Geoscientific Model Development
Geoscientific Model Development
GMD
Articles | Volume 19, issue 14
Articles | Volume 19, issue 14
https://doi.org/10.5194/gmd-19-6777-2026
© Author(s) 2026. This work is distributed under
the Creative Commons Attribution 4.0 License.
https://doi.org/10.5194/gmd-19-6777-2026
© Author(s) 2026. This work is distributed under
the Creative Commons Attribution 4.0 License.
Articles | Volume 19, issue 14
Development and technical paper
 | 
24 Jul 2026
Development and technical paper |  | 24 Jul 2026

Biogeochemistry-Informed Neural Network (BINN v1.0) for improving accuracy of model prediction and scientific understanding of soil organic carbon storage

Haodi Xu, Joshua Fan, Feng Tao, Lifen Jiang, Fengqi You, Benjamin Houlton, Ying Sun, Carla P. Gomes, and Yiqi Luo

Viewed

Since the preprint corresponding to this journal article was posted outside of Copernicus Publications, the preprint-related metrics are limited to HTML views.

Total article views: 17,168 (including HTML, PDF, and XML)
HTML PDF XML Total Supplement BibTeX EndNote
17,013 93 62 17,168 34 21 21
  • HTML: 17,013
  • PDF: 93
  • XML: 62
  • Total: 17,168
  • Supplement: 34
  • BibTeX: 21
  • EndNote: 21
Views and downloads (calculated since 15 Jul 2025)
Cumulative views and downloads (calculated since 15 Jul 2025)

Viewed (geographical distribution)

Since the preprint corresponding to this journal article was posted outside of Copernicus Publications, the preprint-related metrics are limited to HTML views.

Total article views: 17,168 (including HTML, PDF, and XML) Thereof 17,079 with geography defined and 89 with unknown origin.
Country # Views %
  • 1
1
 
 
 
 
Latest update: 14 Aug 2026
Download
Short summary
We developed the Biogeochemistry-Informed Neural Network (BINN) which embeds a process-based model inside an AI framework so the model’s parameters can be learned from empirical data. BINN could recover prescribed parameters in synthetic tests and retrieve key processes controlling the modeled soil organic carbon when applied to ~25 000 soil profiles in US. It operates more than 50 times faster than the traditional Bayesian approach for data-model integration.
Read more
Share