Improving Forest Carbon Sink Accounting Using Integrated Satellite-Ground Observations, Machine Learning, and Ecological Process Modeling

Forest ecosystems are the dominant terrestrial carbon sink and play a critical role in mitigating climate change. However, conventional approaches based on single methods often fail to accurately characterize background carbon dioxide (CO 2 ) concentrations, limiting the reliability of regional carbon accounting. An integrated forest carbon sink monitoring framework was developed and applied in Kunyu Mountain, China, for reconstructing net ecosystem productivity (NEP). Within this framework, satellite observations, mobile ground-based measurements, the Gradient Boosting Regression Trees (GBRT) algorithm, and the Boreal Ecosystem Productivity Simulator model were combined. With the GBRT-derived prior CO 2 field incorporated, the BEPS model’s performance improved: the correlation between simulations and observations increased from 0.571 to 0.802. Results indicate that Kunyu Mountain’s NEP exhibited a sustained increase, with a marked acceleration after 2016, coinciding with China’s 13th Five-Year Plan, leading to a >40% rise in annual carbon sequestration by 2024 relative to 2001. Seasonal analyses likewise indicate that the carbon uptake period became both longer and more intense. Further attribution analysis reveals that ecological conservation policies generated an additional 5%–7% increase (approximately 30–50 gC·m –2 ). This framework provides a novel approach to more accurately account for regional carbon sinks and evaluate the effectiveness of ecological policies.