Assessing the potential of unoccupied aerial vehicles and deep learning to survey wild turkey populations
Fuente:
Wiley Open Access
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Artículo Open Access |
| Published: |
Wiley
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1867017473508769792 |
|---|---|
| author | Allison R. Schumacher Imad Eddine Toubal Gani Rahmon Tim Krock Jason B. Hardin Kannappan Palaniappan Michael E. Byrne |
| author_facet | Allison R. Schumacher Imad Eddine Toubal Gani Rahmon Tim Krock Jason B. Hardin Kannappan Palaniappan Michael E. Byrne Allison R. Schumacher Imad Eddine Toubal Gani Rahmon Tim Krock Jason B. Hardin Kannappan Palaniappan Michael E. Byrne |
| collection | Wiley Open Access |
| contents | Assessing the potential of unoccupied aerial vehicles and deep learning to survey wild turkey populations Allison R. Schumacher Imad Eddine Toubal Gani Rahmon Tim Krock Jason B. Hardin Kannappan Palaniappan Michael E. Byrne Wildlife Society Bulletin Abstract Despite importance as a game species and concerns about regional declines in abundance and productivity, there is no widely accepted method to accurately estimate wild turkey ( Meleagris gallopavo ) abundance that is applicable across landscapes and spatial scales. Unoccupied aerial vehicles (UAVs), also known as drones, have emerged as a new tool for surveying wildlife populations that are difficult to survey using traditional methods, which can be combined with artificial intelligence based deep‐learning methods to automate detection and counting of animals within video data recorded by UAVs. Our objectives were to assess thermal imaging UAV surveys and deep learning as a potential tool for surveying wild turkey populations. We flew UAVs with thermal cameras across several ecoregions in Texas, USA, along 200‐m long transects centered over roost locations of GPS‐tagged wild turkeys and opportunistically located turkey vulture ( Cathartes aura ) and black vulture ( Coragyps atratus ) roosts during winters 2021 and 2022. We trained a deep‐learning algorithm to detect, track, and count wild turkeys in recorded thermal videos, as well as assess discriminability between wild turkeys and vultures. We applied a modified Horvitz‐Thompson estimator that accounts for overall detection probability, false positives, and duplicate detections as functions of weather and habitat variables to correct raw deep‐learning counts and estimate wild turkey abundance in videos. The true population was within the estimated 95% confidence interval of abundance for 81.4% of videos ( n = 59), with an average mean absolute error between estimated and true abundance of 2.6 wild turkeys. The rate of misclassification of wild turkeys as vultures was 0.46, vultures as turkeys was 0.19, although the available training data of vultures was limited ( n = 8 survey nights). Our results suggest winter nocturnal thermal UAV surveys combined with deep learning offers a potentially powerful new tool to survey wild turkey populations with methodology that could be standardized over space and time. Future work should prioritize collecting training data in various landscapes and methods to minimize misidentification of wild turkeys and vultures in regions of North America where those species ranges overlap in winter. 10.1002/wsb.1639 http://creativecommons.org/licenses/by/4.0/ |
| doi_str_mv | 10.1002/wsb.1639 |
| format | Artículo Open Access |
| id | wiley_oa_10_1002_wsb_1639 |
| institution | Wiley Open Access |
| license_str_mv | http://creativecommons.org/licenses/by/4.0/ |
| publishDate | 2025 |
| publisher | Wiley |
| record_format | wiley_oa |
| spellingShingle | Assessing the potential of unoccupied aerial vehicles and deep learning to survey wild turkey populations Allison R. Schumacher Imad Eddine Toubal Gani Rahmon Tim Krock Jason B. Hardin Kannappan Palaniappan Michael E. Byrne Wildlife Society Bulletin Assessing the potential of unoccupied aerial vehicles and deep learning to survey wild turkey populations Allison R. Schumacher Imad Eddine Toubal Gani Rahmon Tim Krock Jason B. Hardin Kannappan Palaniappan Michael E. Byrne Wildlife Society Bulletin Abstract Despite importance as a game species and concerns about regional declines in abundance and productivity, there is no widely accepted method to accurately estimate wild turkey ( Meleagris gallopavo ) abundance that is applicable across landscapes and spatial scales. Unoccupied aerial vehicles (UAVs), also known as drones, have emerged as a new tool for surveying wildlife populations that are difficult to survey using traditional methods, which can be combined with artificial intelligence based deep‐learning methods to automate detection and counting of animals within video data recorded by UAVs. Our objectives were to assess thermal imaging UAV surveys and deep learning as a potential tool for surveying wild turkey populations. We flew UAVs with thermal cameras across several ecoregions in Texas, USA, along 200‐m long transects centered over roost locations of GPS‐tagged wild turkeys and opportunistically located turkey vulture ( Cathartes aura ) and black vulture ( Coragyps atratus ) roosts during winters 2021 and 2022. We trained a deep‐learning algorithm to detect, track, and count wild turkeys in recorded thermal videos, as well as assess discriminability between wild turkeys and vultures. We applied a modified Horvitz‐Thompson estimator that accounts for overall detection probability, false positives, and duplicate detections as functions of weather and habitat variables to correct raw deep‐learning counts and estimate wild turkey abundance in videos. The true population was within the estimated 95% confidence interval of abundance for 81.4% of videos ( n = 59), with an average mean absolute error between estimated and true abundance of 2.6 wild turkeys. The rate of misclassification of wild turkeys as vultures was 0.46, vultures as turkeys was 0.19, although the available training data of vultures was limited ( n = 8 survey nights). Our results suggest winter nocturnal thermal UAV surveys combined with deep learning offers a potentially powerful new tool to survey wild turkey populations with methodology that could be standardized over space and time. Future work should prioritize collecting training data in various landscapes and methods to minimize misidentification of wild turkeys and vultures in regions of North America where those species ranges overlap in winter. 10.1002/wsb.1639 http://creativecommons.org/licenses/by/4.0/ |
| title | Assessing the potential of unoccupied aerial vehicles and deep learning to survey wild turkey populations |
| topic | Wildlife Society Bulletin |
| url | https://wildlife.onlinelibrary.wiley.com/doi/10.1002/wsb.1639 |