Set in stone ? Improvements in stratigraphic data processing and storage

Transparent data reporting is a crucial aspect of an integrated and reproducible stratigraphic approach. However, there is currently no set standard for doing so. Typically, lithological information can still be found as simple images in publications, rather than as importable vector graphics, or better yet, tabulated data. Another problem is that there is often confusion on stratigraphic depth, composite depth and interpreted age: there is no standard format for these parameters, and the way they are obtained is often difficult to replicate. This all impedes the compilation of published data into reliable stratigraphic databases. We suggest that standardised stratigraphic data formats and associated automated tests can solve this issue. We explore this in [1] the case of the quantified data contained in lithological logs, and in [2] age-depth conversion models, or depth-depth models for correlation between sections or for splicing.

We define three types of numerical data in lithological logs:

• Bed nature and thickness (including thickness variations)
• Bed profile (to convey hardness, weathering, grain-size, facies, etc.)
• Discrete feature occurrences (e.g. fossils, minerals, sedimentary patterns)

These data types are not defined based on geological arguments, but on the way they are digitised. Respectively, these are interval data, continuous time-series, and discrete data. Therefore, three data sub-formats can be rigorously defined, and serve as building blocks for a larger single data format that would be comprehensive of all numerical data found in lithologs. Based on this, the existing StratigrapheR package has been updated (version 1.3): it now offers a formalised way of documenting these types of information for lithological logs in a quantified way. StratigrapheR is freely available for R at https://CRAN.R-project.org/package=StratigrapheR.

In StratigrapheR, discrete feature occurrences can be attributed to collections of symbols. Bed profiles can be formatted as time series, and “welded” to the side of lithologs. Layer content can be set as data tables for each bed, and can be illustrated by the colours or patterns of the beds in the litholog. Variations of thickness at the boundary of beds can also be “welded” to these beds in lithologs, as long as the variation does not interfere with the profile’s time series. Using a minimal set of formal rules, all these quantified data types can thus be standardised. We suggest that these concepts can be the basis for universal formats of numerical lithologs.

For age-depth and depth-depth conversion models, we explore the reversibility aspect, i.e., the ability to retrieve the initial signal from one that has been transformed, e.g., correlated, spliced, or tuned. This involves taking the precision in depth and age into account, as well as formalizing hiatuses, and preserving relevant features of the original data/signals in all subsequent data files, such as parts of the signal commonly removed for processing (e.g., turbidites).