Freshwater Data Publishing Guide

There are several spatial scales at which habitat information should be reported with observation data to improve dataset utility and usability. Here, we define these based on definitions in the IUCN Global Ecosystem Typology, which is a hierarchical classification system that groups the world’s ecosystems by realm, biome, and ecosystem functional groups. At the largest scale, realms differentiate between terrestrial, freshwater, marine, subterranean, and atmospheric components of the biosphere, as well as transitional zones between realms. Biomes are ecologically similar subcomponents of realms with broadly similar features. Within biomes, the typology classifies ecosystems by joining those with similar ecological conditions into ecosystem functional groups. Different classification levels in the hierarchy of this typology offer relevant information to better understand observations of freshwater data.

One of the challenges of freshwater data is the prevalence of organisms that make use of multiple realms. For example, a large number of freshwater macroinvertebrates are insects, many of which live in freshwater only during particular life stages (e.g. as immature larvae or nymphs) and live in the terrestrial realm as adults. Some fish species are anadromous or catadromous, which means that part of their life is spent in freshwater and part is spent in marine habitats. Many bird species make use of freshwater for feeding, while breeding and nesting in terrestrial habitats. Some plant species are capable of growing in freshwater systems but also in high-moisture terrestrial habitats. In these and other cases, it is not sufficient to simply know that the species was observed. It is highly relevant to know whether observations of the species were made in and around freshwater or in another realm, as this provides important information about life stage, spatial distribution, and habitat use.

Information about the biome in which an observation was made is also important for understanding the data and for grouping comparable datasets. The IUCN Global Ecosystem Typology separates the freshwater realm into three biomes—rivers and streams; lakes; and artificial wetlands—while groundwater, brackish water, palustrine wetlands, and coastal systems are grouped within transitional realms.

This classification, which describes the type of water body in which the organism was observed, can provide important ecological information to support observations. Within biomes, ecosystem functional groups, which describe ecological conditions (e.g. permanent, seasonal or episodic/ephemeral; freeze-thaw; upland or lowland; large or small), provide further information that is relevant to understanding how comparable ecosystems (and therefore observations) might be. Classification of sampled ecosystems within these groups is necessary to understand the data and facilitate broad-scale assessments of biodiversity.

At a smaller spatial scale, beyond the scope of the IUCN Global Ecosystem Typology, the habitat zones sampled within a water body contribute to biodiversity differences between datasets. Within freshwater bodies, there are natural differences in taxonomic composition among habitat zones that highlight the importance of indicating this information in the supporting metadata. In lakes, these differences are evident among lake zones, which define habitats based on depth and characteristics related to light penetration, oxygen levels, substrates and temperature. For example, the taxonomic composition of macroinvertebrate and algae samples collected from the littoral (shallow shoreline) zone in lakes differs greatly from that of samples collected in the profundal (deep) zone and these samples are generally not comparable. Similarly, natural differences in taxonomic composition may be expected in plankton and fish samples collected from littoral and pelagic (open water) zones of lakes. River mesohabitats differentiate between types of flow, with riffles being fast-flowing shallow rocky areas, runs representing deeper fast-flowing areas, and pools indicating areas of slow-flowing or standing water, all of which might be expected to house different taxa and biomass. Furthermore, benthic samples collected along the margins of a larger river may differ naturally in composition from deeper samples collected from the center of the channel due to differences in substrate and flow conditions. Information about the sampled lake zone or river mesohabitat is therefore necessary to assess comparability of datasets. At the finest scale, information on sampled microhabitats within lakes or rivers (e.g. samples collected from a particular substrate type such as sandy or rocky), when available, can provide a further indication of expected biodiversity patterns. It can also be informative to differentiate between samples collected across multiple microhabitats from those that were specific to a particular microhabitat type.

