In brief
This research tests deep-learning strategies for forecasting global carbon dioxide and methane concentrations six months ahead from satellite and environmental data.
Contribution
A global spatiotemporal forecasting comparison that integrates satellite observations with dynamic and static environmental variables.
Key finding
Graph neural networks delivered the strongest results, showing the value of explicitly representing global spatial relationships.
Models and data
Distinctive research elements
Models and methods
- Graph neural networks
- Spatiotemporal deep learning
Data and evaluation
- Global satellite observations
- Dynamic environmental variables
- Static environmental variables
Author-written overview
Research overview
This research compares deep-learning strategies for six-month global forecasts of carbon dioxide and methane concentrations. The models combine satellite observations with dynamic and static environmental variables and represent measurements distributed across the planet. The comparison tests how temporal learning and an explicit spatial graph contribute to prediction quality. Graph-neural approaches obtain the strongest overall results, indicating that global relationships between observation locations add useful information beyond temporal and contextual inputs alone. The study demonstrates a practical route for integrating heterogeneous environmental information into global greenhouse-gas forecasting.
Cite this work
BibTeX
@article{semper2024global,
title = {Global forecasting of carbon concentration through a deep learning spatiotemporal modeling},
author = {Marc Semper and Manuel Curado and Jose F. Vicent},
journal = {Journal of Environmental Management},
year = {2024},
volume = {371},
eid = {122922},
doi = {10.1016/j.jenvman.2024.122922}
}Always check the publisher record for final volume, issue and page information before citing.