Wildlife Data Sources and Biodiversity Research Methods

Wildlife research data are not interchangeable. A museum specimen can document historical occurrence, a community observation can document a photographed or recorded encounter, a camera trap can document a detection at a monitored station, repeated surveys can estimate site occupancy, capture–recapture can estimate population parameters, telemetry can describe the movements of tagged individuals, and environmental DNA can reveal genetic traces left in water, soil, sediment or air. The correct source depends on the biological question. Presence, occupancy, abundance, density, population trend and geographic range describe different properties and normally require different evidence.

Match the Biological Question to the Evidence

A large biodiversity database is useful only when its records can support the inference being made. Thousands of occurrence records may document where a species has been reported, yet still provide no defensible estimate of how many animals occur in those places. A smaller survey designed around repeated visits, known effort and detection probability may support a stronger ecological estimate.

This distinction separates observations from ecological estimates. GBIF describes occurrence datasets as evidence that a taxon was recorded at a particular place and time, while sampling-event datasets can retain methods, sample size, effort and linked detections. Those additional fields make comparisons among surveys more defensible and can support some forms of non-detection inference. [a]

Wildlife Evidence Comes From Different Observation Processes

The evidence type matters because every observation process filters wildlife differently. A visual survey favors animals that can be seen during the survey period. A microphone favors organisms producing detectable sound. A ground-level camera favors animals that pass within its detection zone. DNA sampling can detect organisms that were never seen, but the DNA may have moved through water, soil, air or another carrier before collection.

Darwin Core formalizes many of these distinctions. Its basisOfRecord term can identify records such as preserved specimens, human observations and machine observations, while Event terms can preserve sampling protocol, sample size and sampling effort. [b]

Occurrence Records Establish Recorded Presence, Not Population Size

GBIF aggregates biodiversity information supplied by institutions and data publishers around the world. Its records can originate from natural-history specimens, field observations, citizen-science observations, machine observations and other sources. This makes occurrence infrastructure useful for locating documented evidence across large spatial and temporal scales. [c]

An occurrence record usually answers a narrow question: a taxon was reported from this place at this time under the information supplied with the record. The record may contain an individual count, but that does not automatically turn it into an estimate of local abundance. A field tally of five animals, for example, describes the animals counted during that event; estimating the number of animals that were present but not detected requires a sampling and analytical design capable of handling that problem.

Sampling Events Preserve the Search Effort

A sampling event can retain the protocol used, the amount of effort, where the event occurred, when it occurred and which taxa were recorded. GBIF currently requires eventID, eventDate, samplingProtocol, sampleSizeValue and sampleSizeUnit for its sampling-event dataset structure and strongly recommends fields including samplingEffort and occurrenceStatus. [d]

Taxonomy Has to Be Fixed Before Records Are Combined

Two datasets can describe the same biological taxon under different names. A species can be transferred to another genus, an older name can become a synonym, one taxon can be split into several species, or several names can later be treated as referring to one species. Merging records by text strings alone can therefore omit valid evidence or combine records that no longer represent the same taxonomic concept.

Catalogue of Life now provides two distinct releases. The Base Release prioritizes expert-curated global sources, while the Extended Release adds programmatically integrated information from additional taxonomic, nomenclatural, regional and other sources to increase coverage. The August 2026 versions preserve this distinction, and Catalogue of Life recommends choosing a release according to the balance between expert-vetted coverage and broader completeness needed for the work. [e]

A reproducible species list should preserve the taxonomic source and version used to resolve accepted names. Otherwise, the same analysis rerun after later splits, synonym changes or source replacements can produce a different species total even when the underlying occurrence records have not changed.

Major Biodiversity Platforms Serve Different Research Tasks

Biodiversity platforms should be selected by the type of evidence they curate rather than by database size alone. An occurrence aggregator, a taxonomic catalogue, a conservation assessment system, a movement-data repository and a barcode library solve different problems.

Marine Records Require Marine Data Context

OBIS harvests marine occurrence records from thousands of datasets and makes them accessible through a single integrated dataset. Its access guidance distinguishes interactive exploration from programmatic analysis of large subsets, for which it recommends the GeoParquet version of the integrated data. [f] Marine records may also carry depth, sampling-event and environmental context that cannot be reduced to a two-dimensional point on a map.

