Dataset Card for NOAA-PIFSC-ESD-ESA-CORAL-ICRA-Dataset

Dataset Details

Dataset Description

This dataset contains annotated benthic coral reef imagery of Isopora crateriformis (ICRA) coral species. The dataset is formatted for object detection tasks.

• Bounding box annotations for ICRA coral colonies
• Stratified train/validation/test splits for robust model evaluation
• License: Public Domain (NOAA Open Data)

Dataset Composition:

• Total Images: 470 images
• Training Images: 327 (69.6%)
• Validation Images: 71 (15.1%)
• Test Images: 72 (15.3%)
• Train/Val/Test Split Ratio: 70:15:15

Annotation Format

• Format: YOLO (You Only Look Once)
• Label Files: .txt files with bounding box annotations
• Annotation Structure: class x_center y_center width height (normalized coordinates)
• Total Annotations: 470 label files (1:1 correspondence with images)

Classes

• Number of Classes: 1
• 0 = ICRA (Isopora crateriformis)

Uses

Direct Use

This dataset is designed for computer vision tasks, specifically:

• Object Detection: Training YOLO or similar models to detect and localize ICRA coral colonies
• Transfer Learning: Fine-tuning pre-trained models for coral-reef benthic imagery
• Baseline Development: Establishing performance baselines for marine habitat monitoring systems
• Research: Academic studies on coral reef ecology and automated monitoring

Suitable Applications:

• Automated coral species identification in underwater imagery
• Monitoring ICRA population distribution and abundance
• Developing AI-assisted tools for marine biologists
• Educational purposes for marine ecology and computer vision

Out-of-Scope Use

• Regulatory or policy decisions without domain expert review
• Habitat trend inference without survey design/statistical context
• **Identification of species other than ICRA or applications not related to benthic habitats
• Real-time autonomous decision-making without human oversight
• Applications outside U.S. Pacific Island regions without additional validation

Getting Started

Dataset Structure

The dataset follows the YOLO directory structure:

NOAA-PIFSC-ESD-ESA-CORAL-ICRA-Dataset/
├── README.md
└── dataset/
    ├── data.yaml          # Dataset configuration
    ├── images/
    │   ├── train/        # 327 images
    │   ├── val/          # 71 images  
    │   └── test/         # 72 images
    └── labels/
        ├── train/        # 327 .txt annotation files
        ├── val/          # 71 .txt annotation files
        └── test/         # 72 .txt annotation files

Using with YOLO

The dataset is ready to use with YOLOv5, YOLOv8, YOLO11 or other YOLO-based frameworks:

# Example with YOLOv8
from ultralytics import YOLO

# Load a model
model = YOLO('yolov11n.pt')  # nano model

# Train the model
results = model.train(
    data='dataset/data.yaml',
    epochs=100,
    imgsz=640,
    batch=16,
    name='icra_detection'
)

# Validate the model
metrics = model.val()

# Run inference
results = model.predict(source='dataset/images/test', save=True)

Data Format

data.yaml:

train: images/train
val: images/val
test: images/test
nc: 1
names: ['ICRA']

Label Format (YOLO format): Each .txt file contains one bounding box per line:

class x_center y_center width height

Where all coordinates are normalized (0-1) relative to image dimensions.

Example:

0 0.4407 0.3601 0.3791 0.2164
0 0.5414 0.4976 0.0995 0.0663

Dataset Creation

This dataset was created to support detection of Isopora crateriformis (ICRA) coral colonies in the Pacific Islands region. The data is aligned with NCRMP/PIFSC benthic image workflows and reproducible splits for model development and validation.

Source Data

Data Collection and Processing

Collection Methodology:

1. Photoquadrat surveys collected by trained divers following NCRMP standardized protocols
2. Images captured at benthic survey sites across U.S. Pacific Island territories
3. Survey design incorporates Stratified Random Sampling (StRS) methodology
4. Quality control procedures applied during field operations

Processing Pipeline:

1. Images retrieved
2. Expert annotation of ICRA coral colonies with bounding boxes
3. Conversion to YOLO format with normalized coordinates
4. Stratified splitting into train (70%), validation (15%), and test (15%) sets
5. Quality assurance reviews by PIFSC analysts

Who are the source data producers?

