Information about the distribution of biological organisms is used to investigate many current global health and human services such as clean water, land preservation and restoration, disease prevention, food safety and security, agricultural pests, drought, land use management, urban planning, and the effects of climate change (Kearney et al. 2014, Sinervo et al. 2010, Arnaud et al. 2016, McGeoch et al. 2016,Nature Publishing Group 2008,Department of Primary Industries and Regional Development: Agriculture and Food 2018). Researchers use locality information associated with vouchered biocollections specimens to evaluate the biological requirements for organisms by linking its presence with many other kinds of data (i.e., climate layers, satellite imagery, and other data resources) (Kearney et al. 2014, Sinervo et al. 2010). In general, researchers create, retrieve, combine, assess quality, clean, and visualize geospatial data before they apply their research methods, such as gap analysis (e.g., taxonomic or geographic, Ariño et al. 2016), species distribution models (e.g., MaxEnt, Anderson et al. 2016) or International Union for Conservation of Nature (IUCN) conservation assessments (Brummitt et al. 2015). Locality information is found on the labels, field books, logs, photographs, etc. that accompany and describe specimens in biocollections. Today, researchers apply geocoordinates immediately when an organism is collected, but before we had readily accessible and reliable Global Positioning System (GPS) units, only a text description of the locality was provided. To make these older specimens useful in modern research practices, geocoordinates (i.e., latitude, longitude) are retroactively applied to a text description of an organism's locality in a process referred to as "georeferencing."

Georeferenced locality data are widely made available through biodiversity data aggregators such as Atlas of Living Australia (ALA), the Global Biodiversity Information Facility (GBIF), Integrated Digitized Biocollections (iDigBio), and VertNet in a digital format. Digital data records for more than 115 million physical specimens have been shared (iDigBio 2018), leading to large amounts of digitally available locality data about organisms. Of those 115 million records, 59 million have coordinates. For example, the points on this map in Fig. 1 indicate geocoordinated specimen location data available through iDigBio for beetles in the genus Cicindela (Coleoptera) from California, USA, with each point representing one or more specimen records found in a collection (Fig. 2).

Figure 1.  

Map created using SimpleMappr (Shorthouse 2010) that illustrates geolocated specimens for Genus=Cicindela in California as found on iDigBio.

Figure 2.  

This specimen record is an example from the University of California Collection Network Symbiota Portal. The large image is an edit of the record to include a medium size version of the image for easier viewing in this article. The portal software is open source and it is freely available for reuse through the Symbiota GitHub repository. The image is an example of a specimen record that includes an image of the specimen with label data. The image is contributed by the UCSB Invertebrate Zoology Collection at the Cheadle Center for Biodiversity and Ecological Restoration. The usage rights for the image is Creative Commons 0 (public domain).

The data captured from biocollection specimens are typically managed by collections staff who care for specimens and digitize the data associated with those specimens, ultimately making the data available for research. This creates a scenario where the people creating the datasets are often not the same people who use the data in research, which creates a need to provide venues of communication and foster understanding between these stakeholders about the implications of each other’s' methods on research products (Tenopir et al. 2015,Zimmermann 2008). Geospatial and biocollections communities benefit from mutual conversations that inform the work of both biocollections staff and the research community, and through exposure to computational technologies, data, and software for working with spatial data. Training aids are specifically needed for biocollections staff to clean and visualize these data to assess its reliability and precision, and they also need to evaluate the data quality feedback that is provided to them after sharing their data with aggregators, which may include issues in the geocoded locality data. Researchers who plan on using the data that is produced need to understand the process of producing the datasets in order to evaluate them for their particular use.

Recognizing the value of biocollections for research, education, and society, a diverse group of scientists outlined a coordinated effort they envisioned as a Networked Integrated Biocollections Alliance (NIBA), which resulted in strategic and implementation plans for the digitization of US collections information (Hanken 2013, Chapman and Wieczorek 2006). In response, in 2011 the US National Science Foundation (NSF) instituted the program Advancing the Digitization of Biological Collections (ADBC). As part of the ADBC, NSF funded the Integrated Digitized Biocollections (iDigBio) 10 year project as the hub for US non-federal collection digitization efforts. iDigBio's missions are to support efforts by biocollections to digitize, mobilize, aggregate, and provide access to biological specimen data for biodiversity research and education globally. Museums holding biocollections collaborate to form Thematic Collection Networks (TCNs) and seek NSF ADBC funding to support digitization of select collections to meet targeted, specific research needs and share this mobilized data via iDigBio.

