A Convolutional Neural Network for the Detection of Gravity Waves in Satellite Observations and Numerical Simulations

Haruka Okui; Corwin James Wright; Peter G. Berthelemy; Neil P. Hindley; Lars Hoffmann; Andrew P. Barnes

doi:10.1029/2025GL115683

A Convolutional Neural Network for the Detection of Gravity Waves in Satellite Observations and Numerical Simulations

Haruka Okui , Corwin James Wright, Peter G. Berthelemy, Neil P. Hindley, Lars Hoffmann, Andrew P. Barnes

Research output: Contribution to journal › Article › peer-review

Abstract

Comparisons between observed and model-resolved gravity waves (GWs) are crucial for evaluating general circulation model (GCM) simulation accuracy and understanding wave characteristics. However, observational noise often obscures waves, complicating such comparisons. To address this, we have developed a GW detection method using a convolutional neural network (CNN). The CNN is trained on Atmospheric Infrared Sounder (AIRS) temperatures with labels indicating wave presence based on Berthelemy et al. (2025, https://doi.org/10.5194/egusphere-2025-455). Their method detects noise-induced pixel-to-pixel variations in horizontal wavelengths; in contrast, the CNN robustly identify waves even when applied to smoothly varying model data. Using this method, we compare stratospheric GWs in boreal winters between AIRS observations and a high-top GW-permitting GCM, Japanese Atmospheric GCM for Upper Atmosphere Research (JAGUAR). The results agree well and exhibit similar interannual variability, with discrepancies also identified, including a more zonally elongated distribution of tropical GWs in JAGUAR. This method is broadly applicable to the future use of satellites for guiding wave-resolving atmospheric model development.

Original language	English
Article number	e2025GL115683
Journal	Geophysical Research Letters
Volume	52
Issue number	11
Early online date	7 Jun 2025
DOIs	https://doi.org/10.1029/2025GL115683
Publication status	Published - 16 Jun 2025

Data Availability Statement

The AIRS temperature retrievals (Hoffmann & Alexander, 2009) are available from https://datapub.fz-juelich.de/slcs/airs/gravity_waves (Hoffmann, 2021). The CNN, a sample code to run the CNN, and processed JAGUAR-as-AIRS data can be downloaded from https://doi.org/10.5281/zenodo.15382321 (Okui, 2025). Version R2022b of MATLAB was used for spectral analysis of gravity waves (https://www.mathworks.com/products/matlab.html). Figures were produced using Python 3.10 (https://www.python.org/).

Acknowledgements

The high-resolution JAGUAR simulations were performed using the Earth Simulator at the Japan Agency for Marine-Earth Science and Technology (JAMSTEC). HO sincerely thanks Kaoru Sato and Shingo Watanabe for their valuable feedback while performing the simulations. This manuscript benefited from discussions at the International Space Science Institute in Bern as part of International Team 567.

Funding

HO acknowledges support from a Japan Society for the Promotion of Science (JSPS) Overseas Research Fellowship. PGB was supported by an URSA studentship awarded by the University of Bath and by Royal Society Grant RF\ERE\210079. CJW was supported by Royal Society Research Fellowship URF\R\221023 and NERC Grants NE/V01837X/1, NE/W003201/1, and NE/Z50399X/1. NPH was supported by NERC Grants NE/W003201/1 and NE/Z50399X/1, and NERC Fellowship NE/X017842/1.

Funders	Funder number
URSA
Japan Agency for Marine-Earth Science and Technology
Japan Society for the Promotion of Science
University of Bath
Natural Environment Research Council	NE/V01837X/1, NE/W003201/1, NE/X017842/1, NE/Z50399X/1
Royal Society	RF\ERE\210079, URF\R\221023

Keywords

general circulation model
gravity waves
machine learning
satellite observations
stratosphere

ASJC Scopus subject areas

Geophysics
General Earth and Planetary Sciences

Access to Document

10.1029/2025GL115683Licence: CC BY

Cite this

@article{705cabf8f40a432eafd29ab442600d9b,

title = "A Convolutional Neural Network for the Detection of Gravity Waves in Satellite Observations and Numerical Simulations",

