A decadal view of biodiversity informatics: challenges and priorities

Until the recent application of molecular technologies to biodiversity studies, almost all information has been labelled with scientific names. Names have a special significance to link information elements and as such, it is important to use them knowingly and to build tools that work with names [36]. As they reflect concepts that change between individuals and over time, names may refer to many different concepts, making them equivocal identifiers. In addition, information is often available only in local databases. The challenge is to find it, harmonise the way it is accessed and make it available in computer-readable formats. Nomenclature, taxonomy, taxa and their biology together constitute a large challenge requiring novel infrastructure and change of usual practices by stakeholders. Numerous initiatives exist to deal with these aspects but progress will require a common agenda to bring about a virtual infrastructure that will reduce the apparent diversity of web resources without reducing the diversity of services required by a diverse user community. While content about taxon names must be assembled by nomenclaturalists, taxonomists and managers of biodiversity information, there is an urgent need for vision-driven architectural and engineering solutions. The GNA’s (Global Name Architecture) [37] current priority is on name-strings (Global Names Index GNI [38]) and name-usage instances^l (Global Names Usage Bank GNUB [39]). The latter does not yet exist but will provide the essential semantic relationships (cross-links) at the nomenclatural level. This focus is entirely appropriate because universal coverage is tractable in the short to medium term. The resolution of names to concepts (see paragraph 3, below) is far more difficult and is likely to be intractable for universal coverage.

A long unorganised list of names is not particularly helpful. Since Linnaeus biologists have used latinised binomial names where the first part (the genus) is shared by a group of similar organisms and the second part, the epithet, differentiates between members of the group (e.g. oak trees belong to the genus Quercus that contains around 600 species). A similar hierarchical classification is followed for genera that are grouped into families, families into orders, orders into classes and classes into phyla. As science advances, however, these relationships change with greater understanding. While it is possible to build hierarchies from instances of name-strings, it is inefficient. The solution required includes a classification bank combined with a name list (see Paragraph 1) to produce a taxonomic hierarchy automatically for groups that have not recently received taxonomic attention.

The Species2000 / ITIS Catalogue of Life [40] is a global taxonomic reference system drawing on content from more than 100 sources. It provides a composite expert view on taxonomic information, providing an authoritative but mutable framework. Names within CoL represent concepts, but there is no link to the concepts themselves and therefore an identification cannot be unequivocally verified. Other classifications with names, such as NCBI taxonomy [41] or the WoRMS systems [42] can also be used as organisational frameworks. Yet each serves its own audiences, revealing the need for multiple systems that are however interoperable. Initiatives such as the Global Names Architecture (GNA) [37] promote the development of an infrastructure capable of linking available information about biological names. iPlant’s Taxonomic Name Resolution Service, TNRS 3.0 [43] corrects and standardises plant scientific names against particular taxonomies. ZooBank [44] is a new initiative to move the process by which new names become recognised into the digital age. Tools for alignment and cross-mapping of taxonomies can only be partially automated, since the domain knowledge held by taxonomists is very difficult fully to codify. Some projects such as i4Life [45] have developed tools that exploit characteristics of biological nomenclature to detect relationships between taxonomies, providing a useful "first draft" cross-map. Nevertheless to be authoritative, future environments must link nomenclators (like ZooBank, IPNI and Index Fungorum, Mycobank), taxonomic compendia (such as CoL and WoRMS), other classifications (a classification bank, perhaps), literature sources (describing species, their attributes, distributions, and common names), and phylogenies, covering the whole spectrum of biodiversity complexity. The taxon name is an access key, but it is essential that it can be linked to other resources, such as descriptions, traits and habitat. Ultimately names are the bridge to the accumulated information built over the past 300 years and trapped in the paper world.

