Source, sink and preservation of organic matter from a machine learning approach of polar lipid tracers in sediments and soils from the Yellow River and Bohai Sea, eastern China

Abstract

Transport and transformation of organic matter (OM) from the river to the marginal sea is a significant part of the global carbon cycle. Biomarkers are of indispensable advantage in precisely identifying the origin of OM that is crucial to understand the organic carbon cycle. Application of more biomarker molecules with mutually confirmable information commonly implies stricter constraint of the source but also brings challenges to the data analysis and interpretation due to a large amount of molecular information. Here we used random forest (RF) classification models to analyze 123 polar lipid biomarkers of six categories, including fatty alcohols, fatty acids, alkan-2-ones, steroids, triterpenoids, and 1-O-monoalkylglycerol ethers (MAGEs) from the sediments and soils in the Yellow River and the Bohai Sea of eastern China. The environmental specificity of biomarkers was assessed based on the effective distinguishment of samples from different habitats by RF models. The sources of polar lipid biomarkers were constrained according to their environmental specificity, and four genetic classifications, i.e., bacteria, algae and zooplankton, terrestrial higher plants, and anthropogenic input were identified. The spatial distribution of OM sources provides a reasonable scheme of the sink for biospheric OM in this typical “land-river-ocean” system. A type of MAGEs as the most important variables for the RF models was effectively used to be a potential bottom-water oxygen proxy to assess the preservation of OM, and ca. 37% of marine in-situ fresh OM was estimated to decompose under varying redox conditions in the surface sediments of Bohai Sea.