For some freshwater species, the life cycle stage of the observed organism is important for understanding community and population dynamics, as well as providing information about life history timing, or phenology. For example, the life cycle stage of insect species provides an indication of whether the observation was of a freshwater or terrestrial habitat for those species that have freshwater juvenile life stages and terrestrial adult life stages. For insects that have multiple freshwater life stages (e.g. beetles that live in freshwater as larvae, pupae, and adults), this information contributes to a greater understanding of population dynamics by indicating the relative proportion of adults and juveniles at the time of sampling. The relative proportion of larvae and pupae for insects with complete metamorphosis, along with the timing of sampling, also provides information about the timing of adult emergence, which is important to track in relation to changes in water temperature.

Life stage details provide similarly valuable information about population dynamics for other organism groups, such as zooplankton and fish. Within zooplankton assemblages, it may be useful to know how many juvenile copepodites and adult copepods are present, and labeling of individuals as nauplii (the first larval life stage for copepods) might be necessary if individuals are too young for species-level identification. For fish, tagging of individuals as young-of-the-year (fish age 0, born within the last year) is important for tracking population dynamics, particularly of threatened or at-risk species. Life stage information is also helpful to understand the timing of important life history events, such as fish migration, spawning, and hatching.

The timing of sampling is highly relevant to understanding life stage information for observed taxa. Full details about the date of sampling (including day, month, and year) are critical for tracking changes in the timing of life history events. Furthermore, information about the season of sampling (recognizing differences between the northern and southern hemisphere) provides context for sampling and allows datasets to be grouped by similar seasonal conditions.

Sampling methods play a major role in determining how comparable different freshwater datasets are and whether freshwater data from different sources can be combined in a meaningful meta-analysis of biodiversity measures and community composition (Lento et al. 2019; Jarvis et al. 2023). For example, different freshwater fish sampling gear types are only effective on a portion of the fish assemblage, and some net mesh sizes fail to capture small-bodied fish species. Combining fish datasets with diverse sampling methods may thus introduce methodological bias into the meta-analysis.

Similarly, the type of sampling equipment used to collect benthic macroinvertebrates has an impact on the collected data. For example, grab samplers (e.g. Ekman and Ponar grabs) and dredges would not be expected to collect comparable samples to fixed area sampling equipment such as Surber and Hess samplers or deployed equipment such as Hester-Dendy samplers. Furthermore, fixed-area samplers may not collect as much diversity as multi-habitat samplers such as kick nets. Mesh size of samplers also plays a role in determining comparability of benthic macroinvertebrate samples, as smaller net mesh sizes capture smaller animals which might increase the number of species at a given site.

For small-bodied planktonic organisms, the net mesh size or the filter pore size is critical for understanding the degree of comparability of different samples. For example, zooplankton nets vary in mesh size, with the larger sizes excluding rotifers and other small-bodied zooplankton and thus underestimating diversity and potentially excluding an entire phylum (Mack et al. 2012; Pansera et al. 2014). Phytoplankton samples are often taken by collecting and filtering a water sample, but the filter pore size will impact whether picoplankton and nanoplankton (among the smallest size classes of phytoplankton) are retained.

Differences in the amount of effort spent sampling, as measured by time, area, or number replicates, ultimately impacts the abundance of collected taxa as well as the probability of collecting more taxa. The greater the effort, the higher the diversity of collected samples (up to a point where additional effort does not increase the number of taxa collected; Gotelli and Colwell 2001). However, it is important to recognize that some compromise is necessary when combining datasets for analysis. While differences in sampling equipment and mesh size can have dramatic effects on the comparability of different datasets, differences in effort may be accounted for in analysis and interpretation.

GBIF defines and supports four classes of datasets: resources metadata (metadata-only datasets), checklist datasets, occurrence datasets, and sampling-event datasets (for detailed definitions and metadata requirements, see Dataset classes and How to choose a dataset class on GBIF?). Differences between dataset classes are defined in terms of the amount of information provided by the data holder. In brief:

Each dataset class allows for different usage of the data. The simpler classes allow for more basic descriptions of the geographic range of available records, observed geographic ranges of taxa, or summaries of expected taxa within a region. In contrast, the most detailed classes (e.g. the sampling-event dataset) allow for the assessment of community composition and biodiversity measures.