The practical identification of an organism relies on the construction of a circumscription of the taxon to be identified, which in turn requires the examination of a range of specimens agreed to belong to the taxon. Before the digital era, the only ways to identify the name of an organism were to use a paper identification key or to consult an expert. New identification techniques have emerged to get to the name of an organism, including matrix keys and ‘smart keys’ that use the locality and time of the year to reduce the number of identification choices. These identification techniques however are labour-intensive and depend on experts to create the necessary circumscriptions, keys and the link to the list of accepted names. Automated identification techniques, like image recognition or in situ DNA analysis, are not yet sufficiently developed to be used routinely and reliably for most organisms. Identification keys always cover a small part of biodiversity and may also be difficult to discover. Developing morphological keys to all organisms is not achievable because no global organisation can establish a central repository, or even coordinate, prioritise and fund the creation of keys. The major priority therefore is to make the necessary descriptive data with their associated range and habitat information freely available. Services can then be created, for instance as 'apps' for the mobile phone market.

Networked services refers to the use of a resource directly over the web, so that one website may call another site for information necessary to carry out its function. Centralised services, on the other hand, concentrate all the associated resources at a single site. Although networked services are desirable to maintain consistency and to focus resources for maintenance (i.e. an authoritative master copy), in practice they are often unable to deliver the speed of response necessary for usability and creating a local copy (a snapshot of a dynamic resource) or developing an independent resource is often the only realistic solution available. Local copies are often the only practical solution for computation but there is no mechanism by which alterations to the primary resource can be effectively propagated to all copies. This inevitably leads to differences between copies. Copies should, therefore, be used over a short timeframe and, if necessary, refreshed. A feedback mechanism is required so that a data user can report an inconsistency to the data owner and, where a correction is suggested, can be easily incorporated into the original dataset. Automated workflows crucially require Web services but working with large datasets in networked services poses technical challenges in the ability to move large volumes of data, the provision of suitable search facilities that minimise the number of host-client interactions, and the bandwidth necessary to keep response times short. Centralised services, such as VertNet [51] assemble large collections of data in a common structure, submitted by individual data creators. The benefit that this brings is economy of scale and the ability to tune the performance of the system. In the context of data cleaning, for example, having the data centrally makes it much easier to compare across data sets and discover inconsistencies. The gain of economy of scale is very important since once a given type of error is identified, rules can be applied for cleaning across the whole data set, therefore avoiding overloading remote services. Once established, though, it is difficult to change the structure and change the purpose, but for large-scale data generation systems, logically centralised services offer significant advantages. One drawback, however, now receiving attention in the genomics and other fields is the issue of time taken to move large datasets to where the resources for computational analysis and modelling are located. Strategies are presently being considered for how to move computation to the data.

A great deal of high-quality software, services and resources have been created over the past decade, but much remains underused, even within the biodiversity informatics community itself. Many projects have relied on traditional routes to publicise their products, primarily through academic publications. It would be undesirable to impose standard applications or resources upon the community. Better to allow users to decide, to select which products best match their requirements. Projects should invest significant resource into marketing their products, engaging with real users and refining the product from user feedback, following the dictum of "release early and release often". Such marketing need not absorb a significant fraction of a project's budget but should be a clear strategy and an integral part of project management.

Open access data and services allow users to remain anonymous for some level of access. Some forms of interaction however, such as posting comments, corrections and some types of services, such as download, often require that users identify themselves. Social media tools like Facebook, Google and Twitter offer common 3rd-party authentication mechanisms that can be used for access control. This has two main advantages: first, it makes every resource easy to access; and second, it is a stronger security check compared to inventing a username and password for each site visited. Nevertheless, some resources will require a stronger form of authentication, for instance where payment is required. As a general principle, access to biodiversity data should normally be unrestricted except where it is essential to protect, for example, location data for rare bird nesting sites.

