蔚蓝海岸大学）
The determination of atmospheric parameters depends on the use of radiative transfer codes (among other elements such as model atmospheres) to compute synthetic spectra and/or derive abundances from equivalent widths. However, it is common to mix results from different surveys/studies where different setups were used to derive the parameters. These inhomogeneities can lead us to inaccurate conclusions. In this work, we studied one aspect of the problem: When deriving atmospheric parameters from high-resolution stellar spectra, what differences originate from the use of different radiative tran…
Accelerating Radiative Transfer for Planetary Atmospheres by Orders of Magnitude with a Transformer-Based Machine Learning Model
Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA 91109, USA（喷气推进实验室、加州理工学院）；NASA Ames Research Center, Moffett Field, CA 94035, USA（NASA阿姆斯研究中心）；California Institute of Technology, Pasadena, CA, 91125, USA（加州理工学院）
Detailed radiative transfer simulations of kilonova spectra play an essential role in multimessenger astrophysics. Using the simulation results in parameter inference studies requires building a surrogate model from the simulation outputs to use in algorithms requiring sampling. In this work, we present KilonovaNet, an implementation of conditional variational autoencoders (cVAEs) for the construction of surrogate models of kilonova spectra. This method can be trained on spectra directly, removing overhead time of pre-processing spectra, and greatly speeds up parameter inference time. We build…
Observations and magnetohydrodynamic simulations of solar and stellar atmospheres reveal an intermittent behavior or steep gradients in physical parameters, such as magnetic field, temperature, and bulk velocities. The numerical solution of the stationary radiative transfer equation is particularly challenging in such situations, because standard numerical methods may perform very inefficiently in the absence of local smoothness. However, a rigorous investigation of the numerical treatment of the radiative transfer equation in discontinuous media is still lacking. The aim of this work is to ex…