To support the effective use of GBIF datasets, whether in simple summaries or more in-depth assessments, there are additional ways to categorize freshwater datasets beyond the four defined GBIF classes. While the GBIF classes largely reflect the amount of available data or metadata, it is important to categorize occurrence and sampling-event datasets based on the type of observation that was made. Based on the type of observation, freshwater datasets can be:

The importance of categorizing freshwater datasets based on the type of observation relates to how the data can be used in further analyses. If data represent opportunistic observations, they can only be used to indicate species presence. Opportunistic observations cannot be used to indicate where a species is not found (e.g. to draw conclusions about its conservation status) nor can they describe the abundance of a species, because no systematic effort has been made to detect the species or quantify its abundance. Caution is therefore advised when combining opportunistic observation data with data from targeted or assemblage sampling, as the conclusions that can be drawn from opportunistic observations are more limited than what might be possible with data that resulted from structured sampling efforts.

Caution is also necessary when combining datasets from organized sampling efforts. Targeted sampling data and assemblage sampling data cannot be compared in terms of diversity or community composition because targeted sampling does not represent an attempt to record all observed taxa and thus does not describe the assemblage as a whole. While the absence of a particular taxon from assemblage sampling data suggests that the taxon was not found in a particular location during the sampling event, its absence from targeted sampling data may simply reflect the fact that it was not the species of interest during sampling and was therefore not recorded.

Freshwater datasets should also be categorized based on the type of data contribution, which we define as:

The type of data contribution has implications for the types of quality checks that may be necessary for datasets retrieved from GBIF. For example, citizen science data may require different quality checks than professional data provided by taxonomic experts or observations from lab-processed samples (Jarvis et al. 2023), particularly for taxonomic groups that must be identified with a microscope. The distinction between community-based research data and citizen science data in our definitions is based on the degree to which there has been training and/or collaboration with professionals, increasing the probability of accurate sampling results. Under these definitions, citizen science data are those collected without training or support from professionals, which are therefore most likely to require quality checks before further data use.

Users who search for data on GBIF may be interested in the general biodiversity of all organisms in a region, but many have an interest in the diversity of a particular organism group. Organism groups are collections of biologically and ecologically similar organisms that are generally grouped together and described as an assemblage. For example, phytoplankton is an organism group that refers to microscopic and planktonic (passive floaters/drifters and weak swimmers that are carried by current) autotrophic (self-feeding) organisms, including algae and bacteria. Benthic macroinvertebrates refers to a group of organisms that can be seen with the naked eye (not microscopic), that have no backbone and that live on the bottom of lakes, rivers, and wetlands, including worms, snails, clams, and aquatic life stages of insects. Generally, freshwater organism groups often comprise more than one order/class/phylum (e.g. benthic macroinvertebrates consist of Trichoptera, Plecoptera, Gastropoda, etc.). The groupings offer a way to refer to particular components of freshwater communities generally studied together.

Adding the organism group to which an observation belongs is a way to make data easier to find and select within GBIF. For example, someone who is interested in phytoplankton diversity would find it useful to be able to select data by the organism group name (phytoplankton) rather than having to search separately for the taxonomic classes that are part of this assemblage. Furthermore, someone who is interested in identifying the spatial distribution of benthic macroinvertebrate sampling data globally would have more success in finding data if each of the taxa of interest (reaching from class to orders) were annotated with the organism group name. Table 1 outlines the organism groups that we recommend adding to freshwater records in GBIF.

Many of the details about sampling methods recommended for inclusion in published freshwater datasets vary depending on the organism group, and applying the labels in Table 1 would facilitate the use of conditional or recommended fields during dataset upload. For example, life stage is a relevant field for benthic macroinvertebrate or fish samples, but not for benthic algae samples. Below, we provide information about relevant fields and sampling details for freshwater organism groups.

An important part of publishing datasets on GBIF is ensuring that sufficient metadata are provided to allow future use of the published dataset. Some metadata are registered at the resource (dataset) level (i.e., the dataset description, version, citation, rights, keywords, contacts, taxonomic and geographic scope) while other metadata can be captured in the records themselves in either occurrence or sampling-event tables.

