Soil Carbon Assimilation Effectively Constrains Carbon‐Cycle Model Forecasting

Abstract

Accurate modeling and prediction of soil organic carbon (SOC) stocks are critical for realistic estimates of climate‐carbon feedback and ecosystem carbon sequestration potential. Process‐based models are widely used for simulating the dynamics of SOC stocks but often have considerable uncertainty. Exploring how process‐based models might be better informed by observations is crucial for improving the ability to forecast SOC. In this study, we assimilated the SoilGrids SOC datasets into the Simplified Photosynthesis and Evapotranspiration model for 39 National Ecological Observatory Network (NEON) terrestrial sites across the continental US using the Predictive Ecosystem Analyzer platform. We explored several strategies to determine the best approach for assimilating both SOC data and other indirect data constraints. Our results show that direct and continuous constraint from SoilGrids SOC data is crucial to achieve more accurate and precise SOC and ecosystem respiration states. The effectiveness of data assimilation is strongly determined by the uncertainty of both models and observations across sites, therefore careful characterization and propagation of uncertainties are important. Indirect constraints on SOC from Moderate Resolution Imaging Spectroradiometer leaf area index, LandTrendr aboveground biomass, and Enhanced Soil Moisture Active Passive soil moisture data also help to reduce the uncertainty in SOC prediction. Overall, our results show that data assimilation represents a promising approach to providing more realistic SOC estimates for regional and global carbon budgets in the future.