The scientific process requires the collection of observations from which hypotheses can be formed and, when necessary, more data to be collected to test them. Adding metadata represents an overhead on current practice but it is essential if data are to be discoverable and re-usable. Metadata are the key to discoverability and provide the context for linking resources. To improve current practices there is an urgent need for i) community agreement on metadata standards for specific purposes and ii) mechanisms to collect and append the necessary metadata, automatically whenever possible, such as the design of workflows that make use of standard services to create data-recording templates. In the short term, the extra effort of metadata production will have to be borne by the data producer, especially in the context of data journals, but tools to automate the production of metadata are conceivable, essentially eliminating the burden of production. A move to Linked Open Data is expected to obviate the need for enlarged metadata by making data more easily discoverable through concept linkage (see paragraph 15 et seq. below on Linked Data).

Appropriate biodiversity informatics tools will generate greater impact than is currently possible from the physical infrastructure of natural history collections, mesocosms, other experimental facilities, long-term ecological monitoring sites and genomics facilities, through much greater digital and on-line access to the facilities than is physically practical. This will enhance the sustainability of the infrastructure, since a large user base is critical for political sustainability.

Two relatively large surveys were conducted to understand how data are treated by scientists across different disciplines: by the PARSE.Insight project [60], with 1202 respondents, and by Science Magazine [61], with 1700 respondents, both with multidisciplinary international responses. From what researchers say about where they store and manage data, it can be deduced that data are not often shared openly. The results show that across all disciplines only between 6-8% of the researchers deposit datasets in an external archive of the discipline/research domain. The most common environment for storing, managing and re-using data is the lab and/or individual working environment, down to PCs and portable storage carriers. The category “server” is probably best understood as a file server of the research organisation behind a firewall and with restricted access for defined groups of registered users. According to Science Magazine, most of the respondents (80.3%) said that they do not have sufficient funding available for data curation. Other reports [4, 62–65] share more insight into data sharing practices by research area and highlight the importance of data sharing becoming normal practice.

The term ‘knowledge organisation system’ (KOS) covers any system for organising information, ranging from the traditional library subject headings to newer approaches like semantic networks and ontologies. Recognising the need for a standards architecture to provide basic interoperability across open systems, the TDWG Technical Roadmaps in 2006, 2007 and 2008 all identified community-supported vocabularies and ontologies, expressing shared semantics of data, as one of three required components; the other two are common exchange protocols and use of persistent identifiers for the data. The TDWG Darwin Core [73] glossary of terms is amongst the most widely deployed biodiversity vocabularies and both its management and relation to the TDWG Ontologies can be used as a model for other vocabularies. The GBIF LSID-GUID Task Group [74] highlighted the need for GBIF to identify sustainable support mechanisms for essential shared vocabularies and commissioned White Paper Recommendations on the use of Knowledge Organisation Systems. GBIF [75] separated the need for ontology management from the lifecycle management of flat vocabularies in such tools as BioPortal. The development, management and governance of such vocabularies remain a challenge for the biodiversity community. As concluded in paragraph 16 and discussed in section 4, the core technologies are available and well understood, but uptake by the community is not ideal. The challenge is to develop and deploy tools within the overall biodiversity informatics infrastructure that make the implementation of knowledge organisation systems effectively invisible. GBIF’s Integrated Publishing Toolkit [76] is one example of a step in this direction. Put simply, what would it take to make knowledge organisation work effectively and what would it achieve if it did?

Biodiversity informatics is inherently a global initiative. With a multitude of organisations from different countries publishing biodiversity data, our foremost challenge is to make the diverse and distributed participating systems interoperable in order to support discovery and access to those data. A common exchange technology, e.g. XML or JSON, may allow the syntactic exchange of data blocks, but both systems also need to understand the semantics of the data being delivered to process it meaningfully. If the data do not share a common reference model, then the exchange requires some brokering or other semantic processing (using tools described in paragraphs 3, 7, 12 and 16 above). For instance, the widely used standard Darwin Core is predicated on the occurrence (either a physical specimen or an observational record) as the unit of information, so is of limited value in the context of metagenomics for example, that may contain information about environmental function without mention of a named taxonomic entity, or information about communities of taxa. It is crucial that future efforts in this area take account of major global initiatives, especially GEO BON, GBIF and Genomics Standards Consortium, as well as novel approaches in eco-informatics, but it is likely that the data models used in these initiatives will also need to be extended [77]. Existing data must either be transformed in a semantically-aware manner to conform to such standards, or software that is aware of the semantic heterogeneity must work with multiple standards.