Community Identification Is a Quality Filter, Not a Population Estimate

On iNaturalist, verifiable observations require an accurate date and location, media evidence, and an organism that is not treated as captive, cultivated or human. Research Grade then depends on community identification and the platform’s Data Quality Assessment rules. [g] That can make the record useful evidence of an observation, but it does not establish local density, long-term occupancy, breeding or a complete regional species inventory.

Field Survey Design Determines What Can Be Estimated

Field methods should be selected after defining the biological variable of interest. A survey designed to produce a species list may not contain the repeated observations needed for occupancy modelling. A point count may produce useful standardized counts without supporting absolute density. A telemetry study can produce millions of measurements while sampling only a small set of animals.

Direct Observation and Standardized Counts

Visual encounter surveys, point counts, transects, timed searches and area searches remain useful when the organisms can be detected directly and survey conditions can be standardized. Their value depends on what was recorded about the event: observer number, duration, distance covered, searched area, weather, season, time of day and target taxa can all change the probability of encountering wildlife.

A raw count is the number detected under the survey conditions. It becomes an estimate of population size or density only when the study design and model account for the probability that animals entered the sample. USGS describes quadrat, line-intercept, distance, repeated-count and capture–recapture approaches among the methods used for population size or density estimation and emphasizes that population estimates must account for sampling probability. [h]

Repeated Surveys Separate Occupancy From Detectability

A species can occupy a site and still remain undetected during one visit. Occupancy methods address this by using repeated observations to estimate both the probability that a site is occupied and the probability that the species is detected when present. The distinction matters for cryptic, nocturnal, seasonal, rare or weather-sensitive animals.

The statistical basis is simple even when the model is not: non-detection is not equivalent to absence when detection probability is below one. Research developed by MacKenzie and colleagues established repeated-survey occupancy methods specifically around this imperfect-detection problem. [i]

Capture–Recapture Follows Individuals Through Repeated Sampling

Capture–recapture links repeated detections to identifiable individuals. Identification can come from physical marks or tags, but modern studies may also use photographs of natural markings or DNA from non-invasive samples. Depending on the model and study design, repeated encounter histories can inform abundance, survival, movement and transitions among states or locations. [j]

Camera Traps Record Detection Events, Not Automatic Animal Counts

Camera traps can operate for long periods with little field disturbance and are especially useful for terrestrial animals that pass through the camera’s detection zone. The biological meaning of the resulting records depends on deployment design. Cameras placed on trails, at water sources, on random grid points or at bait stations sample animal movement in different ways.

One animal can trigger a camera repeatedly. A burst can contain several photographs of the same passage, and the same individual can return many times. Image totals therefore cannot normally be read as animal totals. Studies may group photographs into detection events, identify individuals when markings permit, build detection histories for occupancy analysis, or use designs developed for density estimation.

Camtrap DP addresses the data-management side of this problem by separating deployments, media and observations. Deployment records retain where and when cameras operated and can include camera setup information; media records retain capture time and file information; observation records describe detections derived from the media. [k]

AI Classification Does Not Remove Verification Needs

Automated classification can remove much of the manual work involved in large camera datasets, particularly blank-image filtering and initial taxonomic predictions. Wildlife Insights currently reports an AI training set of more than 35 million labelled camera-trap images covering 1,295 species and 237 higher taxonomic classes. Its system can return higher-level identifications or an unknown result when species-level confidence is inadequate. [l]

Model confidence should remain separate from record verification. Performance differs among taxa and image conditions, and a classifier trained heavily on one species or camera setting may behave differently when animals, backgrounds, lighting or geographic settings change.

Passive Acoustic Monitoring Samples Sound-Producing Wildlife

Autonomous recorders can collect repeated acoustic samples through day and night without an observer remaining at the site. This is useful for animals whose presence can be recognized from sound, including many birds, bats, frogs, insects and marine mammals. Scheduled recordings can also preserve an explicit temporal sampling design.

The observation unit is an acoustic detection, not necessarily an individual animal. One bird can call many times, several animals can call simultaneously, and an animal that remains silent may be present but acoustically unavailable. Wind, rain, vegetation, microphone sensitivity, background noise and distance affect what reaches the recorder.

