AI-assisted quantifications of carbon sequestration in soil and forest biomass of the continental US

Moonshot Award

Proposal Development

PI

Lifen Jiang

Abstract

To quantify soil organic carbon (SOC) storage and forest biomass across the continental US, we have developed BINN (Biogeochemistry-Informed Neural Network), which seamlessly integrates models and big data through neural network. We have applied BINN to retrieve biogeochemical parameters of a soil carbon cycle model from big data to quantify SOC storage in the continental US. We further evaluated how major processes regulate spatial distributions of SOC. We have also applied BINN to quantify the time-dependent efficacy of carbon dioxide removal (CDR) via forest restoration in the continental US. We found that effective forest-based CDR is limited by forest age, with CDR rate peaking within 9.4–31.2 years of forest age and then declining as forests become mature. BINN demonstrates high accuracy and improved computational efficiency. It has a flexible architecture that can be adapted to facilitate other data-model integration research in carbon cycle and beyond.