Room: Poster Area
Date: Wednesday, 20 May 2026
Time: 11:30 - 12:45 CEST
Session code 1BV.4
Biomass supply from agriculture, forestry, and sustainable cropping systems
U.S. Historical Soil Organic Carbon Mapping and Stock Estimation with Machine Learning
Short Introductive summary
High-resolution and temporally consistent SOC information is a prerequisite for credible assessments of sustainable resource strategies. A recent baseline for the U.S. produced 30-m annual SOC maps (1990–2022) and a national 0–30 cm stock ~60.4 Pg C using standardized observations and multi-model benchmarking. While informative, further precision gains are limited by (i) overlaps/duplicates and heterogeneous metadata across databases, (ii) restricted predictor sets that may under-represent soil/site controls, and (iii) modelling choices not optimised for high-dimensional, non-linear relationships. We therefore reposition that baseline as Stage-1 and present Stage-2: duplicate removal and standardisation of datasets, predictor enrichment, and re-benchmarking—followed by an upcoming deep learning stage.
Presenter
Joshua S. FU
The University of Tennessee, Civil and Environmental Engineering Dpt., USA
Presenter's biography
Dr. Joshua S. Fu is Chancellor’s Professor and James G. Gibson Professor of Engineering at the University of Tennessee and a joint appointee at ORNL. His research focuses on soil organic carbon, climate change, AI-driven Earth system modeling, and health. He is a Fellow of AAAS and A&WMA.
Biographies and Short introductive summaries are supplied directly by presenters and are published here unedited
Co-authors:
T.E. Yu, The University of Tennessee, Knoxville, USA
X. Zhang, The University of Tennessee, Knoxville, USA
Session reference: 1BV.4.4