It should be straightforward to assemble a dataset on biodiversity and reach conclusions by linking available information. To understand and model processes, such as the phosphate cycle, requires information at the molecular level over seconds (solubility, diffusion and uptake), kilometre level over years (transportation and availability) and planetary level over geological time (mineral formation, extractability). The integration of all these data resources is necessary to model the cycle, from which policy decisions can be made for the time when cheap mineral phosphate (a fertiliser) is no longer available (in the next few decades) [2]. This example illustrates the complexity of the natural world, and how ‘grand’ is the challenge faced by biodiversity informatics to create a coordinated coupled modelling environment to address health, sustainability and environmental questions [78].

Science is, by its nature, a sceptical process. Data are received at face-value, examined and tested. If the user is satisfied, then the data will be applied. This process is crucial in biodiversity since information can rarely be generated by simple measurements. Concepts (like species), observations (based on human interpretations), proxy data (often originating from sensors) or algorithms (models fit for specific cases) constitute most biodiversity data with their inherent uncertainties and fuzziness. It is vital, then, that information about how the measurement was taken, to the minimum data standard, is included in the associated metadata. Judgement of quality involves an assessment of fitness-for-purpose and therefore cannot be an absolute measure. Data can of course contain both errors of fact, e.g. typographical errors, or errors of design, e.g. collecting data under a flawed methodology. Errors of fact can be detected by various means, e.g. duplicate entry or proof-reading whereas errors of design are more difficult to find automatically. A more significant problem is the accuracy of the data, meaning how precise and complete they are. In measurement it is accepted that a balance might weigh to the nearest 5 g, being a characteristic of the balance. In information terms, lacking a standard for generating the datum, it is harder. For instance, bibliographic citations can have diverse formats that humans can easily resolve to the same publication however computers, by and large, cannot unless given a PID as an information standard. The challenge for biodiversity informatics is to provide appropriate tools for data cleaning^m and to automate procedures for reading data for consistency [79], particularly against standard lists (see paragraph 16 above). Ultimately it is a case of caveat emptor. Users will develop trust in an information supplier and sites may wish to use a voting mechanism, e.g. similar to the supplier rating system on eBay. A system is required for data publishers to display comments from identifiable users (see paragraph 12 above), providing a feedback mechanism, essentially an open peer-review. Exposure to users is the best way to validate data.

Plummeting cost of hardware, increasing use of virtualisation and blurring between fixed / mobile computing and work / domestic environments for computing makes the prediction of preferred computing environments of dubious value. Compiling this white paper has identified no apparent need for bespoke ICT technologies. A continued use of a wide variety of platforms and approaches is to be expected. Biodiversity informatics has many requirements in common with other informatics domains and it is noteworthy that biodiversity research, as in other disciplines has the potential to produce very large and rapidly growing data sets from, for example automated digitisation, remote sensing and genetic sequencing. Although the configuration of existing and planned cross-domain infrastructure such as LifeWatch supports biodiversity informatics well, the domain will place heavy capacity demands on the computing infrastructure in the medium-term. Hardware associated with sensor and data logging is addressed in Appendix 2. Like other domains, biodiversity informatics will require robustness, stability and persistence, so will likely rely on key institutions with long-term funding. Over the core hardware infrastructure lies a spatial data information infrastructure, the biodiversity component of which is largely the topic of this white paper. The leveraging of information from distinct but adjacent domains will be increasingly necessary in the future, such as digital literature resources, image, environmental and climatic information databases. As molecular methods find ever greater uptake, one particular set of resources will become increasingly important to biodiversity informatics: these are the many biomolecular resources that, within Europe, lie within the purview of the ELIXIR infrastructure [80]. While many of the core resources themselves may be sustained with comparatively long-term support, the services built upon these resources must be configured to include biodiversity science use cases. A unified voice in specifying these use cases is required from the biodiversity community. Building the ‘social infrastructure’, however, is a major challenge: we have the technological capability but we need to increase its uptake by the community. For that we need to strengthen considerably the socially connected network of experts spanning the two communities: ICT and biodiversity science.

