A dual large language model framework for forecasting maritime greenhouse gas emissions

Maritime transport is a major source of greenhouse gas emissions, and accurately forecasting them is key to formulating targeted policies such as carbon pricing and emission quotas. The accuracy of existing forecasting models is limited by the challenges they face in processing large multi-source datasets. This study introduces a dual large language model (LLM) framework, MarEmisNet-DualLLM, which integrates a time-series-focused LLM for capturing temporal patterns and a general-purpose LLM for integrating domain knowledge, unstructured text, and contextual reasoning. Empirical tests on three real-world maritime datasets demonstrate that it outperforms baseline methods. The framework could be used by the International Maritime Organization, shipping firms, and ports to support mitigation strategies like route optimization and monitor compliance, thereby advancing maritime decarbonization.