As part of this initiative, the iDigBio Georeferencing Working Group (GWG) was formed to support the biocollections community to implement best practices in the improvement and maintenance of critical location data. The GWG benefits from the contributions from many, including GEOLocate, VertNet, TCNs, and earlier projects (e.g., Georeferencing.org, MaNIS, FishNet2, ORNIS and HerpNET).

In response to biocollection georeferencing needs, the GWG offered two Train-the-Tainers workshops in 2012 and 2013. These workshops were aimed at training biodiversity and collections professionals to use best practices for georeferencing (Bloom et al. 2017), to mobilize biocollection occurrence data, and to encourage those participants to share what they learn with others for the benefit of the collections community as a whole. The primary audience for these workshops included the collaborative museum and herbaria ADBC TCNs staff, collection managers, and curators engaged in active transcription and georeferencing of collections data. An additional workshop, Field to Database (F2DB) (2015), targeted researchers and the collection of new, future biocollections data. F2DB incorporated current best practices for creation and sharing of georeferences and locality information to be born digital, that is mapped to data standards, if possible, in electronic format and georeferenced at the time of collection (Online Computer Library Center (OCLC) 2018). This process results in higher quality data, mobilized more efficiently, and avoids adding to the legacy pile of text locality strings to be data-entered and georeferenced. The Biocode Field Information Management System (Deck and Ewing 2016) is an example project supporting the generation of born digital biodiversity data. Other workshops, such as the joint Synthesys-iDigBio: Digitization Software Training Workshop, have similarly shared tools for biocollection data generation and standardization.

Training provided by the iDigBio project initially focused on community-derived digitization best practice discovery and documentation but has increasingly incorporated innovative use of scientific collections data for research into workshops and symposia design. The iDigBio Georeferencing Working Group (GWG) saw the need to offer a georeferencing workshop that would combine best practices for historical and new data with new lessons on evaluating the resulting biocollections data for their fitness for research and other downstream use.

The major objectives of the workshop were to gain feedback from the participants on research data needs for the future and to enhance and improve the quality of collections geospatial data. Objectives of the 4-day workshop included:

• Demonstration of tools (hardware and software) and geospatial data standards, especially as relating to the Darwin Core standard.

• Discussion of best practices for data repositories (e.g., obstacles and minimization of data loss).

• How to evaluate already georeferenced data for quality, or the fitness of the data for a specific use (fitness-for-use).

• Current tools for visualization and evaluation.

• Introduction of open-source QGIS software and selected plug-ins to participants to demonstrate data visualization methods.

• Sharing best practices for researchers for in-the-field creation of new locality data.

• Gain insight into the challenges faced by researchers for georeferencing through shared research experiences using biocollection geospatial data.

• Gain insight into the challenges data and collection managers experience in generating and managing georeferenced data.

• Gain input from participants about their needs and learning experience during the workshop.

• Gain an understanding of how participants use their learning for research and other present and future research and curation of collections and data needs.

The results presented here represent views from our participants (23 persons) who were selected for their interest in georeferencing and biocollections at the time that they registered for the workshop. All but one of the participants were affiliated with U.S institutions. The participants represented a cross-section of users of biocollections data that include students (graduate and undergraduate), professors, collections managers, curators, an agriculture specialist, and data managers.

Workshop participants self-identified as a researcher, data manager, GIS expert, or some combination of identities. The needs of each of these groups for data management and georeferencing skills were similar, yet each group had particular goals. For example, researchers were interested in the functionality of OpenRefine software for standardizing biodiversity data sets, new tools for efficient field data collection, such as the Biocode Field Information Management System, and efficient georeferencing of large datasets with high-throughput analyses of specimen coordinate accuracy. There was a strong desire for efficient and accurate georeferencing of biocollections to create "research ready" data that is available for complex, data-driven analyses (Seltmann et al. 2017). Developing the skills available through OpenRefine for data cleaning was expressed to be of great benefit for basic cleaning of collections datasets prior to publication and ingestion by data aggregators such as iDigBio and GBIF. Data visualization tools, such as QGIS, for determining the accuracy of geopoints were perceived to be of great benefit to researchers and data/collections personnel alike. Researchers were more interested in more advanced applications, such as the use of R and Python scripting due to the desired downstream applications and analyses.

Workshop participants with a strong GIS background were primarily interested in learning more about the current practices of researchers and data managers in data cleaning, georeferencing, and spatial analysis to identify areas of opportunity for improving the ability of domain experts to assess data quality and gain new insights to visualize their data spatially. The QGIS tutorial was modified in response to the interest in new skills expressed by participants in the first two days of the workshop. For example, a desire to know how to leverage coupled spatial data layers to perform a spatial selection and drill down through several layers of information to choose records that met a specific criterion, was of great interest to the participants. Also, learning how to subset data, edit record entries (e.g., coordinates) using QGIS, and save the cleaned and appropriately georeferenced records as comma-delimited (or comma-separated) text files (CSV) was of great interest and added to the tutorial.

