Machine learning and large language models for life cycle inventory compilation: Current situation and future developments

With increasing requirements for environmental accountability, Life Cycle Assessment (LCA) is becoming key for sustainability reporting. Nevertheless, significant challenges remain regarding data availability, especially for emerging and low-carbon energy technologies, for which Life Cycle Inventory (LCI) data are usually scarce and spread across studies and reports. Common LCI challenges concern the exploitation of available, smaller or larger, LCA datasets and the collection of LCA data from various sources when these are not found in LCA databases. This study explores machine learning (ML), natural language processing, and large language models (LLM) applications to tackle such challenges and estimate missing LCI data. A thorough review of suggested ML and artificial intelligence (AI) applications for LCI compilation is performed, complemented by case studies investigating ML and generative LLM methods to impute or gather missing LCI data for car-driving, power plants, and geothermal energy systems. ML methods can provide more reliable estimations than simple linear regression even for small datasets, while generative LLMs are found to effectively identify and extract LCI information from scientific papers. The potential of ML and AI methods to facilitate LCI compilation and enhance data reliability and availability for the LCA of emerging energy technologies is large, highlighting the crucial role such methods can play for decarbonization. Nevertheless, relevant applications remain in their infancy. More research is needed to construct robust frameworks for large-scale deployment to complement traditional LCI methods, and ensure correct usage of ML algorithms towards automated, accurate, and interpretable AI-assisted LCA. • Data-driven methods can help fill LCI data gaps. • Different algorithms are applicable for different data sizes and structures. • Language models can extract useful information from documents. • Text-based matching can speed database selection. • Combined AI approaches can complement traditional LCI practices.