Harmonizing Terrestrial Carbon Cycle Observations Over CONUS NEON Sites: Assessing the Information Contributions of Multiple Data Constraints

Accurate inventories of terrestrial carbon pools and fluxes are crucial for understanding ecosystem processes, tracking climate change impacts, and meeting the monitoring, reporting, and verification (MRV) requirements in international treaties and voluntary carbon markets. In meeting this need, the fusion of process-based modeling, field data, and remote sensing observations has the potential to provide more accurate and precise estimates than each alone. However, as the number of data constraints on a system increases, different sources of information can interact with each other in complex ways across space, time, and processes. In this study, we undertake a value-of-information analysis to assess the contribution of different observations to reducing carbon cycle uncertainties across pools, fluxes, and spatial domains within the PEcAn carbon cycle data assimilation system. We used a novel block-based Tobit Gamma Ensemble Filter to assimilate four synergistic data constraints, MODIS leaf area index, Landtrendr aboveground biomass, SMAP soil moisture, and SoilGrids soil organic C, into a process-based ecosystem model (SIPNET) at 39 National Ecological Observatory Network sites across the contiguous United States from 2012 to 2021. Results showed that Soil and Wood C contribute most to the total C uncertainty, and not only did we greatly reduce uncertainty and residual error in the directly constrained pools, but many observations also shared information across variables and space, leading to reductions in both uncertainty and residual errors across variables and locations. These indirect constraints helped identify synergies and conflicts among data streams and across space, which provide insights for further constraining carbon inventories. Overall, soil carbon remains the largest source of uncertainty in the overall carbon budget due to both its large size and limited observational constraints.