Automated sound classifiers add another detection layer. False positives can be produced by similar vocalizations or noise, while false negatives can occur when calls are faint, overlap or fall outside the model’s training coverage. A 2026 assessment of landscape-scale biodiversity monitoring methods treated structured and unstructured human surveys, camera traps, eDNA, drones, passive acoustic recorders and satellite remote sensing as distinct data sources because each measures different components of biodiversity and carries different operational constraints. [m]

Environmental DNA Detects Genetic Traces Rather Than Visible Animals

Environmental DNA, or eDNA, refers to DNA shed by organisms into their surroundings. Material can originate from skin cells, mucus, feces, gametes, hair, tissue and other biological material. Sampling can target water, soil, sediment, air or another environmental medium. USGS uses eDNA for biodiversity studies and for detecting threatened, invasive or otherwise difficult-to-detect organisms. [n]

Targeted Assays and Metabarcoding Answer Different Questions

A targeted assay searches for genetic material associated with a selected taxon. Metabarcoding uses markers and sequencing to detect many taxa from the same environmental sample. The latter is suited to community-level surveys, but taxonomic assignment depends on the sequence data, bioinformatic pipeline and available comparison library.

BOLD provides DNA barcode reference data linked to specimen and taxonomic information. Its 2026 data packages contain tens of millions of specimen and sequence records, while its identification service compares submitted barcode sequences against reference libraries. [o] A sequence that has no suitable reference match cannot be converted reliably into a species name merely because sequencing succeeded.

A DNA Detection Has Its Own Spatial Limits

A positive eDNA result does not necessarily locate a living animal at the exact sampling point. Water can transport DNA downstream; sediment can preserve older material; organisms, predators or human activity can transport biological traces; and contamination can create false detections. Conversely, degradation, limited shedding, poor sampling position or laboratory sensitivity can produce non-detection even where organisms occur.

Marine biodiversity infrastructure is now adapting to these more complex event structures. In July 2026, OBIS announced its first eDNA dataset published through Darwin Core Event Core, linking environmental sampling events and DNA-derived observations rather than flattening them into isolated species points. [p]

Tracking Data Describe the Animals That Were Tagged

Telemetry and biologging answer questions that static occurrence records cannot. GPS, Argos, radio telemetry, acoustic telemetry and geolocation can create individual tracks. Animal-borne sensors can add acceleration, depth, temperature, orientation and other measurements.

Movebank stores movement and other animal-borne sensor data together with information about animals, tags and deployments. Its data model explicitly separates a tag from the animal carrying it and distinguishes deployed records from locations collected before or after attachment. [q]

Record volume is especially misleading in tracking studies. Hundreds of thousands of positions can come from a small number of tagged animals. The correct biological sample size depends on the design and inference, not the number of GPS rows.

Remote Sensing Usually Measures Habitat More Directly Than Wildlife

Satellite imagery, aerial imagery, radar and LiDAR can describe vegetation cover, canopy structure, water extent, snow, fire, surface temperature, land-use change and other environmental properties over large areas. These measurements are often used to characterize habitat or environmental conditions associated with animal observations.

Remote sensing can sometimes detect animals directly, particularly large animals, colonies or conspicuous aggregations under suitable spatial resolution. That use should be separated from the more common situation in which remote sensing measures the environment and wildlife data come from another source.

Combining camera detections with forest structure, or telemetry positions with water, elevation and vegetation layers, can test habitat relationships. The environmental layer does not itself establish that an animal occurred in every pixel assigned suitable habitat.

Sampling Effort Changes the Meaning of Wildlife Counts

Two sites cannot be compared fairly from raw detections when one received much more search effort than the other. Survey effort is therefore part of the biological evidence rather than administrative metadata. Darwin Core explicitly provides samplingEffort for recording the amount of work expended during an event. Examples include trap-nights, observer-hours and distance travelled during a survey. [r]

Detection, Non-Detection and Biological Absence Are Separate States

Detected means the survey produced accepted evidence of the target. Not detected means sampling occurred but the survey did not produce an accepted detection. Unknown means suitable sampling information is unavailable. Only the first two states require an actual sampling process; a blank database cell may simply represent missing evidence.

Bias Begins Before the Dataset Reaches an Analyst

Wildlife datasets inherit the geography and behavior of the people and devices that generated them. Accessible roads, protected areas, cities, research stations and popular birding locations can receive more attention than remote terrain. Charismatic or easily identified animals often accumulate more records than small, cryptic or taxonomically difficult groups.

