5.1 Development of a Citation and Usage Tracking System for Greenhouse Gas
ICOS, Integrated Carbon Observation System, is a pan-European research infrastructure with a mission to provide standardised, long term, high precision and high-quality observations on the carbon cycle and Green House Gas (GHG) budgets and their perturbations. The ICOS observing network consists of over 140 observation stations, each related to one or more of the three domains Atmosphere, Ecosystem and Ocean, and operated by its (currently 14) member countries. The collected data is processed and quality controlled at Thematic Centres (one for each domain), before being openly distributed via the ICOS data centre named Carbon Portal (CP).
The ICOS Carbon Portal is designed to manage and/or distribute data objects of a number of different categories. All data objects are assigned a Handle System-based persistent identifier at the time of ingestion into the Carbon Portal repository. The PID of these individual data objects can be resolved via e.g. the handle.net resolver service, which redirects to the object’s landing page hosted by the Carbon Portal. The landing page lists the most relevant metadata of the object, including a direct link to access to the data object.
Objects can be registered with DataCite, either as single objects or as collections of data objects. This means that they are also assigned a DOI and that associated metadata are stored in the DataCite catalogue following the DataCite Metadata SchemaFootnote 6. This allows the data objects to be found also through searches on the DataCite portal, and also provides full integration with the Citation Formatter service from Crossref and DataCite.
Resolving the DOI, e.g. via the resolving services of handle.net and the DOI Foundation, results in a redirect to the landing page of the object or object collection, hosted by the Carbon Portal. The DataCite DOI identifiers have the form “10.18160/<suffix>”, where “10.18160” is the ICOS-specific DataCite prefix, and <suffix> is a globally unique string. (The strings used for DOIs are also computed starting from the data object hash sum, but are designed to be shorter and easier to read for humans. As in the case of the Handle PIDs mentioned above, additional tests for uniqueness are of course performed before submitting the DOI registration request.)
Any updated versions of a given object, for example dynamic, growing time series or corrected data sets, are assigned a completely new PID. In order to provide unbroken provenance chains, the metadata record of the old version is updated with a link to the superseding object, and vice versa. This strategy is applied to data objects of all types described above.
The Carbon Portal provides a landing page (as shown in Fig. 3) for any digital object described in the metadata store – including data sets, but also stations, data type specifications and concepts. All landing pages are created dynamically, i.e. at the moment that their URL is accessed. This means that the displayed information always reflects the current, most up-to-date information. The format, and what information is shown, will vary between the type of objects, with the richest content provided for data objects.
ICOS data is distributed using a Creative Commons Attribution 4.0 International (CC BY 4.0) licenseFootnote 7, which users have to accept before they can access the data. The CC BY license requires end users to give appropriate credit (i.e. citing data when it is used), provide a link to the license, and indicate if changes were made when re-distributing the data. When citing ICOS data, at a bare minimum the data object’s persistent digital identifier should be given in a machine-actionable form (as a HTTP URL, for example http://hdl.handle.net/11676/6PrNhZelwXKHLqO41QRsbheu or https://doi.org/10.18160/RHKC-VP22); this minimal form is sufficient for inclusion in provenance records, but for use in scientific literature much more information (contributors, data set name etc.) is of course required.
5.2 Facilitating Quantitatively Correct Data Usage Accounting
In a world of open and free data sharing, it is often necessary to document the use of data products and give this as a quantitative merit to data producers and providers. Since all entities involved in the data production chain face the challenge of having to find sources of continued funding for their efforts while “selling” their data is not an option. They need to justify to funding agencies and users the relevance of their observations and contributions to data production.
An existing analogy to such a use-based merit system are scientific journal publications, where authors receive merit based on the number of “uses” of the article, i.e. based on the number of citations. Journals are selected by authors and institutions based on the aggregation of those citation scores in recent years, i.e. how visible the result becomes by using a given publication channel. However, aggregating scores, i.e., citation numbers accumulated across repositories, may be difficult to compose if data is stored at different granularities in the different archives.
By analogy to scientific articles, persistent identification of data by Digital Object Identifiers (DOIs) would be a crucial element of such a service for quantitative accounting of data use. However, at least 4 challenges exist:
DOI granularity: This would make usage numbers based on a fine granularity biased in comparison to data identified with a coarser granularity.
Data collections: DOIs can refer to a user defined collection of other datasets, which themselves may be identified by DOIs. The data collection approach makes data very convenient to cite. However, the contribution of different data producers to such a collection can vary significantly.
Accounting mechanism: Indexing agencies will or have been setting up services for counting of use events involving (DOI) identified data. From here, services need to be implemented that break down data use events into the contribution of single data producers, and with a fixed granularity allowing comparisons between data producers.
Nature of data use events: Scientific data can be used in many different ways, e.g. illustration for outreach purposes, trend analysis, constraining models of environmental processes, event analysis, just to mention a few, and data can be accessed once or multiple times for the same use case event. A list of data use types counted towards use accounting, including weighting factors, would typically be agreed on and continuously updated by a cross-domain working group consisting of experts on data production, data management, and data indexing.
In order to meet the challenges, and to work towards implementing accounting services for data use, the project team defined the following tasks in the early project phase:
Data Identification with Homogeneous Granularity in Primary Archives
These DOIs, in this context called primary DOIs, would be used as reference for setting up data use accounting. The ambition in the early project phase was to achieve homogeneous granularity, and thus comparable data use metrics, across repositories and RIs. During the implementation, it turned out that the goal of achieving homogeneous granularity of primary data identifiers across atmospheric RIs was too ambitious. Data products are simply too different in nature among repositories, sometimes even within a single RI. However, primary data identifier granularity should be homogeneous at least within one repository, preferably comparable also among repositories of a single RI.
Transparent Data Accounting When Using Data Collections
When identifying data in larger studies, e.g. global climatology of atmospheric parameters, using primary DOIs, requires quoting hundreds of DOIs, which would be rather inconvenient. The DOI specification provides for coining DOIs for user specified data collections, which are ideally suited to identify data used in larger studies. Ideally, the references would also include further provenance information in order to identify and acknowledge contributors to the data product used.
By interacting with the relevant RDA working group on research data collectionsFootnote 8, this work resulted in a fully finished recommendation for issuing and handling persistent identifiers for data collections that meet the requirement of referencing back to the primary identifiers of the data contained in the collectionFootnote 9.
Performing Correct Accounting of Data Use
For scientific publications, accounting of use is performed by the indexing agencies. If they offer a similar service for data, it needs to be assured that references to collection identifiers are resolved to the primary identifiers to ensure correct accounting of data use. The task involves a dialogue with the indexing agencies to implement this policy. A dialogue with DataCite as an indexing agency collecting use events involving data DOIs revealed that indexing agencies show little willingness to resolve references to primary identifiers contained in collection DOIs when accounting for data use. From the indexing agencies perspective, this approach makes sense due to the issue of heterogeneous granularity of primary data identifiers across or even within domains. As a result, the task needed to be modified. The service of calculating metrics for data use is moved from the indexing agency to the primary data repository. Based on its own primary DOIs with homogeneous granularity, the primary repository can access data use events stored at the indexing agency, resolve references in collection DOIs, and thus calculate data use metrics comparable across the repository. A prerequisite for this approach would be machine-to-machine access to the indexing agencies data holdings by the primary data archives. A dialogue about this is ongoing and needs to be continued in the near future.