Geologic dating methods, saving a few exceptions like varve or tree ring counting, do not directly measure ages or time spans. Instead, the actual observation is typically a geochemical measurement, like a nuclide concentration or isotope ratio. Interpreting the measurement as a geologically useful parameter such as an age or rate then requires some sort of calculation and a variety of independently measured or assumed data such as radioactive decay constants, initial compositions or ratios, nuclide production rates, or nuclear cross sections (Fig. ). These elements form a “middle layer” between the direct observations and the geological information derived from the observations. Middle-layer calculations present a problem for management and analysis of geochemical data because they constantly change as the calculation methods improve and new measurements of the other parameters become available. Even though the geochemical measurements themselves in archived or previously published studies are valid indefinitely, the derived ages become obsolete. This is an obstacle for analysis of geochronological data collected over a long period of time or, sometimes, from multiple laboratories or research groups who have different approaches to middle-layer calculations, because any comparison requires repeatedly recalculating all the derived ages from source data using a common method. This paper describes a prototype computational infrastructure for dealing with this obstacle in one geochronological application – cosmogenic-nuclide exposure dating – that is intended to enable synoptic analysis of diverse data sets by making middle-layer calculations dynamic and transparent to the user.

Cosmogenic-nuclide exposure dating is a geologic dating method that relies on the production of rare nuclides by cosmic-ray interactions with rocks and minerals at the Earth's surface. As the cosmic-ray flux is nearly entirely stopped in the first few meters below the surface, the nuclide concentration in a surface sample is related to the length of time that the sample has been exposed at the surface. This enables many applications in dating geologic events and measuring rates of geologic processes that transport rocks or minerals from the subsurface to the surface or from the surface into the subsurface see review in. The most common of these is “exposure dating” of landforms and surficial deposits to determine, for example, the timing of glacier and ice sheet advances and retreats e.g., or fault slip rates and earthquake recurrence intervals e.g, .

The observable data for exposure-dating applications are (i) measurements of the concentrations in common minerals of trace nuclides that are diagnostic of cosmic-ray exposure, for example, beryllium-10, aluminum-26, or helium-3, and (ii) ancillary data describing the location, geometry, and physical and chemical properties of the sample. Interpreting these measurements as the exposure age of a rock surface is simple in principle: one measures the concentration of one of these nuclides, estimates the rate at which it is produced by cosmic-ray interactions, and divides the concentration (e.g., atomsg-1) by the production rate (atomsg-1yr-1) to obtain the exposure age (yr). It is much more complex in practice, because the cosmic-ray flux (and therefore the production rate) varies with position in the atmosphere and the Earth's magnetic field, and the production rate also depends on the chemistry and physical properties of the mineral and the rock matrix. Production rate calculations are geographically specific, temporally implicit (because the Earth's magnetic field changes over time), and require not only a model of the cosmic-ray flux throughout the Earth's atmosphere but also an array of additional data including atmospheric density models, paleomagnetic field reconstructions, nuclear interaction cross sections, and others. In addition, production rate models are empirically tuned using sets of “calibration data”, which are nuclide concentration measurements from sites whose true exposure age is independently known.

The middle layer for exposure dating, therefore, includes physical models for geographic and temporal variation in the production rate, numerical solution methods, geophysical and climatological data sets, physical constants measured in laboratory experiments, and calibration data. All these elements are the subject of active research: new production rate scaling models and magnetic field reconstructions are developed every 1–3 years, and several new calibration data sets are published each year. The result of this continuous development is that nearly all cosmogenic-nuclide exposure ages in published literature have been calculated with production rate models, physical parameters, or calibration data sets that are now obsolete.

It is unusual for middle-layer improvements to completely falsify or supersede the conclusions of previous research, but it is possible. For example, one common application of exposure dating aims to associate landforms deposited by glacier advances or retreats during the past ∼25 000 years with abrupt climate changes that occurred during that period e.g.,and references therein. Because some of these climate changes are separated by only hundreds or thousands of years, improvements in the middle-layer calculations that change production rate estimates by only a few percent can significantly change the correlation between climate events and exposure-dated landforms see, for example, discussion and examples in. Regardless of the application, however, any use of published data that are more than 1 or 2 years old, or any comparison of data generated at different times or by different research groups, requires complete recalculation of exposure ages from the raw data. As there are tens of thousands of exposure-age measurements in the published literature, this is a major challenge to the use of these data for any sort of synoptic research. This is important because many of the most valuable uses of exposure dating involve large, geographically widespread data sets applied to, for example, analysis of regional and global glacier change e.g., or analysis of ice sheet change and sea level impacts e.g.,.