Georeferencing among data managers and researchers in biocollections generally use the point-radius method (Fig. 3) to account for coordinate uncertainty (Chapman and Wieczorek 2006), although polygons are also commonly used for narrow geographical features (Fig. 3). In the geospatial community, bounding boxes are often used to delineate the extent of resources rather than points with a given radius. Additionally, many field record entries reference named places which can be reconciled against a gazetteer. Participants learned how this presents an opportunity for exploiting existing geometries to use as a record boundary from existing sources, such as OpenStreetMap's Application Program Interface (API), or shape files for publicly available datasets.

Figure 3.  

An illustrative example of the two methods of uncertainty capture when georeferencing specimens. Method A, or polygon, creates a shape around the river (in blue). Method B, or point-radius, creates a circle of uncertainty around the origin. The illustration is based on output from GeoLocate software (Rios 2018) for both polygon and point-radius.

Researchers and data managers were taught how to import existing georeferenced collections into software such as QGIS. Once spatially referenced, it is possible to couple the georeferenced data with existing layers of spatial information that can supplement assessment and analysis. Some layers of interest included administrative boundaries (for checking centroids, or center points of polygons, against existing records), ecoregions, cultivated gardens and zoos, and parks and protected areas. Other researchers expressed interest in acquiring and incorporating data about elevation, federally managed sites, and climate data. A conversation about where to find data resources, like local municipality GIS office websites and ArcGIS Online’s Living Atlas of the World, proved helpful.

On the first day of the QGIS tutorial, the topics covered included importing tutorial data obtained from iDigBio, adding additional layers, and saving a map project. The following day included advanced topics, such as performing a spatial join across layers to query attributes, sub-setting the dataset based on spatial selections and intersection, changing the symbology of the data, and performing summary statistics on the attributes. Additionally, visualizing the data over time and producing heat maps of observation locations provided new views on trends within the dataset that would be impossible to detect by simply viewing the records in a table.

Reactions from the data managers and researchers about the QGIS tutorial was positive. Many were excited about the prospect of incorporating QGIS and other spatial analysis techniques into their georeferencing workflows. While fewer participants indicated interest in using QGIS for data cleaning, many expressed interest in using the techniques for data exploration and public communication. Participants who brought their own datasets to the workshop spent time on the last day running similar analyses against their collections in QGIS and were interested in learning how to perform specific GIS operations to spatialize their research questions. For example, several participants were interested in assessing the spatial distribution of specimens in their study area to find a region with a greater significant number occurences. Others were interested in testing predictions about regions where the specimens were not likely to occur (e.g., studying absence) and using other spatial information to account for the trends in distribution.

Desired skills to develop included coupling QGIS with Python scripting to enable researchers and data managers to work with larger datasets. Researchers who also wanted to incorporate QGIS into georeferencing workflows were interested in plugins such as QGIS Gazetteer Plugin that would allow them to cross-reference observations and flag records for removal from further analysis. Best practices for editing CSV files in QGIS were also desired, and a lack of current consensus surrounding when to remove a point from a dataset resulted in inconsistent practices surrounding editing of data using QGIS. For example, once a point has been flagged because it corresponds too closely to a county centroid, should it be removed from the dataset or can its position be rectified? Should the record be noted but not included in analysis? Ideally, the modified and annotated research dataset would be published with original identifiers to enable linking of the data to the original data record and the physical specimen.

Many researchers and collections managers expressed interest in a follow-up tutorial on more advanced techniques in QGIS and were highly motivated to incorporate QGIS into their workflows upon completion of the tutorial. Interest in using QGIS for georeferencing tasks, data exploration, and public communication indicated that planned future tutorials could help users apply more advanced techniques to their existing research or collections datasets.

We gathered data from the participants using pre, post and follow-up surveys. Results from these were used to gauge enthusiasm from participants for the workshop and the direction participants thought future training should progress. The pre-workshop data was collected online through the workshop application using a google forms (Suppl. material 2). The post workshop (Suppl. material 3) and follow-up surveys (Suppl. material 4) were distributed electronically via Qualtrics Survey software licensed to the University of Florida. The application form, post-workshop surveys, and follow-up survey questions and protocols were reviewed and approved by the University of Florida Human Subjects Committee Institutional Review Board (IRB) (University of Florida IRB201601849).