The world’s biological collections represent the hard core of biodiversity information. All other uses, from identification and naming onwards, are anchored in them. The collections contain an estimated 2–3 billion specimens but less than 10% have been catalogued in databases and much less captured as digital images [81, 82]. This means that more than 90% of the collections are essentially unavailable for use through the Internet. Manually digitising collections represents an effort estimated at up to one million person years, but, with today’s mass-digitisation methodologies, the task is feasible. As shown by multiple virtual herbarium projects [83, 84], the process can be partly automated through imaging techniques. With gazetteer services such as GeoLocate [85], georeferencing can also be computer-assisted. Another good example is the Volunteer site of the Atlas of Living Australia [86] whereby, when a backlog of digital images is available, their transcription is distributed through crowd-sourcing to a large number of volunteers. With help from initiatives like iDigBio [87], we envisage that distributed digitisation infrastructures will become essential parts of most major natural history collections and that dedicated services will be developed for outsourcing this task. A major challenge however is that collections still grow faster than they are being digitised (e.g. through endowments). As private collections must also be digitised by their owners, this requires a new suite of easy and inexpensive tools that can be deployed at large scale. To effectively deliver this research infrastructure service, digitisation requires prioritisation and its own funding channels.

Gathering information about the world around us has been a priority for biodiversity science for many years (see Appendix 3). Observatories will soon operate throughout the biosphere capturing different kinds of data over multiple scales, from microns to planet-wide, from parts of a second to years. It is very important to know the relative and also the absolute position of observed objects and events. This brings special challenges when observing the desired phenomenon and operating in extreme environments, such as the deep sea. The infrastructure for biodiversity data urgently needs more advanced informatics, support - not only mainstream ICT development but also the ability to deal with the specifics of biodiversity features and dataⁿ. It requires informatics to support observations, event detections, species identification, data transfer, storage, filtering and other kinds of data processing. New data-gathering tools that will allow new observatories at all biological scales and sensor networks covering the globe need to be designed, created and tested. There should be automatic processes allowing for feedback from data interpretation back to the observation or detection at site. This combination of techniques and related biodiversity informatics tools is expected to herald a revolution in biodiversity research, resolving much of our current fragmented data coverage and knowledge. Public-private partnerships should be encouraged to enter pre-competitive research and development in this evolution.

Developments in mobile communications offer numerous opportunities for innovation (see Appendix 2). Smartphones and tablet PCs with on-board GPS location can be easily taken into the field, creating opportunities both for innovative data collection and user information services. They are also particularly innovative for reference products such as identification keys^o. Apps like these can be used to generate image-vouchered, location-tagged observations uploaded to central databases^p. Performing science in large virtual communities, where participants have varied levels of expertise requires new techniques for data harvesting, processing, cleansing and validation.

Most biodiversity data that now exists are semi-structured and can be searched with typical search tools (Google, GBIF, etc.). However, these are often designed for use by humans rather than for automated data retrieval tasks and may have in-built limitations or constraints. To make better use of general purpose tools, users may need to use more specialised resources as well. GBIF, for example, supports retrieval by species name but the user may also need to use resources such as the Catalogue of Life to provide alternative names for species-based searching. The volume of data now being searched is so large that it is often not possible to refine keyword-type searches sufficiently to recover the needle buried in the haystack, especially in the absence of widely-used vocabularies. Contextualising information (establishing relationships between data elements) in a resource is possible^q, but currently difficult and slow. The implementation of PIDs (see Paragraph 7) would make the construction of metadata portals much easier. A mature search mechanism that contextualises rather than simply indexing would be far more powerful. A number of newly developed techniques exist, and some are under development, that make extensive use of visualisation methods to detect patterns and issues in data collections. These could be useful for quality and fitness-for-use assessment, especially in very large datasets such as the LTER-Europe data index or the GBIF index and taxonomical nomenclators. Data publishers need to go further in helping users find the data that match their requirements, with the use of PIDs, vocabularies and KOS (see paragraph 17).

