Transnational dependency networks shape trade-offs between decarbonization and economic growth

Enhancing carbon emission efficiency (CEE) is crucial for balancing economic growth with low-carbon transformation. Transnational regional dependencies from global integration offer both challenges and opportunities for collaborative decarbonization. This study presents a multi-objective optimization framework incorporating transnational dependency networks to improve CEE, reduce carbon emissions, and sustain economic development. First, a three-stage super-efficiency model quantifies national CEE. Second, a gravity model integrates economic, population, emission, and geographic factors to build spatio-temporal dependency networks. Third, based on network features, surrogate models for CEE, carbon emissions (CE), and gross domestic product (GDP) are constructed, using explainable machine learning and multi-objective algorithms to identify optimal strategies. A case study of 112 Belt and Road countries (2011–2020) shows: (1) Surrogate models incorporating cross-border dependencies perform well (R² for CEE, CE, GDP: 0.908, 0.927, 0.918); (2) Multi-factor optimization raises CEE by 12.8%, cuts CE by 29.2%, but reduces GDP by 22.7%; (3) Single-factor optimization lessens economic losses (0.48% CEE increase, 28.3% CE and 24.3% GDP reductions). The study’s novelty is: (a) proposing a transnational dependency network framework for efficiency measurement, modeling, and optimization; (b) developing three-dimensional optimization algorithms prioritizing CEE. This research advances theory and methods for low-carbon transition in Belt and Road countries, offering guidance for transnational decarbonization policies.提升碳排放效率（carbon emission efficiency, CEE）是实现经济增长与低碳转型协同平衡的关键所在。全球一体化进程催生的跨国区域依赖关系，为协同脱碳工作既带来了挑战，也创造了发展机遇。本研究构建了一种融入跨国依赖网络的多目标优化框架，旨在提升碳排放效率、降低碳排放总量并维持经济可持续发展。 研究共包含三个核心环节：其一，搭建三阶段超效率模型对各国碳排放效率进行量化测算；其二，引入引力模型，整合经济规模、人口规模、碳排放水平与地理区位等因素，构建时空依赖网络；其三，基于网络特征构建碳排放效率、碳排放（carbon emissions, CE）与国内生产总值（gross domestic product, GDP）的代理模型，结合可解释机器学习与多目标算法筛选最优策略方案。 以2011-2020年的112个‘一带一路’沿线国家为案例研究对象，结果显示：（1）融入跨境依赖关系的代理模型性能优异，碳排放效率、碳排放与国内生产总值的决定系数（R²）分别为0.908、0.927与0.918；（2）多因素优化可使碳排放效率提升12.8%，碳排放总量降低29.2%，但国内生产总值将下降22.7%；（3）单因素优化则可有效减轻经济损失，能够实现碳排放效率提升0.48%、碳排放总量降低28.3%与国内生产总值下降24.3%。 本研究的创新之处在于：（a）提出了面向效率测算、建模与优化的跨国依赖网络分析框架；（b）开发了以碳排放效率为优先目标的三维多目标优化算法。 本研究推动了‘一带一路’沿线国家低碳转型领域的理论与方法发展，可为跨国脱碳政策制定提供科学指导。