Accelerating Radiative Transfer in Planetary Atmospheres by Orders of Magnitude with Neural Network Emulators

Abstract

Radiative transfer calculations are essential for modeling planetary atmospheres. However, standard methods are computationally demanding and impose accuracy-speed trade-offs. High computational costs force numerical simplifications in large models (e.g., general circulation models) that degrade the accuracy of the simulation. Radiative transfer calculations are an ideal candidate for machine learning emulation; fundamentally, radiative transfer is a well-defined physical mapping from a static atmospheric profile to the resulting fluxes, and high-fidelity training data can be created from first-principles calculations. We developed a radiative transfer emulator using an encoder-only transformer neural network architecture, trained on 1D profiles representative of solar-composition hot Jupiter atmospheres. Our emulator reproduced bolometric two-stream layer fluxes with mean test set errors of ∼1% compared to the traditional method and achieved speedups of more than 100× in batched GPU inference. In matched comparisons against a long short-term memory model (LSTM) baseline, a transformer was consistently the more accurate and more robust architecture when training data were limited, while the LSTM achieved better accuracy at full scale. Emulating radiative transfer with machine learning opens up the possibility for faster and more accurate routines within planetary atmospheric models.

Accelerating Chemical Kinetics for Exoplanet Atmospheres using Neural Networks
Isaac Malsky, Xi Zhang, Tiffany Kataria +5
arXiv1 Sep 2026PreprintCC BY 4.0Full textPlanetary science & exoplanetsarXiv
Energy deposition in planetary and exoplanetary atmospheres induced by cosmic rays
Jesse Polman, Ingo Leya
arXiv16 Sep 2026PreprintCC BY 4.0Full textHigh-energy astrophysics Planetary science & exoplanetsarXiv
The 3D Cosmic Shoreline for Nurturing Planetary Atmospheres
Zach K. Berta-Thompson, Patcharapol Wachiraphan, Catriona A. Murray
The Astrophysical Journal29 Oct 2026Peer-reviewedCC BY 4.0Full textSolar & stellar physicsIOP/AASCrossref
Planet formation in chemically diverse and evolving discs II. Chemical fingerprints in planetary atmospheres
E. Pacetti, D. Turrini, E. Schisano +13
arXiv1 Jul 2026PreprintCC BY 4.0Full textPlanetary science & exoplanetsarXiv
Benchmarking Machine Learning Emulators of Stellar Evolution for Precision Asteroseismology
Naomi Gluck, Earl P. Bellinger, Yan Liang +5
The Astrophysical Journal5 Oct 2026Peer-reviewedCC BY 4.0Full textSolar & stellar physicsCrossref

Cite

BibTeX RIS

BibTeX

@article{malsky2026accelerating,
  title = {{Accelerating Radiative Transfer in Planetary Atmospheres by Orders of Magnitude with Neural Network Emulators}},
  author = {Isaac Malsky and Tiffany Kataria and Natasha E. Batalha and Matthew Graham},
  journal = {The Astrophysical Journal},
  year = {2026},
  month = nov,
  doi = {10.3847/1538-4357/aeb019},
  url = {https://iopscience.iop.org/article/10.3847/1538-4357/aeb019},
  note = {CC BY 4.0},
}

RIS

TY  - JOUR
TI  - Accelerating Radiative Transfer in Planetary Atmospheres by Orders of Magnitude with Neural Network Emulators
AU  - Isaac Malsky
AU  - Tiffany Kataria
AU  - Natasha E. Batalha
AU  - Matthew Graham
PY  - 2026
DA  - 2026/11/05
JO  - The Astrophysical Journal
DO  - 10.3847/1538-4357/aeb019
UR  - https://iopscience.iop.org/article/10.3847/1538-4357/aeb019
AB  - Radiative transfer calculations are essential for modeling planetary atmospheres. However, standard methods are computationally demanding and impose accuracy-speed trade-offs. High computational costs force numerical simplifications in large models (e.g., general circulation models) that degrade the accuracy of the simulation. Radiative transfer calculations are an ideal candidate for machine learning emulation; fundamentally, radiative transfer is a well-defined physical mapping from a static atmospheric profile to the resulting fluxes, and high-fidelity training data can be created from first-principles calculations. We developed a radiative transfer emulator using an encoder-only transformer neural network architecture, trained on 1D profiles representative of solar-composition hot Jupiter atmospheres. Our emulator reproduced bolometric two-stream layer fluxes with mean test set errors of ∼1% compared to the traditional method and achieved speedups of more than 100× in batched GPU inference. In matched comparisons against a long short-term memory model (LSTM) baseline, a transformer was consistently the more accurate and more robust architecture when training data were limited, while the LSTM achieved better accuracy at full scale. Emulating radiative transfer with machine learning opens up the possibility for faster and more accurate routines within planetary atmospheric models.
SN  - 1538-4357
SN  - 0004-637X
ER  -

Abstract

Related

Cite