Primary producers are NOAA PIFSC ESD ESA team and collaborators conducting NCRMP field missions and annotation operations in U.S. Pacific Islands regions.

Annotations

Who are the annotators?

Trained NOAA/PIFSC/ESD analyst

Bias, Risks, and Limitations

Known Limitations:

• Geographic Bias: Data is primarily from the U.S. Pacific Island region(Am.Samoa); model performance may degrade when applied to other geographic areas
• Environmental Bias: Imagery collected under specific lighting conditions, water clarity, and seasonal constraints typical of the survey periods
• Temporal Coverage: Data represents specific time periods; coral morphology and environmental conditions may vary across years
• Camera Variability: Images captured with different camera systems across multiple missions may introduce technical variability
• Species Specificity: Model trained only on ICRA; not suitable for detecting other coral species without additional training
• Label Uncertainty: Some annotations may contain uncertainty due to image quality, partial occlusion, or difficult viewing angles

Potential Risks:

• Over-reliance on automated detection without expert validation
• False positives/negatives in novel environmental conditions
• Misapplication to regulatory decisions without appropriate validation
• Extrapolation beyond the trained domain (Pacific Islands, ICRA species)

Recommendations

Model Training:

• Use class-balanced sampling or loss weighting if expanding to multi-class scenarios
• Implement data augmentation to improve robustness (rotation, color jittering, scaling)
• Monitor for overfitting given the moderate dataset size

Evaluation and Deployment:

• Maintain strict train/val/test separation to prevent data leakage
• Validate model performance on independent sites and time periods before operational deployment
• Consider ensemble methods or uncertainty quantification for production use
• Establish confidence thresholds appropriate for your use case

Domain Expertise:

• Always consult marine experts for interpretation of results
• Validate automated detections with manual(human) review, especially for critical applications
• Document model performance metrics and limitations clearly for end users
• Combine automated detection with traditional survey methods for comprehensive monitoring

Personal and Sensitive Information

• No personal, sensitive, or identifiable information is present.
• Images are of marine environments only.

---

Glossary

• ICRA: Isopora crateriformis - A coral species found in Pacific reef ecosystems
• NCRMP: National Coral Reef Monitoring Program - NOAA's long-term monitoring program for coral reef ecosystems
• PIFSC: Pacific Islands Fisheries Science Center - NOAA research center in Honolulu, Hawaii
• ESD: Ecosystem Sciences Division - Division within PIFSC focused on ecosystem research
• StRS: Stratified Random Survey - Statistical sampling methodology used in NCRMP surveys
• ESA: Endangered Species Act - U.S. law under which certain coral species are listed
• Photoquadrat: Standardized photographic method for documenting benthic communities
• YOLO: You Only Look Once - Object detection algorithm and annotation format
• Bounding Box: Rectangular annotation marking object location (x_center, y_center, width, height)

Metadata / Sources

Primary Data Repository:

• Google Cloud Storage: ESA Sites Photogrammetric Imagery
• InPort Metadata Record #76360

Related NCRMP Resources:

• Benthic cover from StRS annotations
• Benthic cover from climate-station annotations
• Benthic images from StRS Sites
• Benthic images from Fixed climate stations

Dataset Card Contact

For questions or inquiries, contact:
Michael Akridge – Michael.Akridge@noaa.gov

Disclaimer

This repository is a scientific product and is not official communication of the National Oceanic and Atmospheric Administration, or the United States Department of Commerce. All NOAA project content is provided on an ‘as is’ basis and the user assumes responsibility for its use. Any claims against the Department of Commerce or Department of Commerce bureaus stemming from the use of this project will be governed by all applicable Federal law. Any reference to specific commercial products, processes, or services by service mark, trademark, manufacturer, or otherwise, does not constitute or imply their endorsement, recommendation or favoring by the Department of Commerce. The Department of Commerce seal and logo, or the seal and logo of a DOC bureau, shall not be used in any manner to imply endorsement of any commercial product or activity by DOC or the United States Government.