BirdNest AI — Biscayne National Park Colonial Bird Monitoring

Monitoring Intelligence

14 Years Across Biscayne Bay

Every survey flight, every nest annotation, every colony — mapped and tracked from 2010 to 2024. Hover the chart, click the map, then dive into the full dataset below.

Nest Detections Over Time

i

Annotations, detections, and flights across 14 survey years — spot colony booms, crashes, and gaps at a glance.

Survey Coverage Map

i

Seven colony sites across Biscayne National Park — every dot is a nest count waiting to be studied.

Explore the full dataset & colony map →

Dataset Overview

The Cartography
of Nesting

Traditional monitoring of coastal waterbirds in remote park areas is labor-intensive and prone to error. Our AI framework automates this process using sub-meter resolution aerial imagery spanning 15 years of Biscayne National Park colonial surveys.

Tap a card to reveal the numbers

+

🎯

Automated Precision

YOLOv5s6 detects nests across thousands of aerial images, replacing slow manual counts.

0.458 best mAP@0.5 on held-out test set

1280px inference resolution (sub-meter aerial)

A single flight that once took analysts days to count by hand is now processed in seconds.

+

🗂️

4-Class Labels

Every nest is labeled by occupancy state — not just "present" or "absent".

138,400 occupied nests

11,238 non-occupied

8,495 chicks · 3,611 eggs

Occupancy state is what tells biologists whether a colony is thriving or collapsing.

+

📅

Temporal Trends

15 years of continuous surveys let us see colonies shift, recover, or disappear.

2010–2024 survey span (14 years)

117 survey flights · 7 mapped colonies

The 2020 dip (just 106 images) marks the COVID-19 survey gap — visible right in the timeline.

Survey Timeline — Images per Yeari

Annotation Class Distributioni

161,744

Total nests

click a slice

Colony Monitoring Sites

The High-Resolution Frontier

7 of 9 colony monitoring sites mapped in Biscayne National Park (2 sites pending GPS coordinates). Click a marker for site details.

Model Performance

Detection at the
Speed of Science

YOLOv5s6 trained on 15,759 aerial images achieves mAP@0.5 = 0.458 on the held-out test set, enabling rapid, consistent nest detection across island colonies.

Architecture: YOLOv5s6 · Image size 1280 px
Dataset split: 70% train / 20% val / 10% test
Best mAP@0.5: 0.458 at epoch 43 / 50
4 classes: occupied · eggs · chicks · non-occupied

Run Live Inference →

0.458

Best mAP@0.5 ⓘ

43

Best Epoch ⓘ

50

Total Epochs ⓘ

mAP@0.5 vs Epoch

Training Loss vs Epoch (box · obj · cls)

Official Training Results (from Dataverse)

Confusion Matrix — Test Set

Live Demo

Run the Detection Model

Upload an aerial photograph to run the YOLOv5s6 nest detection model in real time. Achieves mAP@0.5 = 0.458 · 4 classes · 1280 px inference size.

Upload an aerial photograph to run the YOLOv5s6 nest detection model. The model was trained on all 15,759 images and achieves mAP@0.5 = 0.458 on the held-out test set. It detects four classes: occupied nests, eggs, chicks, and non-occupied nests.

Confidence threshold: 0.25 | Image size: 1280px

📤 Upload Aerial Image

Open Data

Contribute to the Archive

Join the National Park Service and scientific community in refining the models that protect our avian neighbors. Access open-source datasets or deploy detection tools in your own research.

Published Dataset — FIU Research Data

doi:10.34703/gzx1-9v95/UD9HTD

🔗 Open in FIU Dataverse

For Researchers

Access our Python-based detection SDK and datasets. Download the full 113 GB image archive, 161,744 YOLO annotations, trained weights, and training notebooks.

🖼️

Images

15,759 aerial JPEGs

🏷️

Annotations

161,744 YOLO boxes

🤖

Model

YOLOv5s6 weights

API Documentation →

For Public

View real-time nest detection maps and impact reports. Use our live inference demo to run detection on your own aerial imagery. Dataset is in the public domain — free to download, use, and share.

Public domain under 17 U.S.C. § 105 — US federal government works. No license required.

Public Map → Live Demo

Citation

Cite This Work

APA

Juhasz, L., & Guan, B. (2025). Colonial Bird Nest Detection Dataset: Biscayne National Park Aerial Monitoring 2010–2024. Florida International University GIS Center. Prepared for the National Park Service, South Florida/Caribbean Inventory and Monitoring Network. https://doi.org/10.34703/gzx1-9v95/UD9HTD

BibTeX

@dataset{juhasz_guan_2025_birdnest,
  author    = {Juhasz, Levente and Guan, Boyuan},
  title     = {Colonial Bird Nest Detection Dataset:
               Biscayne National Park Aerial Monitoring 2010--2024},
  year      = {2025},
  publisher = {Florida International University GIS Center},
  doi       = {10.34703/gzx1-9v95/UD9HTD},
  url       = {https://dataverse.fiu.edu/dataset.xhtml?
               persistentId=doi:10.34703/gzx1-9v95/UD9HTD}
}

For questions about the data, contact Dr. Kevin Whelan (NPS) — kevin_r_whelan@nps.gov | Dr. Levente Juhasz (PI) — ljuhasz@fiu.edu

Counting What We Cannot Afford to Lose