abstract = "Comparisons between observed and model-resolved gravity waves (GWs) are crucial for evaluating general circulation model (GCM) simulation accuracy and understanding wave characteristics. However, observational noise often obscures waves, complicating such comparisons. To address this, we have developed a GW detection method using a convolutional neural network (CNN). The CNN is trained on Atmospheric Infrared Sounder (AIRS) temperatures with labels indicating wave presence based on Berthelemy et al. (2025, https://doi.org/10.5194/egusphere-2025-455). Their method detects noise-induced pixel-to-pixel variations in horizontal wavelengths; in contrast, the CNN robustly identify waves even when applied to smoothly varying model data. Using this method, we compare stratospheric GWs in boreal winters between AIRS observations and a high-top GW-permitting GCM, Japanese Atmospheric GCM for Upper Atmosphere Research (JAGUAR). The results agree well and exhibit similar interannual variability, with discrepancies also identified, including a more zonally elongated distribution of tropical GWs in JAGUAR. This method is broadly applicable to the future use of satellites for guiding wave-resolving atmospheric model development.",

keywords = "general circulation model, gravity waves, machine learning, satellite observations, stratosphere",

author = "Haruka Okui and Wright, {Corwin James} and Berthelemy, {Peter G.} and Hindley, {Neil P.} and Lars Hoffmann and Barnes, {Andrew P.}",

year = "2025",

month = jun,

day = "16",

doi = "10.1029/2025GL115683",

language = "English",

volume = "52",

journal = "Geophysical Research Letters",

issn = "0094-8276",

publisher = "John Wiley and Sons Inc.",

number = "11",

}

TY - JOUR

T1 - A Convolutional Neural Network for the Detection of Gravity Waves in Satellite Observations and Numerical Simulations

AU - Okui , Haruka

AU - Wright, Corwin James

AU - Berthelemy, Peter G.

AU - Hindley, Neil P.

AU - Hoffmann, Lars

AU - Barnes, Andrew P.

PY - 2025/6/16

Y1 - 2025/6/16

N2 - Comparisons between observed and model-resolved gravity waves (GWs) are crucial for evaluating general circulation model (GCM) simulation accuracy and understanding wave characteristics. However, observational noise often obscures waves, complicating such comparisons. To address this, we have developed a GW detection method using a convolutional neural network (CNN). The CNN is trained on Atmospheric Infrared Sounder (AIRS) temperatures with labels indicating wave presence based on Berthelemy et al. (2025, https://doi.org/10.5194/egusphere-2025-455). Their method detects noise-induced pixel-to-pixel variations in horizontal wavelengths; in contrast, the CNN robustly identify waves even when applied to smoothly varying model data. Using this method, we compare stratospheric GWs in boreal winters between AIRS observations and a high-top GW-permitting GCM, Japanese Atmospheric GCM for Upper Atmosphere Research (JAGUAR). The results agree well and exhibit similar interannual variability, with discrepancies also identified, including a more zonally elongated distribution of tropical GWs in JAGUAR. This method is broadly applicable to the future use of satellites for guiding wave-resolving atmospheric model development.

AB - Comparisons between observed and model-resolved gravity waves (GWs) are crucial for evaluating general circulation model (GCM) simulation accuracy and understanding wave characteristics. However, observational noise often obscures waves, complicating such comparisons. To address this, we have developed a GW detection method using a convolutional neural network (CNN). The CNN is trained on Atmospheric Infrared Sounder (AIRS) temperatures with labels indicating wave presence based on Berthelemy et al. (2025, https://doi.org/10.5194/egusphere-2025-455). Their method detects noise-induced pixel-to-pixel variations in horizontal wavelengths; in contrast, the CNN robustly identify waves even when applied to smoothly varying model data. Using this method, we compare stratospheric GWs in boreal winters between AIRS observations and a high-top GW-permitting GCM, Japanese Atmospheric GCM for Upper Atmosphere Research (JAGUAR). The results agree well and exhibit similar interannual variability, with discrepancies also identified, including a more zonally elongated distribution of tropical GWs in JAGUAR. This method is broadly applicable to the future use of satellites for guiding wave-resolving atmospheric model development.

KW - general circulation model

KW - gravity waves

KW - machine learning

KW - satellite observations

KW - stratosphere

UR - http://www.scopus.com/inward/record.url?scp=105007610353&partnerID=8YFLogxK

U2 - 10.1029/2025GL115683

DO - 10.1029/2025GL115683

M3 - Article

SN - 0094-8276

VL - 52

JO - Geophysical Research Letters

JF - Geophysical Research Letters

IS - 11

M1 - e2025GL115683

ER -

A Convolutional Neural Network for the Detection of Gravity Waves in Satellite Observations and Numerical Simulations

Abstract

Data Availability Statement

Acknowledgements

Funding

Keywords

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this