TaphoSource: An open-source database of experimentally generated bone surface modifications for modeling taphonomic...

Bone surface modifications (BSMs) provide data for modeling predator-prey dynamics and trophic interactions in the fossil record. However, the utility of BSM data depends on our ability to correctly infer which unobservable actions or ecological agents produced these marks, and commonly used qualitative methods may be insufficient to discriminate among BSMs that exhibit substantial macromorphological overlap. Recently, quantitative BSM identification methods have been employed to overcome this taphonomic equifinality, which, despite their methodological rigor, remain underused because they require large databases of experimentally generated BSMs. In the present study, we introduce TaphoSource, an open-source database containing point cloud and measurement data for 942 experimentally generated BSMs, created by replicating five possible taphonomic and depositional actions, as well as 12 fossilized BSMs from the 1.7-million-year-old HWK EE site in Olduvai Gorge, Tanzania. To highlight the utility of this dataset, we used random forest models to distinguish among the five experimentally generated BSM categories based on measurement variability, achieving approximately 73% accuracy. These data were then used to train a random forest model to identify the actions responsible for the 12 fossilized BSMs in the TaphoSource database. We openly disseminate these data so that other researchers can conduct quantitative BSM modeling studies without the need to generate expensive and time-consuming experimental datasets. We anticipate that these data will have broader applications for reconstructing and modeling spatiotemporal trends in carnivory, providing deeper insights into the evolution of both hominin and non-hominin mammalian carnivory and behavior throughout the Plio-Pleistocene.