At present, middle-layer calculations for exposure dating most commonly utilize “online exposure age calculators” developed by, e.g., , , , or that are online forms accessible by a web browser into which one can paste sample information and cosmogenic-nuclide concentrations. The web server executes a script that carries out production rate and exposure-age calculations, and returns results formatted so as to be easily pasted into a spreadsheet. The typical workflow for comparison or analysis of exposure-age data relies on manual, asynchronous use of one or more of these services, in which researchers (i) maintain a spreadsheet of their own and previously published observational/analytical data, (ii) cut-and-paste from this spreadsheet into an online calculator, (iii) cut-and-paste calculator results back into the spreadsheet,and (iv) proceed with analysis of the results. Although the ability to use the online calculators in this way to produce an internally consistent set of results has been valuable in making synthesis of large data sets drawn from multiple sources possible at all, this procedure creates redundancy and inconsistency among separate compilations by many researchers; relies on proprietary data compilations that are, in general, not available for public access and validation; interposes many manual data manipulation steps between data acquisition and downstream analysis; creates a “lock-in” effect in which the effort required to recalculate hundreds or thousands of exposure ages using one scaling method is a disincentive to experimenting with others; and makes it difficult and time-consuming to assimilate new data into either the source data set or the middle-layer calculations.

These disadvantages of the current best-practice approach of manual, asynchronous use of the online exposure age calculators could be corrected, and synoptic visualization and analysis of exposure-age data better enabled, by a data management and computational infrastructure having the following elements.

A data layer: a single source of observational data that can be publicly viewed and evaluated, is up to date, is programmatically accessible to a wide variety of software using a standard application program interface (API), and is generally agreed upon to be a fairly complete and accurate record of past studies and publications, beneath

A transparent middle layer is simply an implementation in which middle-layer calculations are fast enough to be performed dynamically and without any user effort at the time data are requested by an application in the analysis layer. The key property of this structure that a transparent middle layer makes possible is that only observational data (which do not become obsolete) are stored. Derived data (which become obsolete whenever the middle-layer calculations are improved) are not stored but instead calculated dynamically when they are needed. This eliminates unnecessary effort and the associated lock-in effect created by manual, asynchronous application of the middle-layer calculations to locally stored data by individual users and allows continual assimilation of new data or methods into both the data layer and middle layer without creating additional overhead at the level of the analysis application. Potentially, this structure also removes the necessity for redundant data compilation by individual researchers by decoupling agreed-upon observational data (which are the same no matter the opinions or goals of the individual researcher and therefore can be incorporated into a single shared compilation) from calculations or analyses based on those data (which require judgements and decisions on the part of researchers, and therefore would not typically be agreed upon by all users). The subsequent sections of this paper describe the ICE-D (Informal Cosmogenic-Nuclide Exposure-age Database) infrastructure, a prototype implementation of this concept.

The ICE-D transparent-middle-layer infrastructure prototype includes example implementations of all three layers in the transparent-middle-layer architecture. It consists of (i) a networked database server storing observational data needed to compute exposure ages, (ii) a networked Linux server that performs middle-layer calculations with MATLAB/Octave code used in version 3 of the online exposure age calculator described by and subsequently updated, and (iii) a web server that responds to user requests by acquiring data from the database server, passing the data to the middle-layer server for calculation of exposure ages, and returning observations, derived exposure ages, and some related interpretive information to the user (Fig. ). The effect is that a user interacting with the web server can browse and work with large data sets of exposure ages, originally collected and published by many researchers over several decades, without the necessity of managing the data set or repeatedly recalculating all the exposure ages using a common method. Data management and middle-layer calculations are transparent to the user, allowing focus on data visualization, discovery, and analysis.

The ICE-D prototype relies on cloud computing services available at low or zero cost from Google, Amazon Web Services, or other vendors; the current implementation uses Google Cloud Services (https://cloud.google.com, last access: 29 June 2020). The data layer is a MySQL database server provided by the Google Cloud SQL service. The middle-layer is a virtual machine on the Google Compute Engine service running CentOS 7 and the Octave code implementation of the online exposure age calculator, with a new API that facilitates programmatic use of the server. The web server that provides an example of a visualization/analysis layer is Python code running on the Google App Engine framework.

An often noted obstacle to participation in community data management infrastructure e.g, is the conflict between the broad, generalized incentive for an overall research community to develop centralized infrastructure and the immediate incentives of researchers who might, for example, view individually authored publications as more critical to career development objectives. The transparent-middle-layer model for data management has several features that could contribute to resolving this conflict. First, as discussed above, the separation of agreed-upon observational data from interpretive calculations or analysis makes the data compilation itself agnostic with respect to differences of approach or opinion among researchers, thereby reducing potential disincentives to participation in database development. Researchers with different approaches could simply develop different middle-layer and analysis-layer elements. Second, from the perspective of an individual researcher, the transparent-middle-layer infrastructure can make it substantially faster and easier to carry out time-consuming or difficult tasks (e.g., statistical analysis, generating statistical or graphical comparisons of new and existing data, comparing data with model predictions) that are required to achieve individual goals (e.g., writing successful proposals or publishing high-impact papers). In fact, more than 25 % of sample records in the ICE-D:ANTARCTICA database at this writing are unpublished data incorporated at the request of a number of researchers, and this may be evidence that the ability to use the analysis layer in tasks such as paper writing, proposal preparation, or sharing data with collaborators provides a positive incentive for user engagement with the project. User engagement with centralized data management systems should represent a trade – users provide a service to the community by making data available, and in exchange they are provided with services that help them to fulfill their own individual goals faster, better, and more easily. A transparent-middle-layer infrastructure can facilitate this exchange.