A machine learning method for estimating atmospheric trace gas concentration baselines

Gerrand, Kirstin; Fillola, Elena; Manning, Alistair J.; Arduini, Jgor; Krummel, Paul B.; Lunder, Chris R.; Mühle, Jens; O'Doherty, Simon; Park, Sunyoung; Prinn, Ronald G.; Reimann, Stefan; Young, Dickon; Rigby, Matthew

doi:10.5194/amt-19-5387-2026

Articles | Volume 19, issue 16

https://doi.org/10.5194/amt-19-5387-2026

© Author(s) 2026. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/amt-19-5387-2026

© Author(s) 2026. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 19, issue 16

Research article

|

19 Aug 2026

Research article |

| 19 Aug 2026

A machine learning method for estimating atmospheric trace gas concentration baselines

Kirstin Gerrand, Elena Fillola, Alistair J. Manning, Jgor Arduini, Paul B. Krummel, Chris R. Lunder, Jens Mühle, Simon O'Doherty, Sunyoung Park, Ronald G. Prinn, Stefan Reimann, Dickon Young, and Matthew Rigby

Supplement

https://doi.org/10.5194/amt-19-5387-2026-supplement

Data sets

A Machine Learning Method for Estimating Atmospheric Trace Gas Concentration Baselines: Training Data Kirstin Gerrand, Elena Fillola Mayoral, Alistair Manning, Matthew Rigby https://doi.org/10.5281/zenodo.20483977

The dataset of in-situ measurements of chemically and radiatively important atmospheric gases from the Advanced Global Atmospheric Gas Experiment (AGAGE) and affiliated stations (Version 20250123) Ronald Prinn et al. https://doi.org/10.60718/0FXA-QF43

Model code and software

A Machine Learning Method for Estimating Atmospheric Trace Gas Concentration Baselines: Code repository Kirstin Gerrand, Elena Fillola Mayoral, Matthew Rigby https://doi.org/10.5281/zenodo.20484075

Short summary

To analyse long-term trends in atmospheric trace gas concentrations, it is important to identify data points minimally affected by local pollution sources or air masses carried from other latitudes or altitudes. Traditional methods for detecting these “baselines” are computationally expensive or lack a basis in physical principles. This paper introduces a machine-learning method that uses meteorological data and offers significantly lower computational costs compared to physics-based techniques.