Key conversations at the workshop centered around capturing historical versus future biocollections data. All participants, data researchers and data providers noted changes occurring in biocollections data workflows, yet outstanding challenges continue. For all of these reasons and more, there exists a need for ongoing initiatives to address these topics.

• Significant quantities of specimen data collected remain on paper specimen labels, in notebooks, catalogs, and field cards, limiting online open access and resulting in patchy datasets.
• At the same time, new specimens are still being accessioned into biocollections in solely paper-based data formats, thus further contributing to a largely inaccessible backlog of biodiversity specimen data.
• Backlogged legacy data is labor-intensive, expensive, and sometimes difficult to georeference, and often has significant uncertainty.
• Not all georeferences (new or legacy) come with the metadata needed for researchers (or algorithms) to effectively evaluate geospatial data fitness-for-use.
• Many collections undertake the time-consuming process to provide uncertainty data using point-radius or polygons, however, these data seem to be underused in scientific analysis.

Collections and researchers need workflows that reduce unnecessary future collections data management and speed access to data. On this topic, our discussions focused on best practices for creating new georeferences when collecting specimens, georeferencing existing collection locality data, and methods for use and evaluation of geospatial coordinates from historic biocollections records. Optimistically, technology is making it possible for new specimen data to be born-digital (August et al. 2015) and biocollections are beginning to develop policies and subsequent workflows (Nicole Fisher, Digital Collections and Informatics, National Research Collections Australia (NRCA), Personal Communication, August 2018) to prevent the creation of an even bigger collection legacy data backlog (La Salle et al. 2016, Young 2015). There is a push to capture coordinates and other relevant data upfront using mobile devices and other methods, saving time, money and providing valuable, much-needed, more accurate data, faster (Morrison et al. 2017). Born-digital data has the potential to speed access to georeferenced specimen data, although those using these data will still need to evaluate usefulness and accuracy as born-digital does not necessarily equate to high geospatial data quality.

Finding materials and expertise can be challenging and time-consuming. Development of new tools and workflows continues. Participants appreciated consolidated resources like those found at iDigBio and the Biodiversity Catalogue. Aggregation of this information saves time and effort. Some topics and tools covered stood out from the rest for all participants. These include: template generation for data collection, such as the Biocode Field Information Management System, application development with Open Data Kit (mobile-first data collection platforms), and discussion and development of a data validation process (Suppl. material 7).

Hands-on interaction with new software tools and APIs for data refinement, along with in-person interactions enabled by the workshop improved the efficiency of the learning process. Sharing standards of practice and needs across disciplines highlighted the need for changes and transparency in future data collection, evaluation, and data sharing processes.

Our workshop experience suggests the need to improve creation, use, and publication of uncertainty data. Whether expressed as a polygon or point-radius, this information appears to be underused and quite time-intensive to create. The biocollections and research communities are encouraged to utilize best practices for creating and sharing coordinate uncertainty information, and the collections community need research examples using these data in order to justify such effort. This may require future training in the research use and value of uncertainty data.

Data aggregators like GBIF, Atlas of Living Australia, and iDigBio are working through the Biodiversity Science and Standards TDWG/GBIF Biodiversity Data Quality Interest Group to harmonize biocollections data quality (DQ) feedback. Aggregators provide these standardized DQ assertions: 1) to those browsing the data online, 2) to the data providers, and 3) in any downloaded datasets. However, this DQ process, along with the tools, methods, and data standards used during data capture and publishing are not always understood in the research community. For instance, data aggregators currently use taxonomic name assemblages to facilitate searching and indexing of aggregated data. But as a result, the taxonomy may not be what some expect or need (Mesibov 2018,Franz and Sterner 2018). These issues and others, can affect data use and the impressions of the data value (Maldonado et al. 2015,Mesibov 2018). Lacking the understanding of the DQ process can make it difficult for researchers to evaluate and use the data, and make it tricky for researchers to offer fruitful feedback. This issue is now a priority of the TDWG Biodiversity Data Quality Interest Group which recognizes this barrier to the use of collections data and is actively working to resolve some of these issues.

Some possible actions to advance and speed up the use of biocollection data in research are listed here, and the worldwide biodiversity collections network, industry, and groups like the iDigBio GWG are encouraged to collaborate to support:

• increased outreach and training of necessary skills, standards, and literacy in the biodiversity data community,
• georeferencing of specimens at the time of collection (including uncertainty and source of the coordinates),
• further development and integration of tools such as GEOLocate into collection management systems,
• streamlining batch processing (Guo et al. 2008),
• the development of shared locality services and more gazetteer resources to reduce repeated georeferencing efforts and improve usefulness, and
• the development of techniques to publish and link georeferenced research data sets to the original occurrence records and physical specimens.