Oceanic Iron fertilization: a bibliometric analysis of the literature based on Natural Language Processing techniques

Since John Martin proposed the "Iron Hypothesis" in 1990 as a means of increased ocean carbon (C) uptake, much research has been done on the role of iron (Fe) in the oceans. Fe dynamics have been incorporated in biogeochemical modules of the latest generation climate models. The science of Fe Fertilization—the stimulation of ocean surface phytoplankton productivity by improved Fe bioavailability—continues to be of considerable research interest for paleo-theories of climate change and for the modern climate as an atmospheric carbon dioxide removal (CDR) technique.

We perform a scientific and bibliometric analysis of the Fe fertilization literature published since 1990 using techniques from the Natural Language Processing community not often used in oceanography. Our dataset is comprised of Web of Science titles, abstracts and keywords.

We start by a 'term class' frequency analysis for relevant science terms. We perform correlations among these term classes to track relevant oceanography issues: links in macro-nutrient and Fe uptake; Si:Fe ratio changes; efficiency of C export; and secondary side effects (e.g. increased dimethylsulfide or nitrous oxide, or shifts in phytoplankton groups).

We present a novel use of Latent Dirichlet Allocation (LDA), an unsupervised learning algorithm. LDA facilitates bibliometric analysis on a corpus of papers by automatically classifying articles into categories. We check our LDA results by manually categorizing the papers published into an emergent 4-category classification (in-situ experiments; climate model development; paleoclimate study; and CDR). We present priority shifts through time in the literature using the manual and LDA categories.

Since 1990, we find that discussion in the category of in-situ experiments steadily increases until 2010. Following the 2010 United Nations moratorium on large-scale Fe fertilization experiments we observe an increase in the fraction of documents discussing Fe fertilization as a CDR method. The frequency of papers in the climate model category increases with the availability of GEOTRACES data, which provides Fe measurements for validating new generation ocean models: Fe behavior in models remains poorly constrained. The proportion of papers categorized as 'paleoclimate' remains constant throughout the period studied.