In publications, either paper or PDF, information is often embedded in text blocks or tables in a way that inhibits its re-use. Semantic technologies (data mining) offer potential for liberating such data, but have not yet demonstrated the necessary flexibility or speed needed for broad uptake in the verbose, descriptive disciplines of taxonomy and ecology. Perversely, it is also difficult to extract information from highly condensed scientific writing such as taxonomic descriptions because this style of writing relies on implicit context in order to be understandable. New tools will be needed that use the vocabularies, ontologies and KOS described above to establish context between data elements, then to extract and assemble those elements into a format suitable for the user's purposes. Copyright held by commercial publishers remains a serious obstacle to recovering the non-copyright factual data. Some older publications that are not in copyright are being digitised, but serious issues remain over the rate at which the historic legacy literature is being captured, the completeness of the digitised literature and ease of access to this literature. Errors in the Optical Character Recognition (OCR) process mean that search of these archives will return an incomplete result set [88]. This presents an additional level of difficulty over and above the problems with extracting information from the born-digital literature.

For many analyses it is often necessary to aggregate data from several sources. Several data-aggregating initiatives have emerged in the last two decades for various areas of biodiversity informatics. Some of these initiatives were done on a project basis, while others were embedded in national structures, making them more reliable sources of information in the long-term. Examples include GBIF for primary data, Encyclopedia of Life (EOL) [89] for species descriptions, uBio [90] and Global Names Index [38] for names usages, CoL [40] for validated species names and Europeana [91] for multimedia resources. These data-aggregation initiatives have important beneficial side effects for biodiversity informatics such as enhancing the availability, standardisation and duplication (‘backup’) of data. Aggregation problems remain, similar for all these initiatives, such as hidden duplication, proper attribution, harvesting and storage. Each aggregrator has tended to solve these on their own by developing their own data provider network and internal infrastructure. Increasingly, though they are recognising the need to streamline and to avoid duplication of effort (for some examples see [18]). To facilitate integration, it will be valuable to develop a well-known catalogue of large-scale resources, with associated metadata, including a concept map.

When combining data from different sources and domains, interoperability is clearly not the only obstacle to analysing complex patterns of biodiversity. The accessible biodiversity data today come largely from repositories and individual researchers, and are generally of high quality with respect to reliability. The quality is rather low however, with respect to consistency. In other words the data aggregated today has been collected for different purposes and on different spatio-temporal scales leaving significant gaps in the assembled data sets and seriously hampering the analysis of complex patterns with data from diverse domains. Gaps can be filled by developing more comprehensive biodiversity observatory networks (BONs) and associated e-tools to support the collection, aggregation, and discovery of data from these observatories. There is also a fundamental need to re-consider what we already have – to ‘invert the infrastructure’ [92],p.20] – to re-examine and re-formulate existing data to make it more homogeneous, to remove non-biodiversity factors (e.g. compensating for differing observation technique) and to make it suited to the kinds of analyses we foresee for the future.

Virtual research environments (VREs), or Virtual Laboratories are online systems helping researchers to carry out their work. They include environments both to publish data (e.g. Scratchpads [93]) and to execute operations on data (e.g. myExperiment [94]) or both (e.g. AquaMaps [95] and iMarine [96]). VREs also include facilities to support collaboration between individuals. The challenge is to build integrative flexible e-Science environments using standardised building blocks and workflows, with access to data from various sources. Just as with physical laboratories, different kinds of VREs are possible, ranging from general-purpose to the highly specialised. A general VRE for wetland studies can be customised to a specific geographical area and populated with relevant datasets. A VRE specialised for a single scientific objective e.g. to find an optimal way of sequestering carbon in a forest would be equipped with workflows based on highly specialised simulation tools such as Biome-BGC [97]. For a successful uptake of VREs, they must generate immediate benefit for their users. For casual users the interface(s) must be simple and intuitive. For developers, there must be a usable pool of services and other resources that can be linked simply (e.g. BioVeL [98]). VREs must perform functions that people find useful. VREs as envisaged here, also act as social networking applications and have a central role in making some of the available technology described above usable or better, invisible to the majority of users.

