Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Emission-GPT: A domain-specific language model agent for knowledge retrieval, emission inventory and data analysis

Jiashu Ye, Tong Wu, Weiwen Chen, Hao Zhang, Zeteng Lin, Xingxing Li, Shujuan Weng, Manni Zhu, Xin Yuan, Xinlong Hong, Jingjie Li, Junyu Zheng, Zhijiong Huang, Jing Tang

TL;DR

Emission-GPT presents a domain-specific LLM agent built to retrieve, organize, and analyze atmospheric-emissions information by integrating retrieval-augmented generation, structured prompting, and function calling. The system rests on a curated knowledge base of over 10,000 documents, processed into millions of context chunks to support accurate inventory methods, emission-factor recommendations, and data-driven analyses. Evaluation combines automated, expert, and case-study assessments, showing high factual alignment and actionable outputs while highlighting challenges in multilingual context and retrieval precision. The work demonstrates a practical, extensible platform that lowers entry barriers for emission inventories and scenario-based analyses, with potential for real-time decision support and policy assessment.

Abstract

Improving air quality and addressing climate change relies on accurate understanding and analysis of air pollutant and greenhouse gas emissions. However, emission-related knowledge is often fragmented and highly specialized, while existing methods for accessing and compiling emissions data remain inefficient. These issues hinder the ability of non-experts to interpret emissions information, posing challenges to research and management. To address this, we present Emission-GPT, a knowledge-enhanced large language model agent tailored for the atmospheric emissions domain. Built on a curated knowledge base of over 10,000 documents (including standards, reports, guidebooks, and peer-reviewed literature), Emission-GPT integrates prompt engineering and question completion to support accurate domain-specific question answering. Emission-GPT also enables users to interactively analyze emissions data via natural language, such as querying and visualizing inventories, analyzing source contributions, and recommending emission factors for user-defined scenarios. A case study in Guangdong Province demonstrates that Emission-GPT can extract key insights--such as point source distributions and sectoral trends--directly from raw data with simple prompts. Its modular and extensible architecture facilitates automation of traditionally manual workflows, positioning Emission-GPT as a foundational tool for next-generation emission inventory development and scenario-based assessment.

Emission-GPT: A domain-specific language model agent for knowledge retrieval, emission inventory and data analysis

TL;DR

Emission-GPT presents a domain-specific LLM agent built to retrieve, organize, and analyze atmospheric-emissions information by integrating retrieval-augmented generation, structured prompting, and function calling. The system rests on a curated knowledge base of over 10,000 documents, processed into millions of context chunks to support accurate inventory methods, emission-factor recommendations, and data-driven analyses. Evaluation combines automated, expert, and case-study assessments, showing high factual alignment and actionable outputs while highlighting challenges in multilingual context and retrieval precision. The work demonstrates a practical, extensible platform that lowers entry barriers for emission inventories and scenario-based analyses, with potential for real-time decision support and policy assessment.

Abstract

Improving air quality and addressing climate change relies on accurate understanding and analysis of air pollutant and greenhouse gas emissions. However, emission-related knowledge is often fragmented and highly specialized, while existing methods for accessing and compiling emissions data remain inefficient. These issues hinder the ability of non-experts to interpret emissions information, posing challenges to research and management. To address this, we present Emission-GPT, a knowledge-enhanced large language model agent tailored for the atmospheric emissions domain. Built on a curated knowledge base of over 10,000 documents (including standards, reports, guidebooks, and peer-reviewed literature), Emission-GPT integrates prompt engineering and question completion to support accurate domain-specific question answering. Emission-GPT also enables users to interactively analyze emissions data via natural language, such as querying and visualizing inventories, analyzing source contributions, and recommending emission factors for user-defined scenarios. A case study in Guangdong Province demonstrates that Emission-GPT can extract key insights--such as point source distributions and sectoral trends--directly from raw data with simple prompts. Its modular and extensible architecture facilitates automation of traditionally manual workflows, positioning Emission-GPT as a foundational tool for next-generation emission inventory development and scenario-based assessment.

Paper Structure

This paper contains 25 sections, 10 equations, 10 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Data and methods
  3. Pipeline for Emission-GPT
  4. Dataset preparation and preprocessing
  5. Workflow of RAG
  6. Emission factors recommendation
  7. Providing targeted references to standards and guidelines
  8. Querying for emission factors
  9. Ranking the recommended emission factors
  10. Tools design and prompt engineering
  11. Tool functionality design
  12. Prompt engineering for structured invocation
  13. Results and discussion
  14. Model performance evaluation and benchmark datasets
  15. Automatic evaluation of knowledge base performance
  16. ...and 10 more sections

Figures (10)

  • Figure 1: Pipeline for Emission-GPT.
  • Figure 2: Structured prompt engineering framework for controlled function invocation.
  • Figure 3: Model performance automatic evaluation of Emission-GPT. (a–-f) EvalScope evaluation metrics (average faithfulness, answer relevancy, semantic similarity, context relevance, context precision, context recall) of Emission-GPT under different question categories and question difficulties.
  • Figure 4: The comparative winning rates of Emission GPT, GPT-4o and DeepSeek R1 across different dimentions
  • Figure 5: Radar plots of model-specific performance across the five evaluation dimensions for 3 difficulty levels
  • ...and 5 more figures