EIannino/lion-responses-to-aerial-monitoring-dataset

Curated by: Elena Iannino

Language(s): English (metadata and documentation)

Dataset Overview

This dataset contains multi-modal UAV-based wildlife observations collected in Ol Pejeta Conservancy, a privately managed conservation area in central Kenya. The primary target species is the African lion, observed within open savannah and bushland ecosystems. The dataset includes synchronized RGB and thermal video recordings captured using a UAV platform during controlled experimental approaches. Drone flights were conducted at multiple fixed altitudes to evaluate both animal behavioral responses to UAV disturbance and to enable multi-scale visual wildlife monitoring. The data were collected over a 15-day field campaign (October–December 2024) during daylight hours (08:00–19:00). Observations span multiple lion prides across heterogeneous savannah habitats dominated by acacia and mixed grassland vegetation. This dataset was created to support research in wildlife monitoring, UAV-based ecological sensing, and animal behavior analysis, with a particular focus on understanding how large carnivores respond to aerial robotics in real-world conservation environments.

Supported Tasks and Applications

This dataset supports the following computer vision, ecological, and robotics tasks:

🤖 Computer Vision Tasks

• Object Detection (bounding boxes around lions in RGB and thermal imagery) • Instance Segmentation (potential pixel-level labeling of individuals in future annotations) • Multi-Object Tracking (consistent identity tracking across video frames) • Re-identification (individual lion recognition) • Behavior Recognition (classification of responses to drone presence, social behaviors) • Cross-modal Learning (RGB–thermal fusion for robust detection under variable vegetation and lighting)

🌿 Ecological Applications

• Population abundance estimation (group size estimation from aerial imagery) • Behavioral analysis (quantifying UAV-induced disturbance responses) • Habitat use patterns (space-use inference within savannah and bushland environments) • Species distribution modeling (within protected conservation areas) • Wildlife monitoring and conservation decision support • Human–wildlife interaction studies (response to aerial systems and anthropogenic noise)

🤖 Robotics Applications

• UAV perception system benchmarking in real-world wildlife environments • Aerial robotics safety and disturbance minimization studies • Sensor fusion benchmarking (RGB + thermal + telemetry integration)

Dataset structure

Directory organization

lion-responses-to-aerial-monitorin-dataset/
└── dataset/
  ├── data/
  │   ├── AJ02/  
  │   │   ├── DJI_20241107134254_0001_S.MP4   # 704.6 MB
  │   │   ├── DJI_20241107134254_0001_S.SRT
  │   │   ├── DJI_20241107134254_0001_T.MP4   # 144 MB
  │   │   ├── DJI_20241107134254_0001_T.SRT
  │   │   ├── DJI_20241107134254_0001_V.MP4   # 2.02 MB
  │   │   ├── DJI_20241107134254_0001_V.SRT
  │   ├── AJ03/  … (3 clips × {MP4, SRT})
  │   ├── AJ04/  … (3 clips × {MP4, SRT})
  │   ├── AJ05/  … (3 clips × {MP4, SRT})
  └── metadata/
      ├── darwin_core_occurrences.csv   # Darwin Core — one row per clip  
      └── darwin_core_events.csv        # Darwin Core — one row per mission 

Data instances

Darwin Core Event Tables:

data/darwin_core_occurrences.csv — one row per video clip.

data/darwin_core_events.csv — one row for the full mission.

Raw Videos (data/raw/mission_1//.{MP4,SRT}):

Original 4K footage as recorded on-board, with DJI sidecar files:

MP4 — H.264/H.265-encoded 4K (3840 × 2160) video at ~30 fps SRT — DJI subtitle telemetry file; one entry per frame with GPS, altitude, camera settings, and UTC timestamp (source for the occurrence CSVs) File naming follows the DJI convention: DJI_YYYYMMDDHHMMSS_NNNN_D where NNNN is the clip index on the SD card.

Data Fields

Key field groups:

🌿 Darwin Core Event Fields

data/metadata/darwin_core_occurences.csv:

• eventID
• eventDate
• eventTime
• decimalLatitude, decimalLongitude, coordinateUncertaintyInMeters, geodeticDatum
• locality, habitat
• samplingProtocol, Altitude (mAGL)

data/metadata/darwin_core_events.csv:

• occurrenceID
• eventID
• scientificName, kingdom, phylum, class, order, family, genus, species
• taxonRank
• Count_Individuals, Count_AdultFemales, Count_AdultMales, Count_cubs # counts are for eventID, not for occurrenceID
• Behaviors

Platform and Mission Specifications

🚁 Platform Details

Type

UAV (Unmanned Aerial Vehicle)

