AirGPT: pioneering the convergence of conversational AI with atmospheric science
Jun Song, Chaoyong Ma, Maohao Ran
Abstract
Large language models (LLMs) face significant limitations in specialized scientific domains due to their inability to perform data analysis and their tendency to generate inaccurate information. This challenge is particularly critical in air quality management, where precise analysis is essential for addressing climate change and pollution control initiatives. To bridge this gap, we present AirGPT, a computational framework that integrates conversational AI with atmospheric science expertise through a curated corpus of peer-reviewed literature and specialized data analysis capabilities. Through a novel architecture combining natural language processing and domain-specific analytical tools, AirGPT achieved higher accuracy in air quality assessments compared to standard LLMs, including GPT-4o. Experimental results demonstrate superior capabilities in providing accurate regulatory information, performing fundamental data analysis, and generating location-specific management recommendations, as validated through case studies in metropolitan areas such as Beijing.