China unveils new large model for carbon emission accounting

China unveiled a new large model for carbon emission accounting on Wednesday, marking a major technological breakthrough in the global field of carbon accounting.

Carbon emission accounting is a crucial basis for international climate compliance, a cornerstone of global carbon pricing and an essential prerequisite for meeting peak carbon and carbon neutrality goals.

The model is the world's first panoramic carbon emission accounting system covering production-side, consumption-side and natural sources, according to its developer, the Shanghai Advanced Research Institute under the Chinese Academy of Sciences (CAS).

It is designed to address bottlenecks in traditional carbon accounting, including high knowledge barriers, cumbersome data processing, long timeframes and low resolution. By leveraging generative artificial intelligence (AI), it aims to reshape the paradigm of carbon accounting, dynamically depicting global carbon flows and traceability, thereby comprehensively enhancing the country's technological discourse power in global climate governance.

Built on a scientific foundation model (ScienceOne) from the CAS, the new model rests on three pillars. They are eight proprietary datasets that enable high-frequency data updates and fusion; an independently developed methodology using a large language model (LLM)-powered agent with multi-agent collaboration to ensure accuracy and a hybrid computing cluster that optimizes resources across internal servers and external centers.

Currently, the model's service interface provides a vertical-domain LLM with 32 billion parameters and an intelligent database, offering both conversational and programming interfaces.

Five intelligent agents with specific functions have been developed, which can respectively realize digital simulation and optimization of industrial system processes, trade carbon transfer accounting, life cycle assessment, natural source accounting and uncertainty analysis.

Among them, the life cycle assessment agent can independently complete the entire process of goal and scope definition, inventory analysis, accounting and result interpretation, enabling automated product carbon footprint accounting.

Based on the model, a high-precision national-level carbon holographic map has been initially developed. Taking 2022 as an example, under the scientifically equitable new accounting framework, the greenhouse gas emissions of China, the United States and Japan were adjusted by -17.7%, +15.2% and +7.2%, respectively, compared with the results of traditional Intergovernmental Panel on Climate Change (IPCC) production-side accounting.

The model also revealed that the default emission factors of the European Union's Carbon Border Adjustment Mechanism systematically overestimate Chinese product emission factors, highlighting the need to improve accounting accuracy and adopt localized factors.

It has also accurately calculated the emission reduction contributions of China's green products to the world. For instance, the country's exported wind turbines and photovoltaic products generated approximately 2 million tonnes of carbon emissions during the production phase in 2024, yet delivered approximately 350 million tonnes of carbon reduction benefits globally during their operational phase.

The model provides critical support for China's compilation of national greenhouse gas inventories, the development of its national carbon market, the green transition of key industries and the country's response to international carbon policies. It also contributes Chinese expertise to global efforts to establish a fairer and more scientifically sound system of carbon accounting and responsibility allocation.