Guidelines to give the agent¶

Prompting with the documentation standards we wanted to follow, rather than describing the desired output from scratch, was the single biggest time saver. These are the documents we kept in the agent's context (or packaged as a reusable skill) for every documentation task.

Resource	Use
FAIR² Drones Data Standard (repository): `TEMPLATE.md`, `QUICKSTART_GUIDE.md`	Dataset card structure and required metadata for drone and multi-sensor wildlife datasets: Darwin Core Event and Occurrence fields, platform and sensor metadata, annotation and quality fields
Imageomics dataset card template, part of the Imageomics Guide (Campolongo et al., 2026)	Base card template for biodiversity and computer-vision datasets, which the FAIR² template extends
`guidelines/ACKNOWLEDGEMENTS.md`	The funder acknowledgements and disclaimer, to be copied verbatim, never paraphrased
Hugging Face dataset cards	YAML front matter and card conventions
Darwin Core terms	Term definitions for Events and Occurrences, so column names, event types, and the verification-status vocabulary follow the standard
KABR behavior telemetry dataset	A completed FAIR² Drones card and processing scripts to imitate
The project's own field notes, sensor configuration tables, and processing logs	The facts

Local copies¶

The FAIR² Drones files are in the repository under guidelines/, ready to hand to an agent:

FAIR² Drones quick-start checklist

FAIR² Drone Dataset Card: Quick-Start Guide¶

For researchers creating or documenting drone-based wildlife datasets

Below is a streamlined checklist for creating a WildFAIRx compliant drone dataset card. Review the full dataset card template for detailed guidance.

Before You Start¶

[ ] Gather image/video files and annotation files (if applicable)
[ ] Collect drone/sensor specifications (model numbers)
[ ] Locate flight logs or mission notes
[ ] Find research permits and approval numbers
[ ] Create species list with scientific names (use GBIF)
[ ] Set aside 2-3 hours for completion

Choose Your Template¶

Select the appropriate template based on your primary use case:

Object Detection: Detection Template (~2 hours) - Core + Darwin Core + COCO annotations
Multi-Object Tracking: Tracking Template (~2.5 hours) - Detection + MOT format + ID protocols
Behavior Recognition: Behavior Template (~3 hours) - Detection + Ethogram + temporal labels
Robotics Benchmarking: Platform Template (~2.5 hours) - Core + Full telemetry + Minimal annotations
Multiple Tasks: Comprehensive Template (~3-4 hours) - All modules

YAML Front Matter¶

[ ] License: Specify license type (e.g., cc-by-4.0)
[ ] Pretty Name: Provide descriptive dataset name
[ ] Task Categories: List relevant tasks (e.g., object-detection, image-classification)
[ ] Tags: Include relevant tags (e.g., wildlife-monitoring, species names, locations)
[ ] Size Categories: Specify dataset size (e.g., n<1k, 1k<n<10k)

Dataset Overview¶

[ ] Title: Clear, descriptive title
[ ] Description: 2-3 paragraph summary of dataset purpose and content
[ ] Authors: List curators/authors with affiliations
[ ] Contact Information: Provide contact details or point to discussion forum
[ ] Repository: Link to GitHub repository
[ ] Homepage: Provide link if available
[ ] DOI: Assign or note as pending

Data Structure¶

[ ] Directory Structure: Document file organization with tree diagram
[ ] File Formats: List formats for images (e.g., JPG, PNG) and annotations (e.g., COCO JSON, YOLO)
[ ] Naming Convention: Explain file naming pattern with examples
[ ] Data Splits: Describe train/val/test splits with counts and creation method
[ ] Example Files: Link to at least one representative image

Consistent Naming Conventions

Ensure file naming is consistent across videos, telemetry files, and occurrence data: - Use the same date format throughout (e.g., YYYY_MM_DD or DD_MM_YY) - Match session/flight identifiers exactly between CSV files and data folders - Document any naming changes (e.g., session_1 → flight_1) to maintain data linkages - Test that scripts can locate files using your naming pattern

Darwin Core Metadata¶

Event Records¶

For each survey location/date, document:

[ ] eventID: Unique identifier per mission
[ ] eventDate: Date in ISO format (YYYY-MM-DD)
[ ] eventTime: Time with timezone
[ ] decimalLatitude/Longitude: Coordinates in WGS84
[ ] coordinateUncertaintyInMeters: GPS accuracy (typically 5-10m)
[ ] locality: Study site description
[ ] habitat: Habitat type
[ ] samplingProtocol: Survey method (e.g., "UAV transect at 60m AGL")
[ ] sampleSizeValue/Unit: Coverage area and units
[ ] samplingEffort: Duration or effort metric

Extracting GPS Data from Telemetry

If your videos have embedded telemetry (SRT files, EXIF data, or flight logs), you can extract GPS coordinates programmatically: - Use exiftool for EXIF GPS data from images - Parse DJI .SRT files for frame-level GPS coordinates - Extract launch points, min/max bounds, and altitude ranges - Aggregate video-level data to session/mission-level events - See KABR scripts for Python examples

