Enhanced Identification of Marine Anthropogenic Carbon Accounting and Climate Neutrality Using GNNs and Knowledge Graph

This project presents an engineering implementation of CE&CRGNN, a hybrid framework for ship emission quantification and coastal carbon accounting. The framework addresses limitations of satellite-based point-source detection and bottom-up emission inventories that rely on subjective or unreliable data sources.

CE&CRGNN integrates remote sensing–based ship recognition with graph neural networks and knowledge graph reasoning. Ships are automatically classified from satellite imagery and linked to structured emission knowledge to enable robust and scalable carbon estimation. This design reduces dependence on manual surveys and AIS data while improving consistency in emission inference.

In addition to ship emissions, the project includes a module for assessing marine aquaculture carbon absorption, supporting integrated analysis of emission sources and carbon sinks. Based on this framework, a port-level carbon offset strategy is designed, connecting port activity and regional economic indicators to carbon responsibility allocation with an emphasis on fairness and coordination.

The released files include model architectures, data preprocessing workflows, knowledge graph construction logic, and experimental configurations, supporting reproducibility and further methodological development.