AI for Natural Methane (AI4NM): A Community Effort to Harmonize Natural Methane Datasets Using Knowledge-Guided Machine Learning
Welcome to the AI for Natural Methane (AI4NM) Working Group — a collaborative community sponsored by the Environmental Data Science Innovation & Inclusion Lab (ESIIL). Our mission is to advance global understanding of natural methane sources and sinks by harmonizing model and observation datasets using knowledge-guided machine learning (KGML) methods.
Our Goals
- Harmonize modeled and observed natural methane datasets into an AI-ready data pool.
- Do science with harmonized dataset and KGML to improve our understanding of natural methane sources and sinks.
- Grow AI literacy via open KGML packages and tutorials.
Working Group Team (As of October 2025)
Co-PIs
Youmi Oh (youmi.oh@noaa.gov), Licheng Liu (lliu223@wisc.edu), Sparkle Malone (sparkle.malone@yale.edu), Gavin McNicol (gmcnicol@uic.edu)
All Team
Alison Hoyt, Ammara Talib, Avni Malhotra (group lead), Ben Gaubert, Ben Riddell-Young, Bradley A. Gay, Colin Quinn, Danielle Potocek, Eric Ward, Etienne Fluet-Chouinard, Fa Li, Fenghui Yuan, Gavin McNicol (group lead), Housen Chu, Jennifer Watts, Jianqiu Zheng (group lead), Kayla Borton, Kelly Wrighton, Kendalynn Morris, Kevin Rozmiarek (group lead), Kunxiaoji Yuan, Kyle Arndt, Licheng Liu, Lori Bruhwiler, Madeline Scyphers, Michael Yonker, Minkyu Moon, Nicole Cai, Qianlai Zhuang, Qing Ying, Qing Zhu (group lead), RongChao Dong, Sara Knox, Shuo Chen (group lead), Sparkle Malone (group lead), Stefan Metzger, Wang Shaoyu, Xiaowei Jia, Xueying Yu, Yi Yang, Yiming Sun, Youmi Oh, Yujie Liu, Zhen Zhang, Zichong Chen
- Our affiliation can be found: https://docs.google.com/spreadsheets/d/1xVoMTReT30rbvdaD4w0QPra3FjtPc4I5zj12ZM48NB0/edit?usp=sharing
Photos from our Workshops in Oct 2024 (left) and in Oct 2025 (right)
Motivation of Our Project
- Atmospheric methane (CH4) grew at the fastest rate ever recorded in 2020-2022. Slowing or reversing the accelerating growth in atmospheric CH4 will require an improved understanding of the global CH4 budget. Natural CH4 emissions are responsible for ~40% of the total global CH4 budget but remain the most uncertain factor.
- Our working group aims to build a novel Knowledge-Guided Machine Learning (KGML) framework that integrates scientific knowledge from process-based models and machine learning to harmonize simulated and observed datasets from global wetlands and soil sinks.
Goal of our Working Group
- The objective of this working group is to synthesize multiple measured and simulated datasets using a KGML framework to better constrain natural CH4 fluxes from wetlands and soil sinks.
- This KGML framework will be designed to integrate scientific knowledge from bottom-up and top-down models, machine learning models and multi-source data through knowledge-guided architecture pretraining and training.
- Specifically, we will harmonize the following four types of natural CH4 datasets within the KGML framework; We will use the bottom-up and top-down estimates of natural CH4 sources and sinks to pretrain the model. We then use observation data from chamber and FLUXNET-CH4 measurements to fine-tune the model.
Code Repository
https://github.com/CU-ESIIL/AI-for-Natural-Methane
Contributing to This Repository
We welcome contributions from all group members. To maintain the quality and integrity of the repository, please adhere to the following guidelines:
- Make sure all commits have a clear and concise message.
- Document any major changes or decisions in the meeting notes.
- Review and merge changes through pull requests to ensure oversight.