Current browse context:
astro-ph.IM
Change to browse by:
References & Citations
Astrophysics > Instrumentation and Methods for Astrophysics
Title: Accurate Machine Learning Atmospheric Retrieval via a Neural Network Surrogate Model for Radiative Transfer
(Submitted on 5 Mar 2020 (v1), revised 25 Jan 2021 (this version, v3), latest version 3 May 2022 (v4))
Abstract: 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.
Submission history
From: Michael Himes [view email][v1] Thu, 5 Mar 2020 04:48:01 GMT (2153kb,D)
[v2] Mon, 9 Mar 2020 18:42:30 GMT (2153kb,D)
[v3] Mon, 25 Jan 2021 17:35:26 GMT (4801kb,D)
[v4] Tue, 3 May 2022 16:28:39 GMT (4800kb,D)
Link back to: arXiv, form interface, contact.