Quantifying Ecological Tipping Points of Benthic Foraminifera Across the Permian-Triassic Boundary: A Machine Learning Approach

The Permian-Triassic mass extinction (PTME) represents the most severe ecological crisis of the Phanerozoic. Although previous studies have indicated that the decline of benthic foraminifera during this period was driven by the synergistic effects of multiple extreme environmental events, precisely quantifying the dominant factors and ecological tipping points driving shifts in community structure remains a challenge due to the differential responses of distinct morphogroups to environmental stress. To address this, we constructed a high-resolution paleoecological dataset integrating multi-proxy geochemical records (e.g., paleotemperature, total organic carbon (TOC), and redox state) and relative abundance data of benthic foraminiferal morphogroups from classical P-T sections across South China. We first applied Redundancy Analysis (RDA) to resolve the overall response patterns of community structure to environmental gradients. Subsequently, we deployed a Random Forest machine learning pipeline to accurately quantify the relative contribution of each environmental factor to specific morphogroups and identify the dominant drivers. Ultimately, by coupling Partial Dependence Plot (PDP) analysis and mathematically isolating the confounding effects of environmental multicollinearity, this study precisely extracted the marginal ecological tipping points of different ecological functional groups in response to single environmental drivers. This study not only provides a quantitative reference for exploring the ecological crisis mechanisms of the P-T boundary but also offers a methodological reference for assessing the vulnerability of benthic ecosystems under extreme greenhouse climates.