Influences of paleoclimatic changes on organic matter enrichment mechanisms in freshwater and saline lacustrine oil shales in China: A machine learning approach

Lacustrine oil shales are valuable unconventional resources, with their organic matter (OM) enrichment closely linked to paleoclimatic conditions. However, the mechanisms controlling OM enrichment in freshwater and saline lacustrine environments remain underexplored. This study represents the first application of the random forest (RF) classifier to investigate how paleoclimatic changes influence OM enrichment mechanisms in lacustrine oil shales in China. By analyzing the bulk geochemical (total organic carbon (TOC), trace and major elements) and molecular biomarker data from seven representative lacustrine oil shales, we demonstrate that the RF classifier can effectively discriminate between freshwater and saline lacustrine oil shales. The results indicate that the TOC content is the most significant parameter, with freshwater shales having higher TOC values than their saline counterparts. Further analyses suggest that OM enrichment in these two types of lacustrine settings was driven by a combination of factors, including paleoproductivity, paleowater conditions, and terrigenous inputs, all modulated by paleoclimatic changes. Freshwater lacustrine shales were generally deposited during humid and warm periods with higher atmospheric CO₂ levels and greater precipitation, which led to higher paleoproductivity, reduced terrigenous influx, rising lake levels, and improved OM production and preservation. In contrast, saline lacustrine shales were deposited during dry and cold periods, when increased water salinity and diminished paleoproductivity resulted in lower OM production, reduced preservation, and decreased TOC content. These findings provide key insights into the role of paleoclimate in shaping OM-rich lacustrine shales and establish a framework for evaluating organic-rich shale deposits. The study also underscores the potential for future geochemical datasets to enhance classification models and broaden applications of machine learning in the exploration of hydrocarbon resources.