.png){kind=icon}
.png){kind=icon}
Return
Paleoclimate controls on organic matter enrichment in Mesozoic lacustrine shales: A synthesis based on geochemical big data and machine learning
.png){kind=icon}
PRE
AI
C
Z
X
F
C
X
DOI：10.1016/j.earscirev.2026.105636.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Mesozoic lacustrine shales represent significant source rocks in global unconventional oil and gas exploration. These shale systems record the complex responses of terrestrial organic matter (OM) enrichment under greenhouse climatic conditions. However, the mechanisms by which paleoclimate modifies the key factors controlling OM enrichment on a global scale remain unclear. More importantly, the thresholds at which these factors promote OM enrichment have not yet been established. To address this, geochemical data from 35 Mesozoic lacustrine basins worldwide were compiled. An interpretable machine-learning framework combining CatBoost and Shapley Additive Explanations (SHAP) was then developed to quantify the nonlinear relationships between total organic carbon (TOC) and paleoclimate, paleosalinity, redox conditions, paleoproductivity, terrestrial input, and sedimentation rate. The results show that paleoclimate emerges as the most significant predictor of OM enrichment in lacustrine environments, as it strongly regulates variations in paleosalinity, redox conditions, paleoproductivity, terrestrial input, and sedimentation rate, thereby leading to diverse OM enrichment patterns. The study further identified several thresholds favorable for OM enrichment, including CIA > 79.3, Sr/Ba between 0.6 and 1.77, MoEF between 0.78 and 13.42, Ti < 0.35%, Nixs > 21.54 ppm, and LaN/YbN > 1.34. Once Ti exceeds 0.35%, its diluting effect becomes dominant and leads to a marked reduction in OM enrichment potential. These findings refine the quantitative model of climate-driven OM enrichment and provide new insights into assessing hydrocarbon source-rock potential. Additionally, this study demonstrates the potential of interpretable machine learning for identifying hidden geological regularities from multidimensional geochemical big data.
Journal
E
IF:
10
Papers:
3.8K
Citations:
4.2W
Organization
Cited Papers