Hardware

• Manufacturer: DJI Enterprise • Model: DJI Mavic 3 Enterprise T • Weight: 0.92 kg • Max flight time: 45 min • Max speed: 21 m/s • Wind resistance: 12 m/s

Autonomy

• Mode: Semi-autonomous • Navigation: Manual experimental approaches with hovering • Collision avoidance: Yes (omnidirectional obstacle sensing) • Return-to-home: Yes

Payload

• Max payload: Integrated payload system (non-modular) • Gimbal: Yes, 3-axis stabilized gimbal (tilt, roll, pan)

📷 Sensor Specifications

Primary Sensor: Integrated RGB + Thermal Imaging System

RGB Camera

• Type: RGB • Manufacturer: DJI • Model: Mavic 3T Wide Camera • Resolution: 8000 × 6000 pixels (48 MP) • Sensor size: 1/2-inch CMOS • Focal length: 24 mm equivalent • Field of view: 84° • Frame rate: 4K at 30 fps • Bit depth: 8-bit JPEG/H.264 video

Thermal Camera

• Type: Thermal • Manufacturer: DJI • Model: Mavic 3T Thermal Camera • Resolution: 640 × 512 pixels • Sensor type: Uncooled VOx Microbolometer • Frame rate: 30 Hz • Infrared wavelength: 8–14 μm • Focal length: 40 mm equivalent • Field of view: 61° DFOV

Spectral Bands

Spectral Bands

Band	Wavelength	Purpose
RGB Visible	~400–700 nm	Visual identification and behavioral monitoring; detection and counting of lions under varying vegetation and lighting conditions.
Thermal Infrared	8–14 μm	Visual identification and behavioral monitoring; detection and counting of lions under varying vegetation and lighting conditions.

Calibration

• Calibrated: Factory calibrated • Method: Manufacturer calibration

Synchronization (multi-sensor)

• Method: Integrated hardware synchronization within DJI imaging system

🗺️ Mission Parameters

Flight Specifications

• Altitude: 20–120 m AGL • Speed: 3–10 m/s • Flight pattern: approach followed by stationary hover or manual tracking • Coverage per mission: Opportunistic wildlife encounters; no fixed-area survey • Image overlap: Not applicable • Ground sampling distance: Variable depending on altitude

Environmental Conditions

• Temperature range: 12–28°C • Weather: Typical savannah conditions; likely clear to partly cloudy • Time of day: Morning to evening (08:00–19:00) • Season: Short rainy / transitional season (October–December)

🔍 Sampling Protocol

Survey Design

The study used opportunistic wildlife monitoring combined with controlled experimental drone approaches. Lion groups were located using GPS collars, VHF telemetry, and opportunistic sightings reported by field personnel and safari guides.

Flight Operations

All flights were conducted by licensed drone pilots operating from vehicles with open roofs. Before each mission, operators located lions visually or through telemetry data and launched the drone vertically to a predetermined altitude. Flights maintained fixed altitudes between 20 and 120 m AGL.

Safety and ethical procedures were followed to minimize disturbance:

• Flights ceased pursuit if lions exhibited flee responses. • Hovering duration was minimized to operational needs. • Drone use complied with Kenyan aviation regulations.

Data Collection

Data were collected over 15 field days between October and December 2024.

Collection procedures included:

• Manual drone approaches at randomized altitudes • Simultaneous RGB and thermal video recording • Live aerial counting of lion groups • Behavioral response observations • Opportunistic individual identification imagery

Quality Control

• Real-time visual verification of imagery quality • Cross-validation between aerial and ground-based lion counts • Identification imagery reviewed for individual recognition markers • Data collection supervised by experienced wildlife monitors familiar with OPC lion populations

📋 Permits and Compliance

Permits Obtained

• Kenyan Civil Aviation Authority (KCAA) operational permission • Wildlife Research and Training Institute (WRTI) Research Permit No. WRTI-0431-06-24 • National Commission for Science, Technology and Innovation (NACOSTI) License No. NACOSTI/P/24/41404

Regulations Followed

• Kenyan UAV operational regulations • Wildlife research permitting requirements • Conservancy operational protocols

Ethics Approval

• Not required

Animal Welfare Protocol

• Experimental altitude treatments designed to assess disturbance thresholds • Flights terminated or modified when lions showed strong negative responses • No prolonged pursuit during flee behavior

Dataset Creation

Curation Rationale

Motivation for Creating the Dataset

This dataset was created to investigate the use of drone-based RGB and thermal imagery for lion monitoring in savannah ecosystems and to evaluate behavioral responses of lions to UAV operations.

Scientific Questions Driving Data Collection

• How do lions respond behaviorally to drones at different flight altitudes? • Can drones improve wildlife population monitoring and counting accuracy? • Can thermal imaging support detection and identification of lions under field conditions?

