China’s marine carbon sink capacity assessment and potential projection: a machine learning approach

Unofficial AI-generated summary based on the public title and abstract. Not an official translation.

🔬研究者:The machine learning framework (XGBoost with SHAP) for carbon sink projection is methodologically novel and can be replicated in other regions.

🏢実務担当者:Coastal managers and carbon project developers can use the identified drivers and threshold effects to optimize mariculture and conservation for carbon sequestration.

🏛政策担当者:Chinese policymakers can leverage scenario analysis to align marine resource management with 'Dual Carbon' goals; global policymakers see a model for blue carbon accounting.

The intensification of global climate change poses severe challenges to ecosystems and human development. Marine carbon sinks, as a critical natural climate solution, have placed their potential assessment and trend prediction at the centre of global climate governance and policymaking. As the world’s largest carbon emitter, China urgently requires scientifically grounded identification of the incremental potential and regulatory pathways of marine carbon sinks to achieve its “Dual Carbon” goals. This study employs panel data from 11 coastal provinces and municipalities in mainland China, specifically Liaoning, Hebei, Tianjin, Shandong, Jiangsu, Shanghai, Zhejiang, Fujian, Guangdong, Guangxi, and Hainan (2005–2022) and integrates multidimensional indicators spanning environmental conditions, human activities, and policy measures. In this study a predictive framework that combines machine learning with interpretability tools was also developed. Using XGBoost to capture complex nonlinear relationships, the model achieves a prediction accuracy of 95.7%, and SHAP analysis was applied to quantify the marginal contributions and threshold effects of key drivers. Key findings include the following: (1) The number of natural reserves, mariculture areas, and total wastewater discharge are identified as core drivers, while chlorophyll-a concentration and the number of research personnel serve as important moderators—each exhibiting distinct “ecological thresholds”. (2) Multi-scenario projections for 2023–2032 indicate that the Green Development scenario yields the highest annual carbon sink potential (4.0061 million tC), surpassing the Business-As-Usual (3.2133 million tC) and Economy-Priority (3.0872 million tC) scenarios. The latter shows an initial decline of 13.4% due to deviation from ecological thresholds. (3) Significant regional heterogeneity is observed: the Northern Coastal Economic Belt is dominated by mariculture, with EP ≈ BAU > GP; the Eastern Coastal Economic Belt is primarily driven by urbanisation rate. With GP substantially outperforming others, the Southern Coastal Economic Belt follows a dual-core-driven pattern of mariculture and sea surface temperature, where GP demonstrates both optimal and stable outcomes. This research provides a scalable, data-driven approach for projecting marine carbon sink dynamics, offering actionable insights for adapting coastal management to climate change and evidence-based policy formulation in China and for other maritime regions.

• openalex https://doi.org/10.3389/fmars.2026.1839741first seen 2026-05-17 07:27:02 · last seen 2026-05-20 05:16:33