Data for: Drone surveys of birds foraging in intertidal habitats: A proof-of-concept in Moreton Bay, Australia

We used a DJI Mavic 3 Enterprize to survey a 650m x 250m section of intertidal mudflat surrounding Minjerribah (-27.51, 153.41), an island in Moreton Bay off the east coast of Australia. We flew the drone at 80m height, capturing images at a ground sampling distance of 0.00429mmpix-1, covering 16m x 12m of ground per image. Images were captured in 7 straight lines spaced 75m apart, oriented perpendicular to the shoreline, flight speed was 4ms-1. The drone pilot had their remote pilots licence and flights were approved by the Department of Environment and Science, Ausralin Government, the Queensland Animal Ethics Committee (2021/AE000474), and the Quandamooka Yulluburrabee Aboriginal Corporation. During surveys, I also observed the birds with a spotting scope from the drone launch point 250m away from the site and recorded the number and species of birds within the study site as well as any signs of disturbance. I have four years of experience surveying birds across intertidal mudflats. To make viewing the transects easier, I mosaiced the images from the transects together to create 7 long mosaics, one for each transect. To orthorectify and mosaic the images I used Web Open Drone Map. Within the WebODM application, I uploaded the images from a transect, then created the orthomosaic using the settings: crop: 0, fast-orthophoto: true, ignore-gsd: true, orthophoto-resolution: 0.429, sfm-algorithm: planar. Because orthorectifying the images can introduce distortions that can reduce the quality of the images, I also mosaicked the images without orthorectifying them. To do this, I used Python to manually paste the images into one large image, using the drone positional data stored in the image Exif to position each image correctly. The birds within the images were detected and identified both manually, and using the computer vision tool developed by Wilson et al. (2025). The results of this survey are given in the supporting material provided here. Code CS1_Transect_Spacing.R: Code used to determine error associated with transect drone wildlife surveys based on transect spacing and wildlife abundance, density and clustering. CS2_Mosaic_Images.py: Code to stitch raw drone survey images into a single image for each transect. CS3_Modify_Mission.py: Code to modify the survey kmz file generated by DJI Pilot 2 Application to change the survey height and camera zoom. Data TS1_Species: List of the native bird species found in the Moreton Bay local government area. TS2_Computer_Orthomosaic: Computer vision tool counts of the birds in the orthomosaic images. TS3_Computer_Mosaic: Computer vision tool counts of the birds in the mosaic images. TS4_Manual: Manual counts of the birds in the orthomosaic images. TS5_Ground: Ground counts of the birds conducted at the same time as the drone survey. TS6_Computer_Assessment: Assessment of the computer vision tool performance compared to the manual counts of drone images. TS7_Results: Summary of the results of the survey.

我们使用DJI Mavic 3 Enterprise（大疆御3行业版）对澳大利亚东海岸莫顿湾内明杰里巴岛（Minjerribah，坐标：-27.51, 153.41）周边650米×250米的潮间带泥滩区域开展野外调查。本次无人机飞行高度设定为80米，单张图像的地面采样距离（ground sampling distance）为0.00429毫米每像素，单幅图像覆盖地面范围16米×12米。调查沿垂直于海岸线的7条直线样带完成，样带间距为75米，飞行速度为4米每秒。无人机驾驶员持有官方遥控飞行员执照，本次飞行作业已获得澳大利亚环境与科学部、昆士兰动物伦理委员会（批准编号：2021/AE000474）以及昆达穆卡尤卢布拉比原住民公司（Quandamooka Yulluburrabee Aboriginal Corporation）的许可。

调查期间，笔者在距调查点250米的无人机起飞点位使用单筒望远镜开展鸟类观测，记录研究区域内的鸟类物种与种群数量，同时记录各类干扰活动迹象。笔者拥有4年潮间带泥滩鸟类野外调查的工作经验。

为便于样带数据的可视化与分析，笔者将每条样带的拍摄图像拼接为7幅长全景拼接图，对应每条飞行样带。为完成图像正射校正与拼接作业，本研究使用了Web Open Drone Map（简称WebODM）工具：在WebODM应用中上传单条样带的图像集，随后通过以下参数生成正射影像：裁剪阈值设为0、启用快速正射影像模式、忽略地面采样距离参数、正射影像分辨率设为0.429、运动恢复结构（SfM）算法采用平面模式。考虑到正射校正流程可能引入图像失真，降低影像质量，笔者同时开展了未经过正射校正的图像拼接工作：借助Python程序，利用图像EXIF元数据中存储的无人机位置信息，手动将单张图像对齐拼接为完整大图。

图像中的鸟类通过人工目视识别，以及Wilson等人（2025）开发的计算机视觉工具两种方式完成检测与物种鉴定。本次调查的完整结果详见本文配套的补充材料。

本次研究附带的代码与数据集详情如下：
1. "CS1_Transect_Spacing.R"：用于基于样带间距与野生动物丰度、密度及集群特征，计算样带式无人机野生动物调查误差的统计代码；
2. "CS2_Mosaic_Images.py"：用于将单条样带的原始无人机调查图像拼接为单幅完整图像的代码；
3. "CS3_Modify_Mission.py"：用于修改DJI Pilot 2应用生成的调查KMZ文件，以调整飞行高度与相机变焦参数的代码；
4. "TS1_Species"：莫顿湾地方政府辖区内原生鸟类物种列表；
5. "TS2_Computer_Orthomosaic"：正射影像中鸟类的计算机视觉工具计数结果；
6. "TS3_Computer_Mosaic"：未经过正射校正的拼接影像中鸟类的计算机视觉工具计数结果；
7. "TS4_Manual"：正射影像中鸟类的人工计数结果；
8. "TS5_Ground"：与无人机调查同期开展的地面鸟类实地计数结果；
9. "TS6_Computer_Assessment"：计算机视觉工具性能评估数据集，对比计算机视觉工具与无人机图像人工计数的结果差异；
10. "TS7_Results"：本次鸟类调查的结果汇总数据集。