Gap Filled by This Dataset

Existing wildlife drone datasets rarely combine: • Experimental disturbance assessment • Thermal and RGB imagery • Large carnivore monitoring • Simultaneous aerial and ground-based validation The dataset contributes rare UAV observations of free-ranging African lions under operational conservation conditions.

Intended Use Cases

• Wildlife detection and counting • Animal behavior analysis • Thermal object detection • Conservation technology evaluation • UAV disturbance assessment • Multi-modal wildlife computer vision research

Source Data

Data Collection and Processing

Field Collection

Planning

Sites consisted of naturally occurring lion locations within Ol Pejeta Conservancy, Kenya. Sampling was opportunistic but informed by GPS collar telemetry and ongoing lion monitoring operations.

Collection

• Drone launches performed from field vehicles • RGB and thermal imagery collected simultaneously • Behavioral observations recorded during flights • Data collected during daylight operational windows

Post-Processing

• Review and filtering of usable imagery/video • Cross-referencing aerial observations with ground counts

Software and Tools Used

• Flight planning/control: DJI Pilot 2

Who are the source data producers?

Field Team

• Elena Iannino as pilot, Simon Irungu as assistant pilot, Kelvin Mutethia as ground observer. • Licensed UAV pilot(s) with A1/A2/A3 European certification or equivalent

Local Collaboration

• Ol Pejeta Conservancy staff and monitoring teams • Collaboration with Kenyan wildlife research and regulatory institutions • Research conducted under Kenyan governmental authorization

Annotations

This is a raw telemetry dataset with no animal detection boxes, track identities, or behaviour labels. Researchers wishing to add annotations (detection boxes, identities, behaviours) can use tools such as CVAT. Personal and Sensitive Information Human Subjects: • Flights conducted in a managed conservancy away from public areas Wildlife and Location: • Target species African lion Panthera leo is classified Vunerable by IUCN 2025 • Location corresponds to a well-managed, access-controlled conservancy (Ol Pejeta) • Full GPS coordinates included to support scientific replication Security: • No security concerns • Data collected in coordination with Ol Pejeta Conservancy management

Considerations for Using the Data

Dataset Statistics

Survey Summary

Property	Value
Session Date	October–December 2024 (15 field days total)
Location	Ol Pejeta Conservancy
GPS Coordinates	Approximately 0°00′ N, 36°54′ E
Elevation (ASL)	Mean elevation ~1,810 m above sea level
Target Species	African lion (Panthera leo)
Aircraft	DJI Mavic 3 Enterprise T
Drone Altitudes (AGL)	20, 40, 60, 80, 100, and 120 m
Flight Approach Type	Manual experimental approach flights
Flight Speed	3–10 m/s
Mission Duration per Encounter	Minimum 5-minute hover over lion groups; 5–10 minute follow duration for moving groups when applicable
Data Collection Period	08:00 AM – 07:00 PM
Video Modalities	Simultaneous RGB and thermal recording
Flight Launch Platform	Vehicle-mounted deployment (open-roof vehicle)
Lion Localization Method	GPS collars, VHF telemetry, and opportunistic sightings
Pilot Qualification	Licensed UAV pilot (A1/A2/A3 European license or equivalent)
Synchronization Method	Integrated onboard synchronization between RGB and thermal sensors
Telemetry Availability	Embedded DJI flight metadata and GPS telemetry
Environmental Conditions	Savannah grassland and bushland habitat; daytime conditions
Temperature Range	Mean minimum 12°C; mean maximum 28°C
Weather Conditions	Typical dry-to-transitional savannah conditions
Survey Design	Opportunistic monitoring combined with controlled altitude exposure experiments

Bias, Risks, and Limitations

Geographic Bias

Data were collected exclusively within Ol Pejeta Conservancy, a protected savannah ecosystem in central Kenya. Results may not generalize to: • Human-dominated landscapes • Dense forests • Desert ecosystems • Other lion populations with different habituation levels

Temporal Bias

Flights occurred only during daytime hours (08:00–19:00) from October–December 2024. The dataset does not capture: • Nocturnal behavior • Seasonal variation outside the short rainy/transitional season • Long-term temporal dynamics

Species Bias

The dataset focuses exclusively on African lions (Panthera leo). Other species are absent or incidental.

Environmental Bias

Most imagery originates from relatively open savannah and bushland habitats where aerial visibility is comparatively high.

Detection Bias

• Detection probability likely varies with vegetation density and altitude. • Lions under tree canopy or dense shrub cover may be partially or fully obscured. • Thermal contrast may vary depending on ground temperature and time of day.