Freshwater datasets published on GBIF should include the GBIF dataset class (listed as type of dataset: resources metadata, checklist, occurrence, or sampling-event) in the metadata. We recommend adding the type of observation (opportunistic observation data, targeted sampling data, or assemblage sampling data (see §1.2.2)) and the type of data contribution (professional data, community-based monitoring data, or citizen science data) to the occurrence dataset (see <<§2.2>> and <<§3.1.1>>). These categories reflect the opportunities and limitations of each dataset for large-scale data compilation and biodiversity assessment more accurately than the GBIF dataset classes. Table 2 indicates which of these categories can be applied to occurrence or sampling-event datasets. Note that the freshwater data categories may apply to different GBIF dataset classes depending on the amount of information available in the dataset, as indicated below.

Opportunistic observation data are not collected as part of a planned sampling event, e.g. they are not collected through a structured effort to describe the assemblage composition or estimate the geographic distribution or population size of a particular species. Instead, these data may represent secondary observations of non-target species or casual observations of species. Opportunistic observations are grouped as occurrence datasets under GBIF’s dataset classification system because there are no specific sampling methods to report (Table 2). Opportunistic observation data include presence-only records or counts, but the latter is not particularly meaningful without information about the planned effort that can quantify abundance.

Targeted species sampling occurs as part of a planned sampling event but is focused on the collection of a particular species or a subset of species. Assemblage sampling is similarly part of a planned sampling event, but effort is made to record all species observed during the event. Both targeted sampling data and assemblage sampling data are likely to be grouped as sampling-event datasets in GBIF (Table 2), as the sampling effort is documented following a protocol. However, whether these data are grouped as occurrence datasets or sampling-event datasets depends on whether the details and methods of sampling are available.

Under the definition provided in §1.1, most citizen science data are categorized as opportunistic observations. These observations are generally not made as part of an organized sampling effort following specific protocols (such an organized effort would generally constitute community-based monitoring), and there are no sampling methods to report. In contrast, professional data and community-based research data are generally collected as part of an organized sampling effort with a sampling protocol and can be grouped as either occurrence datasets or sampling-event datasets, depending on whether or not event data are published (Table 2).

GBIF requires metadata in XML format corresponding to the GBIF Metadata Profile, which is based on the Ecological Metadata Language (EML). All GBIF dataset classes require the same set of metadata for each dataset (Table 3).

It is useful to know that when datasets are downloaded individually from GBIF, the XML metadata file is included and metadata fields from this table are automatically added to the occurrence file. When data are selected for download from within a polygon (thereby choosing datasets from multiple studies over a given geographic area), less of the metadata is provided in the occurrence table, but the permanent link to the data selection (provided by GBIF with the data download) allows the user to explore metadata for each individual project.

As outlined in §1.1, there are additional metadata fields that are necessary to describe details about the dataset, including where, when and how the data were collected. Some of this information can be reported within the resource metadata, while other fields may be better associated with the occurrence or sampling-event datasets.

Habitat descriptions should at minimum include the realm and biome to indicate whether observations were made in freshwater and in what water body type. For example, these fields may indicate that a semi-aquatic plant was found adjacent to a pond rather than in the pond. The habitat zone is also required to indicate comparability of data, as for organism groups such as benthic macroinvertebrates and zooplankton, assemblage composition will differ naturally in different lake zones and river mesohabitats.

The amount of sampling method information that is required to make informed decisions about data comparability and data selection also differs among organism groups. In some cases, minimal sampling method information is required for datasets to retain usability and broad compatibility. Additional information is particularly needed for organism groups in which methods or equipment may selectively sample only a subset of size classes or taxa. For example, mesh size of sampling nets is important for zooplankton, benthic macroinvertebrates, and fish, as taxa and age classes may be excluded from larger mesh sizes. For phytoplankton, filter pore size is similarly important to ensure different sets of data are focused on a similar portion of the phytoplankton assemblage. Sampling equipment type is highly relevant for benthic macroinvertebrates and fish and can have an impact on the degree of comparability among samples. For microscopic organism groups, it might also be necessary to report the microscope magnification used when processing samples. For some other organisms groups such as macrophytes, amphibians, reptiles, birds, and mammals, the method itself may provide the most relevant information about sample comparability. Across all organism groups, sampling effort, measured as sampled area, time, catch per unit effort, or other similar measures, can be used to standardize estimates of abundance of taxa, even if sampling methods differ. All of these details improve the utility of dataset published on GBIF and can facilitate large-scale analyses of data from different sources.