• Spatial bias: survey effort is concentrated in accessible or preferred locations.
• Temporal bias: sampling is concentrated in certain seasons, times of day or years.
• Taxonomic bias: some groups receive far more observation, collection and identification effort.
• Observer bias: expertise and identification ability differ among observers.
• Sensor bias: camera position, microphone sensitivity, trigger settings and equipment failure affect detections.
• Digitization bias: specimens may exist in collections without being digitized or georeferenced.
• Publication bias: collected records may remain in local databases, project files or restricted repositories.

These biases do not make biodiversity data unusable. They define which comparisons require correction, additional sampling or narrower claims. Opportunistic records can still be excellent occurrence evidence while remaining poorly suited to direct comparisons of abundance among locations.

Cleaning Data Should Preserve Uncertainty Instead of Erasing It

Data cleaning is not simply the removal of unusual points. Records should be checked against the intended biological and analytical use. Duplicate identifiers, impossible coordinates, swapped latitude and longitude, historical locality errors, captive records, taxonomic mismatches and records outside the chosen time period can all affect an analysis.

An outlying locality can be an error, but it can also be evidence of dispersal, introduction, a vagrant animal, an overlooked population or a genuine range extension. When the original record can be inspected, identification history, media, specimen information, coordinate uncertainty, collector notes and dataset provenance provide stronger grounds for a decision than geographic surprise alone.

Historical evidence also needs to remain historical. A specimen collected many decades ago can be valid evidence that a species occurred at that place in the past without establishing current occupancy. Converting all dated records into one timeless presence layer removes biological information.

Standardized Fields Make Independent Datasets Easier to Combine

Biodiversity integration depends on retaining the meaning of fields across institutions and projects. Darwin Core supplies defined terms for occurrences, events, locations, identifications, taxa, organisms and related information. This allows a museum, field survey or sensor project to publish data with recognizable fields rather than requiring every user to reverse-engineer a project-specific spreadsheet.

Standardization does not make two surveys methodologically identical. Two projects can both use Darwin Core while applying different survey protocols, target groups and definitions. Interoperability helps records travel between systems; ecological comparability still depends on the sampling process.

Camera-trap data show why method-specific extensions matter. Camtrap DP retains deployments, media and observations rather than forcing every image into a flat occurrence table. eDNA publishing is moving in a similar direction by retaining environmental sampling events and their derived detections. The shared goal is to keep biological observations attached to the event that produced them.

Conservation Assessments Are Syntheses, Not Raw Survey Records

Conservation assessments can combine distribution, population, habitat, threats and other evidence into an evaluated species account. Their output answers a different question from an occurrence portal. A taxon can have few downloadable observations because of low digitization or limited public reporting without being globally rare, while a widely recorded species can still face population decline.

The IUCN Red List also applies defined spatial standards to mapped assessment data. The current Mapping Standards and Data Quality document is version 1.21, issued in November 2025, and includes standardized treatment of spatial attributes such as presence, origin and seasonality. [s] These attributes matter because a species can be native in one part of a mapped area, introduced in another, resident in one region and seasonal elsewhere.

Combining Methods Works Best When Their Differences Remain Visible

No single wildlife method detects every taxon or measures every biological property. Camera traps can complement acoustic recorders because ground-moving animals and vocal animals enter the observation process differently. eDNA can reveal organisms that visual surveys miss, while direct observation may provide behavior, age class or exact encounter context that a DNA sequence cannot. Telemetry describes movement in detail but only for tagged animals; remote sensing can place those tracks within changing habitat conditions.

Combining datasets therefore requires more than appending rows. Taxonomic names must be reconciled. Spatial precision should be compatible. Historical and recent evidence should remain distinguishable. Survey effort and detection processes should not disappear. Duplicate records passed between aggregators need to be identified where they would affect the analysis.

Data integration is strongest when each observation retains what generated it: taxon, place, time, evidence type, sampling event, protocol, effort, identification history and data provenance.

From Wildlife Observations to Biodiversity Measures

Raw wildlife records can be transformed into broader measures only after decisions about taxonomy, sampling, spatial scale, time and observation error. A photograph can become a verified species occurrence. Repeated detections can contribute to an occupancy estimate. Standardized counts can contribute to abundance models. Repeated community samples can be used to estimate change in community composition.