Experience suggests that for an effective outcome in biodiversity informatics, a balance of top-down and bottom-up approaches is required. It is also important to remember that there is small benefit to asking end-users for their requirements, when they may not be aware of the benefits that new technologies can bring. The FP5 funded project ENBI (European Network for Biodiversity Information, funded 2003–2005) [99] concluded that a modular infrastructure could provide both the architecture and the sustainability to overcome the partial and ad hoc solutions developed in the past twenty years, and designed the LifeWatch infrastructure for biodiversity and ecosystem research (an ESFRI project [100]). This approach, followed for a decade, has not led to the development of trusted repositories or stable funding, and therefore has not generated user confidence. Thus far, all biodiversity information projects share a common problem, viz. how to keep the service running after the funding period. If people don't have confidence that an environment will last essentially indefinitely, or at least as long as it is relevant to enter information, then why would they invest their time and effort contributing to the common system? Paper publication is perceived, unrealistically, to last indefinitely and is the yardstick by which people judge any new approach. Publishing in PDF format is conceptually equivalent to paper, although it’s easier, faster and cheaper to distribute copies and its longevity has not been demonstrated. New paradigms are making data available in forms that can be readily re-used, e.g. Pensoft Publishers [56] and GBIF's Integrated Publishing Toolkit [76]. It may be possible that international organisations, such as GBIF, or large national institutions, such as natural history museums, will agree to underwrite data services on a care-and-maintenance basis, while the underlying software goes open source but with institutional oversight. Ultimately, to build the crucial user confidence, service managers need to invest far more than they have thus far in marketing to create new social norms.

The traditional citation system provides a mechanism to measure impact and provenance but it applies to a publication unit rather than code from a software library or data from a repository. Systems are required that will generate comparable metrics from the new open science resources (see paragraph 15). Data contributors would benefit by knowing the number of people and/or projects who used those data (impact). Code developers will benefit by knowing how widely their code is being used (impact). Users want to be able to drill down and find who wrote the code or generated the data (provenance). The ability to credit the data creator or code author is the primary basis for trust in the quality of the data or service. The following challenges need to be resolved: first, we need a system of attribution that is robust in a distributed network, easily achieved by the use of PIDs and author identities (see paragraphs 7 and 11). Second, licensing is poorly understood in the community, both by producers and consumers. For data, the flavours of Creative Commons licenses that involve "non-commercial" clauses make risk-averse consumers wary of using material, even when free use was the intention of the original contributor. For software, the terms of open source licencing and free-use are similarly subtle. In both cases, there is a widespread failure to understand the distinction between licensing and copyright. Third, copyright often creates a barrier to data use and re-use, although in academic work no instances of case law have been identified, so guidance is based on commercial publishing case law, predicated on financial loss. The wider Open Science movement is pushing hard to clarify this situation and biodiversity data should benefit from the increasing widespread liberalisation.

Career progression is enormously influenced by citation metrics as a proxy for impact and, more than anything else, this keeps us tied to a paper publication model. Products that users want, e.g. identification keys, are often used without citation and contribute nothing to career progression. People are too often not sharing their data freely, but save it for their close collaborators: they need to be given new tools that facilitate data sharing in the long run, but keep them in control while the research is still active. New metrics need to be defined that measure how often a data set is used and where conclusions based (in part) on those data appear. This through-tracking requires, at the very least, two of the fundamentals discussed above in Section 1, the use of PIDs to track the data and the development of a system to identify contributors (= authors). Ultimately, this is the single largest problem we face in persuading people share their data.