Machine learning emulators of dynamic global vegetation models for fast gridded carbon budget predictions
Abstract
The land carbon sink explains most of the variability of the atmospheric CO2 growth rate, but the Global Carbon Budget estimates it with dynamic global vegetation models (DGVMs) run once a year. Emulators could provide it sooner if they reproduce each model outside its training period. Here we show that convolutional neural network emulators of 19 of these models, driven by ERA5 climate and their own past predictions, reproduce the global land sink of these models and its spread over two years not used in training. Global monthly net biome production (NBP) anomalies were reproduced with a median R^2 of 0.81. The error of the multi-model mean annual NBP was at most 0.23 GtC yr^-1. At the grid-cell level, the median R^2 of monthly anomalies was 0.30, partly because of errors that accumulate in the emulated NBP history.
Related
- From buffer-based tables to dynamic response curves for soil-specific lime requirement estimation using machine learning
- Improved permafrost soil representation in a dynamic global vegetation model enhances simulations of carbon dynamics and vegetation structure in boreal forests of Northeast China
- Dynamic evaluation of blast-induced slope stability in open-pit iron mines using deterministic and machine learning approaches
- A spatial downscaling framework for vegetation optical depth based on geographically weighted stacking ensemble learning
- SamudrACE: Fast and Accurate Coupled Climate Modeling With 3D Ocean and Atmosphere Emulators
BibTeX
@misc{ke2026machine,
title = {{Machine learning emulators of dynamic global vegetation models for fast gridded carbon budget predictions}},
author = {Piyu Ke and Xiaofan Gui and Stephen Sitch and Zhu Liu and Xun Li and Jiang Bian and Pierre Friedlingstein and Philippe Ciais},
howpublished = {arXiv},
year = {2026},
month = nov,
doi = {10.48550/arxiv.2611.04536},
eprint = {2611.04536},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2611.04536},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Machine learning emulators of dynamic global vegetation models for fast gridded carbon budget predictions
AU - Piyu Ke
AU - Xiaofan Gui
AU - Stephen Sitch
AU - Zhu Liu
AU - Xun Li
AU - Jiang Bian
AU - Pierre Friedlingstein
AU - Philippe Ciais
PY - 2026
DA - 2026/11/03
PB - arXiv
DO - 10.48550/arxiv.2611.04536
UR - https://arxiv.org/abs/2611.04536
AB - The land carbon sink explains most of the variability of the atmospheric CO2 growth rate, but the Global Carbon Budget estimates it with dynamic global vegetation models (DGVMs) run once a year. Emulators could provide it sooner if they reproduce each model outside its training period. Here we show that convolutional neural network emulators of 19 of these models, driven by ERA5 climate and their own past predictions, reproduce the global land sink of these models and its spread over two years not used in training. Global monthly net biome production (NBP) anomalies were reproduced with a median R^2 of 0.81. The error of the multi-model mean annual NBP was at most 0.23 GtC yr^-1. At the grid-cell level, the median R^2 of monthly anomalies was 0.30, partly because of errors that accumulate in the emulated NBP history.
ER -