Text mining carbonates: a record of dolomite fabrics through geologic time

Abstract:

The “dolomite problem,” the observation that dolomite, a ubiquitous mineral in deep time, does not form readily from modern oceans, has two end-member solutions. One hypothesizes that ocean chemistry varied significantly throughout Earth history and ancient oceans promoted dolomite precipitation. The other suggests that the abundance of ancient dolomite can be attributed to late diagenetic alteration of limestones. These hypotheses can be tested using petrographic textures: primary/early diagenetic dolomite is fabric-retentive, while late diagenetic dolomite is fabric-destructive. To quantify the importance of each of these hypotheses throughout geologic time, we use the machine reading infrastructure of geoDeepDive coupled to Macrostrat to construct a record of the prevalence of various dolomite fabrics since ~2 Ga. Using text mining and natural language processing of scientific literature, we rapidly determine the age, location, mineralogy, and petrographic characteristics of rocks that have been described in published papers; we focus on dolomites and related sedimentary facies (e.g., stromatolites, evaporites). In a North American dataset, we find no statistically significant change in the relative abundance of dolomite across the Precambrian-Cambrian boundary, despite evidence that Precambrian dolomites are generally more fabric-retentive than their Phanerozoic counterparts. The co-occurrence of stromatolites and dolomite in carbonate units also begins to shift around this time: while >50% of Proterozoic dolostones contain stromatolites, only ∼20% of Paleozoic dolomitic units do. These results suggest that, although the relative abundance of dolomite was constant across the Precambrian-Cambrian boundary, the mode of dolomite formation likely changed in the early Phanerozoic.