Occurrence Records¶

[ ] Scientific Names: Verify all species names with authority via GBIF
[ ] Taxonomic Hierarchy: Complete Kingdom, Phylum, Class, Order, Family, Genus, Species for each taxon
[ ] Occurrence Table: Create table linking events to species observations

Platform Specifications¶

[ ] Platform Type: Drone type (multirotor, fixed-wing, hybrid)
[ ] Manufacturer and Model: Full platform identification
[ ] Physical Specs: Weight, dimensions, flight time, max speed, wind resistance
[ ] Sensor Details: Camera manufacturer, model, resolution, sensor size, focal length, field of view
[ ] Additional Sensors: List any thermal, LiDAR, multispectral sensors
[ ] Gimbal: Type and axes of stabilization
[ ] Autonomy Mode: Manual, waypoint, or fully autonomous
[ ] Flight Features Used: Grid, orbit, follow, terrain-following, etc.

Mission Parameters¶

[ ] Flight Altitude: Range in meters AGL
[ ] Flight Speed: Speed in m/s
[ ] Flight Pattern: Description (grid, transect, adaptive)
[ ] Coverage: Area covered per mission
[ ] Image Overlap: Forward/side overlap percentages
[ ] Environmental Conditions: Weather, temperature, wind, visibility
[ ] Telemetry Data: Available flight logs and formats (GPS, IMU, battery, etc.)
[ ] Permits: Research permits, IRB/IACUC approvals, aviation regulations followed
[ ] Animal Welfare: Minimum altitudes, disturbance protocols

Annotations¶

Format and Coverage¶

[ ] Supported Tasks: Detection, tracking, segmentation, behavior, re-ID, keypoints
[ ] Annotation Format: Specify format for each task (COCO, MOT, etc.)
[ ] Label Set: List all classes/species, behaviors, attributes
[ ] Total Counts: Images, annotations, annotations per image (min/max/avg)
[ ] Per-Class Distribution: Count per class/species

Quality Metrics¶

[ ] Creation Method: Manual, semi-automatic, or automatic
[ ] Annotation Tool: Software used (CVAT, Label Studio, etc.)
[ ] Annotators: Who created annotations (experts, students, crowd workers)
[ ] Quality Assurance: Number of annotators, inter-annotator agreement, review process
[ ] Confidence Scores: Whether included in annotation files
[ ] Known Issues: Annotation gaps, difficult cases, systematic biases

Annotation Difficulty¶

[ ] Occlusion: Percentage of instances with none/partial/heavy occlusion
[ ] Crowd Density: Distribution across sparse/moderate/dense scenarios
[ ] Scale Variation: Range of object sizes
[ ] Environmental Challenges: Glare, shadows, motion blur, etc.

Dataset Statistics¶

[ ] Temporal Coverage: Date range, seasons represented
[ ] Spatial Coverage: Number of locations, total area surveyed
[ ] Class Balance: Distribution across classes/species
[ ] Baseline Results: Performance metrics if available

Limitations and Considerations¶

[ ] Known Biases: Geographic, temporal, species, environmental
[ ] Limitations: Technical constraints, coverage gaps, quality issues
[ ] Recommendations: Guidance for appropriate dataset use
[ ] Ethical Considerations: Privacy, animal welfare, cultural sensitivities
[ ] Reporting Issues: Link to issue tracker or community forum

Licensing¶

[ ] License Details: Confirm license choice and any special conditions
[ ] Component Licenses: Note if images, annotations, or code have different licenses
[ ] Attribution Requirements: Specify how to cite the dataset

Citation¶

[ ] Dataset Citation: Provide BibTeX entry for the dataset
[ ] Associated Paper: Include citation for related publications

Acknowledgements¶

[ ] Funding Sources: List grants and funding agencies
[ ] Contributors: Acknowledge field teams, annotators, collaborators
[ ] Institutional Support: Note supporting organizations

Optional Sections¶

[ ] Glossary: Define technical terms or specialized calculations
[ ] Additional Information: Any other relevant context
[ ] Related Datasets: Links to complementary datasets
[ ] Multimodal Linkages: Connections to other sensor data

Validation Checklist¶

WildFAIRx Compliance¶

[ ] DOI assigned or pending
[ ] License clearly stated
[ ] All required (*) fields completed in template
[ ] Machine-readable YAML front matter filled
[ ] Contact information provided

Darwin Core Compliance¶

[ ] Event records complete (dates, locations, protocol)
[ ] Occurrence records have scientific names with authorities
[ ] Taxonomic hierarchy filled (minimum to family level)
[ ] Sampling effort quantified
[ ] Coordinates in WGS84 with uncertainty

Practical Completeness¶

[ ] Directory structure documented
[ ] File naming convention explained
[ ] Data splits clearly defined
[ ] Annotation format specified
[ ] At least one example image linked
[ ] Known limitations acknowledged

