In the 3D-ABC project, we develop a foundation model (FM) focusing on three-dimensional above and below ground terrestrial carbon stocks. With this artificial intelligence (AI) model, we target the accurate mapping and quantification of terrestrial carbon stocks in vegetation and soils at high spatial resolution. By integrating multimodal remote sensing datasets and addressing complex challenges such as multi-dimensionality and multi-resolution in FMs, the 3D-ABC FM aims to provide a seamless understanding of vegetation and soil carbon distribution.
Our initial approach targets eight specific downstream tasks that partially build on top of each other and grow in complexity, focusing on either continental-scale demonstration areas or global data. Model performance will be evaluated in continental-scale regions, including the Amazon rainforest and Arctic-Boreal permafrost regions.
Partners
Six Helmhotz Centres collaborate in the 3D-ABC project.
Downstream Tasks
The 3D-ABC project includes eight defined downstream tasks.
Publications & Data
Explore the results and findings from the 3D-ABC project.
The Amazon Region
The Amazon is a critical global carbon reservoir, storing vast amounts of carbon in above-ground biomass and playing a key role in regulating the global carbon cycle. Forest biomass is a direct indicator of carbon storage, and quantification is critical for understanding carbon stock distribution and carbon dynamics over time. [read more]
The Arctic-Boreal Region
In the Arctic-Boreal region, permafrost soils store more carbon than is currently present in the atmosphere. These below-ground carbon stocks are poorly understood due to spatial heterogeneity and limited sampling across vast areas. 3D-ABC will generate high-resolution soil carbon maps for the permafrost region. [read more]
The 3D-ABC FM integrates large-scale global remote sensing datasets, including satellite-based multispectral imagery, InSAR coherence data, and 3D lidar point clouds and waveform data from space, aircraft, and ground-based platforms. It also incorporates climate reanalysis information, digital elevation data, as well as local lidar and field data on vegetation, soils, and carbon flux parameters. High-resolution forest models will be used to benchmark carbon fluxes.
Foundation Model (FM)
To accommodate the diverse modalities of data assembled for the 3D-ABC FM and to address the varying prediction requirements of multiple downstream tasks, the AI model employs an adaptive architecture. This architecture integrates a multi-modal input processor, an FM encoder, an adaptive fusion neck, and task-specific prediction heads. [read more]
High-Performance Computing (HPC)
High-Performance Computing is a crucial component of the 3D-ABC project, playing a key role in the training, validation, and fine-tuning of the 3D-ABC FM. The extensive geospatial datasets are processed and integrated in the 3D-ABC model on the HPC infrastructure at the Forschungszentrum Jülich. Computing primarily targets the JUWELS Booster and JUPITER infrastructure. [read more]
Interested in learning more? Explore the project structure, including our Work Packages and Deliverables, and find out how we contribute to the HFMI initiative.
Please reach out to us for more details!
Project
Discover the project’s structure and key components.
Contact
Get in touch for questions or collaboration opportunities.