GDGT-based determination of paleoenvironments via machine learning

Summary Glycerol Dialkyl Glycerol Tetraethers (GDGTs) are a group of molecules that have been successfully used for paleoclimate reconstructions in a wide range of environments. GDGTs have shown to be useful biomarkers as they are widely available, and their structure is relatively resistant to diagenetic alteration. Specific GDGTs are associated with particular environments, which suggests that they could potentially be used to classify ancient depositional conditions in cases where the exact paleoenvironment is not known. In this work we present a comprehensive branched and iGDGT dataset from published and novel analyses, with more than 1000 modern samples from marine, lake and river sediments as well as soils and peat samples. Unsupervised and supervised machine learning techniques were applied to this dataset to generate GDGT and environmentally based clusters and train a classification algorithm. The trained algorithm has the potential to be applied to samples formed in unclear or transitional depositional environments, allowing us to identify the potential environment in which they were formed. Furthermore, the classification model can help determine which environmental calibration is most appropriate for the setting, thereby improving paleoenvironmental reconstructions.