Common Mistakes to Avoid¶

Vague Descriptions ❌ "We used a drone to collect wildlife images" ✓ "DJI Matrice 300 RTK with Zenmuse H20T camera flew grid patterns at 60m AGL"

Missing Geographic Precision ❌ "Collected in Tanzania" ✓ "Serengeti National Park, Mara Region (-2.3456, 34.8123 ±5m)"

Unclear Sampling Effort ❌ "Multiple flights" ✓ "45 missions totaling 30 flight hours, covering 2,500 hectares"

Incomplete Species Names ❌ "elephants, zebras, giraffes" ✓ "Loxodonta africana, Equus quagga, Giraffa camelopardalis"

Undocumented Splits ❌ "Split into train/val/test" ✓ "Stratified by location and season: missions 1-300 (train), 301-350 (val), 351-400 (test)"

Hidden Biases ❌ "Representative wildlife dataset" ✓ "Dry season only; large-bodied species overrepresented; morning flights bias against nocturnal species"

Time-Saving Tips¶

Before Starting: - Gather all information before opening the template - Copy from existing paper methods sections - Look up drone/camera specs online from manufacturer sites

While Completing: - Start with easy sections (Overview, Structure) to build momentum - Mark sections to revisit with TODO notes - Use EXIF data from images for missing camera/GPS information - Estimate reasonably if exact values unavailable (note as approximations)

Automation Opportunities: - Write scripts to extract GPS/altitude from telemetry files (SRT, EXIF, flight logs) - Generate occurrence records from detection annotations - Calculate statistics (image counts, annotation distributions) programmatically - Aggregate frame-level or video-level data to session-level events - Validate Darwin Core format compliance automatically

For Missing Information: - Document what's unavailable rather than leaving blank - Contact original research team if retrofitting dataset - Use "not available" or "not recorded" explicitly

Resources¶

Validation Tools¶

Python scripts for template generation and validation (coming soon)
Darwin Core export tools for GBIF submission
HuggingFace conversion utilities

External Resources¶

Getting Help¶

Questions, Comments, or Concerns?

For assistance: - Report issues or unclear sections via GitHub Issues - Contribute example cards or improvements - Share feedback on the template

Ready to start?

Download the appropriate template for your task
Gather your information using the checklist above
Set aside 2-3 hours to complete the card
Follow the template section by section
Validate your completed card
Publish your WildFAIRx compliant dataset!

Packaging as a skill¶

The guidelines are small enough to include verbatim. A minimal skill directory for a coding agent:

skills/dataset-documentation/
├── SKILL.md            # when to use, the steps, the review checklist
├── TEMPLATE.md         # FAIR² Drones dataset card template
├── QUICKSTART_GUIDE.md # FAIR² Drones checklist
├── ACKNOWLEDGEMENTS.md # funder language, copied exactly
└── dwc_terms.md        # the Darwin Core terms the project uses, with definitions

SKILL.md should state the order of operations (inventory, summarize, cross-check, draft, review), the rule that numbers are computed rather than copied, and the rule that missing information is recorded as missing.

References¶

Wilkinson, M. D. et al. (2016). The FAIR Guiding Principles for scientific data management and stewardship. Scientific Data 3, 160018. https://doi.org/10.1038/sdata.2016.18
Wieczorek, J. et al. (2012). Darwin Core: an evolving community-developed biodiversity data standard. PLoS ONE 7(1), e29715. https://doi.org/10.1371/journal.pone.0029715
Kline, J. and Campolongo, E. G. (2026). FAIR² Drones Data Standard for Wildlife Datasets. https://imageomics.github.io/fair_drones/
Kline, J. M., Potlapally, A., and Campolongo, E. G. (2026). SmartWilds: Integrating, Aligning, and Documenting Multi-Modal Field Sensing Data. Under review.
Campolongo, E. G., Thompson, M. J., Zhang, N., Duan, Y., Bradley, J., & Lapp, H. (2026). Imageomics Guide (Version 2.3.1) [Computer software]. https://doi.org/10.5281/zenodo.17817610
Barnas, A. F. et al. (2020). A standardized protocol for reporting methods when using drones for wildlife research. Journal of Unmanned Vehicle Systems 8(2), 89–98. https://doi.org/10.1139/juvs-2019-0011

To cite the standard and the paper:

@misc{fair_drone_standard,
  title        = {FAIR² Drones Data Standard for Wildlife Datasets},
  author       = {Jenna Kline, Elizabeth Campolongo},
  year         = {2026},
  publisher    = {GitHub},
  howpublished = {\url{https://github.com/Imageomics/fair_drones}}
}

@article{kline2026smartwilds,
  author  = {Kline, Jenna M. and Potlapally, Anirudh and Campolongo, Elizabeth G.},
  title   = {SmartWilds: Integrating, Aligning, and Documenting Multi-Modal Field Sensing Data},
  year    = {2026},
  note    = {Under review}
}