GEO BON’s Essential Biodiversity Variables organize biodiversity measurements into six classes: genetic composition, species populations, species traits, community composition, ecosystem functioning and ecosystem structure. The system is intended to connect underlying biodiversity observations with standardized measurements that can be compared across space and time. [t]

The transformation can be represented as observation → standardized record → analytical estimate → biodiversity variable → broader indicator. Each step adds interpretation. The final indicator should therefore remain traceable to the observations, methods and assumptions that produced it.

A Defensible Wildlife Dataset Preserves More Than Species Names

A useful dataset allows later users to reconstruct what was observed and under what conditions. The most valuable fields depend on the method, but several information groups recur across wildlife research.

• Taxonomic identity: recorded name, accepted name where resolved, identification history and taxonomic source.
• Evidence type: specimen, human observation, machine observation, genetic detection or another clearly defined record basis.
• Place: coordinates or locality, spatial reference information and uncertainty where available.
• Time: event date or interval rather than an undated assertion of presence.
• Sampling event: the visit, deployment, collection event or sample that produced the observation.
• Protocol: the method used to search for or detect organisms.
• Effort: time, distance, area, volume, trap exposure or another measure appropriate to the method.
• Detection status: whether the taxon was detected, not detected under a defined survey, or simply lacks data.
• Provenance: dataset, institution, project, identifier and original provider.
• Version and license: enough information to reproduce the data selection and determine permitted reuse.

When these properties are retained, wildlife evidence can be reused without pretending that every record measures the same thing. A specimen can remain a specimen, a camera detection can remain tied to its deployment, a GPS location can remain tied to an individual, and a non-detection can remain tied to the survey that produced it. That separation is what allows different evidence streams to be combined without turning them into unsupported claims about distribution, abundance or population change.

Sources and Verification

1. [a] GBIF — Dataset Classes — Used to distinguish occurrence, checklist and sampling-event datasets and the types of inference each data structure can support.
2. [b] TDWG — Darwin Core Quick Reference Guide — Used for definitions of basis of record, Event information, sample size and sampling effort.
3. [c] GBIF — What Is GBIF? — Used for the scope and origin of biodiversity records aggregated through the GBIF network.
4. [d] GBIF — Data Quality Requirements for Sampling-Event Datasets — Used for sampling-event fields, survey effort and presence or non-detection context.
5. [e] Catalogue of Life — Releases — Used for the distinction between the expert-curated Base Release and the broader Extended Release.
6. [f] Ocean Biodiversity Information System — Access OBIS Data — Used for the structure and access methods of the integrated marine biodiversity dataset.
7. [g] iNaturalist — Data Quality Assessment and Research Grade — Used for verifiable-observation and community-identification requirements.
8. [h] U.S. Geological Survey — Abundance: Population Size and Density Estimation — Used for methods that estimate abundance or density while accounting for sampling probability.
9. [i] U.S. Geological Survey — Estimating Site Occupancy Rates When Detection Probabilities Are Less Than One — Used for the distinction between site occupancy and imperfect detection.
10. [j] U.S. Geological Survey — Capture-Recapture Methodology — Used for abundance, survival and movement inference from repeated encounters with identifiable animals.
11. [k] TDWG — Camtrap DP — Used for the separation of camera deployments, media files and biological observations.
12. [l] Wildlife Insights — Artificial Intelligence Models — Used for current camera-trap classification coverage, training-data scope and model-performance limitations.
13. [m] Ashton-Butt et al. — Assessing Potential Data Sources for Landscape-Scale Terrestrial Biodiversity Indicators — Used for current comparison of field surveys, camera traps, eDNA, drones, passive acoustics and satellite remote sensing as monitoring data sources.
14. [n] U.S. Geological Survey — Environmental DNA — Used for eDNA origin and its application to biodiversity, threatened-species and invasive-species detection.
15. [o] BOLD — DNA Barcode Data Packages — Used for the scale and function of current DNA barcode reference data.
16. [p] OBIS — First eDNA Dataset Based on Event Core — Used for the July 2026 implementation of event-based publication for DNA-derived marine observations.
17. [q] Movebank — Manual — Used for animal movement, animal-borne sensor data and deployment-based data management.
18. [r] TDWG — Darwin Core List of Terms — Used for the current definition and examples of sampling effort.
19. [s] IUCN Red List — Mapping Standards and Data Quality — Used for the current version of Red List spatial standards and standardized presence, origin and seasonality attributes.
20. [t] GEO BON — Essential Biodiversity Variables — Used for the six EBV classes and the relationship between biodiversity observations and standardized biodiversity measurements.