Checklist datasets are not necessarily specific to a location and do not always represent individual observations. However, they can be location specific (like the country-specific Global Register of Introduced and Invasive Species (GRIIS) checklists) and may contain occurrences (example). Checklist datasets may provide detailed information about species geographic distributions or species profiles that describe species characteristics.

Checklist datasets can be used to make it easier to select all freshwater assemblage data from a particular region, if a checklist of the freshwater species in the region is available. For more information on the possible types of checklists and application of checklists, see GBIF Checklist Datasets and Data Gaps.

There are fewer required fields for checklist datasets compared to other dataset classes (Table 4). As noted, higher classifications (e.g. kingdom) are helpful to include in the data to ensure taxa are not misassigned to the wrong name in the taxonomic backbone.

Occurrence datasets provide information about observations of taxa and the locations where they were found (Table 5). Although only coarse location information is required, the recommended best practice is to always provide coordinates (decimalLatitude and decimalLongitude in decimal degrees), a geodeticDatum which will automatically be interpreted to WGS84 when data are published to GBIF, and a measure of the uncertainty around the coordinates (coordinateUncertaintyInMeters).

Occurrence datasets can be provided as presence data (e.g. a “1” for a site where the taxon was observed) or as counts in the field individualCount (Table 5). Counts in this case refer to situations where there is not an effort to estimate the total abundance of the taxon (e.g. by collecting a sample), but instead, numbers of individuals are recorded (tallied) as individuals are encountered. This could include point counts (e.g. in bird surveys, when an observer counts the number of individuals of each species that is viewed or heard) or opportunistic observations. When an effort is made to estimate, for example, abundance, density or biomass as part of targeted or assemblage sampling, these measures should be recorded in the field organismQuantity with units recorded in organismQuantityType (Table 5). Ideally, such occurrence datasets should also be accompanied by sampling-event datasets to provide details on sampling methods. Finally, if effort has been put into recording true absences (e.g. through systematic and/or extensive sampling procedures), then presence or absence can be recorded in the field occurrenceStatus (Table 5). These distinctions will facilitate meta-analysis of data collected in a similar manner or will allow for data to be adjusted as needed for analysis (e.g. all data converted to presence data).

Sampling-event data are structured and systematic surveys (i.e. periodical or singular surveys, routine or one-time environmental monitoring) that must include metadata describing sampling methods (Table 6). Please note that each event dataset consists of two files: the sampling-event dataset and the associated occurrence dataset. The associated occurrence dataset looks like the one in §2.1.2. but needs to be amended with the eventID (mandatory; identifying the event and linking the two datasets) and the occurrenceStatus (recommended to indicate whether a taxon was present or not detected at a site).

Sampling methods are described in the sampling-event dataset with the field samplingProtocol, which provides a name/link to a specific protocol and/or description of the protocol (Table 6). The recommended best practice is to have a separate event for each sampling method used. In addition to describing the protocol, the field sampleSizeValue and sampleSizeUnit can be used to indicate the spatial or temporal extent of sampling for the described sampling event, as a measure of sampling effort for each event. In addition, the field samplingEffort can be used to record the total effort spent on the event, for example, when there were multiple nets, multiple microhabitats sampled, or multiple periods of time over which sampling occurred. Additional details about sampling methods are recommended to be included in the freshwater DwC extensions described in §3.1.

Most terms that we suggest are urgently needed for other realms as well, which is why most terms do not have a "freshwater" precursor. Those terms that are specific to freshwater or specifically needed to support assessment of freshwater data include "freshwater" as a precursor. For all recommended terms, we have provided freshwater examples.