Technical Limitations

Image Quality

• Thermal imagery has substantially lower spatial resolution than RGB imagery. • Motion blur may occur during tracking flights or high-speed approaches. • Zoom adjustments and changing gimbal pitch may introduce framing variability.

Coverage Gaps

• No nighttime flights • Limited seasonal coverage • Opportunistic rather than systematic spatial sampling

Ethical Limitations

Animal Welfare

The study explicitly evaluated drone-induced disturbance; therefore, some flights intentionally approached animals at low altitudes. Behavioral disturbance was monitored, and flights were modified or terminated when strong negative responses occurred.

Recommendations

Best Practices for Using This Dataset

For Detection/Tracking Models

• Account for altitude-dependent object scale variation. • Use multi-scale augmentation during training. • Consider separate processing pipelines for RGB and thermal imagery. • Temporal tracking methods may improve detection stability in moving groups.

For Ecological Analysis

• Apply detection probability corrections where possible. • Consider habitat visibility effects when estimating group size. • Avoid extrapolating results beyond savannah ecosystems without additional validation.

For Transfer Learning

• Models trained on 20–120 m AGL imagery may require fine-tuning for other flight altitudes. • RGB-thermal fusion may improve robustness in partially vegetated habitats. • Domain adaptation may be necessary for different UAV platforms or ecosystems.

What This Dataset Should NOT Be Used For

• Estimating absolute population density without accounting for imperfect detection • Generalizing lion behavior across all ecosystems or seasons • Developing systems for unethical wildlife tracking or targeting • Drawing conclusions about nocturnal lion ecology • Inferring long-term behavioral responses from short-duration exposure experiments

Licensing Information

Dataset License: CC BY 4.0 (Creative Commons Attribution 4.0 International) Citation Requirement: Please cite the dataset and the associated paper if you use this data (see Citation section). Code License: MIT License for scripts in this repository

Citation

Dataset

@misc{lion_drone_dataset2025, author = {Elena Iannino}, title = {Lion Responses to Aerial Monitoring Dataset}, year = {2025}, publisher = {Hugging Face}, url = {https://huggingface.co/datasets/EIannino/lion-responses-to-aerial-monitoring-dataset}, }

Paper

@article{lion_drone2025, title = {Drone-based aerial monitoring improves demographic accuracy of lion groups with low behavioral impact}, author = {Iannino, Elena and others}, year = {2026}, note = {in preparation} }

FAIR² Drones Standard

@article{kline2025fair2, title = {Toward a FAIR² Standard for Drone-Based Wildlife Monitoring Datasets}, author = {Kline, Jenna and others}, year = {2025}, note = {In preparation} }

Acknowledgements

We thank: • Ol Pejeta Conservancy management and wildlife monitoring teams • Licensed UAV pilots and field researchers • Kenyan regulatory agencies for permitting support • Wildlife Research and Training Institute (WRTI) • National Commission for Science, Technology and Innovation (NACOSTI) • Data collection team: Simon Irungu and Kelvin Mutethia This work is supported by the WildDrone MSCA Doctoral Network funded by EU Horizon Europe under grant agreement no. 101071224, and by the Innovation Fund Denmark for the project DIREC (9142-00001B).

Validation and Quality Metrics

🤖 AI-Readiness Validation

• Machine-readable metadata: Partial • Structured annotations: TBD • Train/val/test splits: TBD • Class distribution documented: Partial • Data loading code provided: TBD • Example notebooks provided: TBD

🌿 Darwin Core Validation

• Event records documented • Coordinates recorded in UTM • Sampling protocol described • Scientific naming follows accepted taxonomy (Panthera leo)

FAIR² Compliance Checklist

Principle Status

Findable: Pending DOI and registry indexing Accessible: Intended open-access release Interoperable: Standard metadata structure used Reusable: Licensing and provenance partially documented AI-Ready: Multi-modal structured imagery suitable for ML workflows

Glossary

• AGL: Above Ground Level • Darwin Core: Biodiversity metadata standard maintained by TDWG • FAIR²: FAIR principles extended for AI-ready datasets • GSD: Ground Sampling Distance • AGL: Above ground level altitude • UAV: Unmanned Aerial Vehicle • TDWG: Biodiversity Information Standards organization

Dataset Card Authors

Elena Iannino Dataset Card Contact • Primary Contact: Elena Iannino • GitHub: https://github.com/Eiannino • HuggingFace: https://huggingface.co/datasets/EIannino/lion-responses-to-aerial-monitoring-dataset Version History • v1.0.0 (2026-06-08): Initial dataset card draft This dataset card follows the FAIR² Drone Data Standard (Kline et al., 2025) and is modelled on the KABR Behavior Telemetry dataset card.