Structuring five decades of Antarctic biogeographical records from scientific literature

Comprehensive records of Antarctic biogeography (i.e., where biological organisms exist) are fundamental to systematic conservation of Antarctic biota and understanding how their distribution may be altered by climate-driven environmental change.

Scientific publications on physiology, genetics, biochemistry, and other aspects of Antarctic biota often contain accidental biogeographical information (e.g., sampling location for specific taxa in the material and methods section). However, Antarctic biogeographical databases typically don’t capture this vast collection of biogeographical data that exists in the scientific literature as unstructured information.

Leveraging recent developments in machine learning, the Evolving Biogeography Register (EBR) was developed to fill this gap. Broad search terms were used to retrieve articles that may contain accidental or explicit biogeographic data. Such broad search terms allow unmatched sensitivity for relevant articles, but also retrieve a vast number of irrelevant results. To address the latter challenge, the EBR informatic pipeline uses machine learning algorithms to predict whether individual articles are likely to be relevant and to identify biogeographical records.

In other words, the EBR informatic pipeline helps focus human annotators on articles that are most likely relevant and directs them to plausible biogeographical data in those articles. Consequently, EBR generates biogeographic records that are at least as reliable as those obtained through conventional literature review.

The initial register contains 25,000+ unstructured biogeographical records from more than 250,000 scientific articles, but was limited to the Ross Sea region. There is an opportunity to work with international collaborators to use the EBR informatic pipeline to generate an unprecedented collection of biogeographical records across the whole of Antarctica.