fossilbrush: An R package for automated detection and resolution of anomalies in palaeontological occurrence data

Abstract

1. Fossil occurrence databases are indispensable resources to the palaeontological community, yet present unique data cleaning challenges. Many studies devote significant attention to cleaning fossil occurrence data prior to analysis, but such efforts are typically bespoke and difficult to reproduce. There are also no standardised methods to detect and resolve errors despite the development of an ecosystem of cleaning tools fuelled by the concurrent growth of neontological occurrence databases.2. As fossil occurrence databases continue to increase in size, the demand for standardised, automated and reproducible methods to improve data quality will only grow. Here, we present semi-automated cleaning solutions to address these issues with a new R package fossilbrush. We apply our cleaning protocols to the Paleobiology Database to assess the prevalence of anomalous entries and the efficacy and impact of our methods.3. We find that anomalies may be effectively resolved by comparison against a published compendium of stratigraphic ranges, improving the stratigraphic quality of the data, and through methods which detect outliers in taxon-wise occurrence stratigraphic distributions. Despite this, anomalous entries remain prevalent throughout major clades, with often more than 30% of genera in major fossil groups (e.g. bivalves, echinoderms) displaying stratigraphically suspect occurrence records. 4. Our methods provide a way to flag and resolve anomalous taxonomic data before downstream palaeobiological analysis and may also aid in the automation and targeting of future cleaning efforts. We stress, however, that our methods are semi-automated and are primarily for the detection of potential anomalies for further scrutiny, as full automation should not be a substitute for expert vetting. We note that some of our methods do not rely on external databases for anomaly resolution and so are also applicable to occurrences in neontological databases, expanding the utility of the fossilbrush R package.

Bayesian analyses indicate bivalves did not drive the downfall of brachiopods following the Permian-Triassic mass extinction

“…gr ., “” ) with identical primary and accepted generic names were discarded, while those with different names (i.e., the species was re-assigned to another genus) were retained. In addition, absolute ages (maximum and minimum ages) of the occurrences were updated based on ages in the Geological Time Scale 2020 93 using the chrono_scale function of the fossilbrush R package 94 . Next, occurrences with a high temporal uncertainty (>10 Myr) but not from an international stage were removed from the dataset.…”

palaeoverse: A community‐driven R package to support palaeobiological analysis

“…In tax_check, Jaro distances are calculated via the stringdistmatrix function from the stringdist package (van der Loo, 2014). This function is provided to help researchers perform a spell check on their dataset, with additional functionality available in the fossilbrush package (Flannery‐Sutherland, Raja, et al, 2022). However, it should be made clear that this is no replacement for thorough taxonomic vetting.…”

“…Today, palaeobiologists commonly use code to clean (e.g. Flannery‐Sutherland, Raja, et al, 2022; Zizka et al, 2019), analyse (e.g. Guillerme, 2018; Kocsis et al, 2019), and visualise data (e.g.…”

Paleoecology and evolutionary response of planktonic foraminifera to the mid-Pliocene Warm Period and Plio-Pleistocene bipolar ice sheet expansion

“…Species were assigned the speciation and extinction datums in accordance with Aze et al (2011) and Fenton and Woodhouse et al (2021), and all species occurrences located outside of these assigned stratigraphic ranges were removed. This range trimming was applied to eliminate much of the occurrence data likely attributable to misidentification and/or reworking that may create artificial "tails" within speciation and extinction data (Liow et al, 2010;Lazarus et al, 2012;Flannery-Sutherland et al, 2022). The trimming of taxa resulted in a dataset of 239 317 planktonic foraminiferal occurrences.…”