Evaluating the efficacy of drone-based thermal images for measuring wildlife abundance and physiology
Data files
Nov 22, 2023 version files 2.04 GB
-
AI_Seal_-_Sheet1.csv
-
All_Seal_Temperature_Data_Combined.csv
-
JM_data.csv
-
README.md
Dec 01, 2023 version files 6.02 GB
-
AI_Seal_-_Sheet1.csv
-
All_Seal_Temperature_Data_Combined.csv
-
eSeal_imagery_for_manuscript.zip
-
JM_data.csv
-
README.md
Abstract
Monitoring the population dynamics and behaviors of wildlife is crucial for effective conservation. Although drones can provide a promising alternative to traditional monitoring methods, validation studies must be done to quantify the accuracy of drone-based abundance and distribution estimates in various biological systems. Here, we investigate the use of drones equipped with high-resolution Red-Green-Blue (RGB) and thermal cameras, along with machine learning techniques, for assessments of abundance and physiology in northern elephant seals (Mirounga angustirostris). Aerial images of N=3,415 northern elephant seals were collected at Año Nuevo Reserve during N=24 drone flights, along with ambient air temperatures, wind speed, and time-of-day data. The two-dimensional footprints and surface temperatures of seals were measured from the images. Machine learning algorithms were applied to detect seals in the imagery, and model performance was evaluated. Our findings indicate that seal detection was more accurate using RGB images compared to Thermal images, but that Thermal images could be used to determine that time of day and ambient temperature (but not wind speed or body size) strongly influenced seal external skin temperature. In other words, RGB and Thermal cameras have different strengths and weaknesses that should be carefully considered when designing research studies. Our study highlights the promising integration of drones, thermal imaging, and machine learning for wildlife research, contributing to faster, safer, cheaper, less disruptive, and more accurate wildlife monitoring and conservation efforts.
README: Evaluating the efficacy of drone-based thermal images for measuring wildlife abundance and physiology
https://doi.org/10.5061/dryad.g4f4qrfwp
Code/Software
- Drone Dryad 2023_10_17.R can be used to reproduce the statistical analyses and figures in the paper.
Description of the data and file structure
- AI Seal - Sheet1.csv headers: (number of seals counted for each drone flight)
- Date = drone flight date
- Type = method for image analysis
- Number = number of seals
- All Seal Temperature Data Combined.csv (seal temperature and weather to analyze relationships between the two)
- Date = drone flight date
- Time = drone flight time
- Location = beach location of drone flight
- Seal ID/Number = unique identifier
- Seal Temperature (C) = external skin temperature measurement, in degrees Celcius
- Ambient Temperature (C) = air temperature measurement, in degrees Celcius
- Wind Speed (MPH) = wind speed, in miles per hour
- JM_data.csv (seal size and temperature data to analyze relationships between the two)
- filename = name of image
- type = image type (RGB or Thermal)
- date = drone flight date
- poly_id = unique identifier for each polygon
- class = seal
- area tempc_mean = seal area, meters squared, in Celcius
- tempc_median = median temperature of polygon, in Celcius
- tempc_max = maximum temperature of polygon, in Celcius
- tempc_min = minimum temperature of polygon, in Celcius
- tempc_sd = sd temperature of polygon, in Celcius
- tempc_n = number for polygon, in Celcius
- tempc_var = variance of polygon temperature, in Celcius
- temp_cov = covariance of polygon temperature, in Celcius
The zipped eSeal_imagery_for_manuscript folder contains many imagery and annotation files, as described below:
aligned_imagery
combined
- Includes four 4-channel tiff files that have with red, green, blue, thermal channels.
- "20200224 FINAL_transparent_reflectance_grayscale_warped.tif"
- "20200403 Thermal FINAL_transparent_reflectance_grayscale_warped.tif"
- "20200610_MBBL_Thermal2_transparent_reflectance_grayscale_wraped.tif"
- "20200618 MBBL Thermal Final_transparent_reflectance_grayscale_warped.tif"
manual_annotations
- Include shape files with manually drawn polygons for each seal corresponding to each image. These shape files were converted from cvat XML files and originally annotated on the RGB imagery. These shape files should align with the combined, RGB, and thermal imagery in the aligned_imagery folder.
predicted_polygons
- Includes shape files from model predictions from the RGB model trained as part of this project. There is one file for each of the combined images
rgb
- These are the original RGB images
thermal
- These are the thermal images after the warping process so that the results align with the RGB imagery and the shape files in the aligned imagery folders.
original_thermal
manual_annotations
- Includes shape files for the original thermal images. These shape files were converted from cvat XML files and originally annotated on the thermal imagery. These shape files will **not **align correctly with the warped imagery in the aligned imagery folders.
thermal
- 4 thermal images
manual_annotation_polygons.csv
This is a CSV with 1 row per polygon vertex with the following columns
- Filename - the name of the image file that these polygons were originally annotated on
- Class - seal or NULL for areas with no seals
- x - the x coordinate in pixels from right
- y - the y coordinate in pixels from top
- Height - the hight in pixels of the image
- Width - the width in pixels of the image
- Order - the plotting order of the vertices
- Poly_id - an id for each polygon
In all datasets, NA means that no measurements were made.