Large language models for environmental research: systematic literature screening, relational knowledge mining, and quantitative data extraction

Environmental research is facing a growing challenge of data abundance coupled with knowledge fragmentation. Evidence on pollutant exposure, toxicological effects, and pollution control is rapidly expanding, but it remains scattered across heterogeneous literature sources and reported in inconsistent formats. This fragmentation limits the throughput and efficiency of traditional manual literature screening and data extraction. Large language models (LLMs) have demonstrated methodological potential in natural language understanding, contextual integration, and structured output generation, offering new opportunities for automated environmental evidence extraction and integration. This review examines the capabilities of LLMs in three interconnected tasks: systematic literature screening, relational knowledge mining, and quantitative data extraction. Specifically, this review summarizes progress in how LLMs can support the identification of relevant studies, the extraction of relationships among multi-level entities, and the conversion of unstructured texts into structured quantitative datasets. Furthermore, applications across environmental exposure, toxicological effects, and pollution control are discussed, and key limitations regarding evaluation frameworks, result traceability, and expert oversight are highlighted. Current evidence suggests that while LLMs can substantially improve data extraction efficiency and structured data generation, their reliable deployment still requires clear task definition, structured constraints, rule-based validation, and human review. In conclusion, this review highlights that LLMs serve as an important methodological bridge between rapidly expanding environmental literature and reusable environmental knowledge systems.