Harvesting Global Oceanic Lead Data Using Machine Learning: Standardization, Alignment and Spatio-temporal Puzzle Assembly with Large Language Model

The scarcity and inherently challenging nature of acquiring observational data has long hampered our understanding of elemental lead (Pb) distribution across global oceans. Recent advancements in data mining and machine learning offer new directions to address this challenge. In this study, we capitalize on our team's existing data mining system, DataExpo, to collate Pb-related data sets accessible via the internet, aiming to depict a comprehensive picture of global oceanic Pb. The collected datasets are heterogeneous in nature, exhibiting variances across data descriptions, formats, and types due to disparate data sources. To facilitate congruity in data amalgamation, Llama2, a generative large language model (LLM), is used to achieve semantic understanding to standardize and align the differing data table headers. Following the standardization and alignment processes, the aggregated data are spliced according to geographical location and temporal period, creating a complete spatio-temporal puzzle of global oceanic lead. The ultimate portrayal of this panoramic view promises to shed light on the global cycling pathways of Pb within our oceans. Among the anticipated outcomes are the identification of origins of Pb contamination and the elucidation of potential cycling pathways. These findings are critical for targeted interventions to mitigate contamination effects and enhance our understanding of bio-geochemical processes. Further advancements in global oceanic lead estimates are foreseen through iterative enhancements of our generative large language model application methodology.