International maritime transportation is a major yet complex source of greenhouse-gas emissions, whose systemic drivers and network formation mechanisms are not fully captured by existing, often isolated, methodologies. To bridge this gap, this study develops a multi-scale, integrated analytical framework. We first employ an environmentally extended multi-region input–output model to quantify global maritime embedded carbon flows (2000–2020). We then combine a high-precision machine-learning model (MLP) with SHapley Additive exPlanations (SHAP) analysis to identify key drivers, and finally apply a weighted exponential random-graph model to uncover network generative mechanisms. Our analysis yields three pivotal insights that offer new perspectives beyond conventional approaches: (1) The global flow network exhibits a polarized core–periphery structure centered on major hubs like China, Singapore, and the United States. (2) Bilateral flow intensity is primarily driven by asymmetric economic structures, operating through robust nonlinear (e.g., U-shaped, inverted U-shaped) channels rather than linear relationships. (3) Network formation is co-driven by homophily in consumption and heterophily in industrial structure, with geographic distance a persistent barrier. These findings directly inform international climate policy: they advocate for expanding emission responsibility to include major consumer nations and logistics hubs, and call for policies that account for the nonlinear, structural drivers of carbon exchange. The machine learning code and data have been uploaded to GitHub. URL: https://github.com/zhaoliangovo/Project-of-global-maritime-embedded-carbon-flow-network.
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号