Barro Colorado Island 50-ha Plot Aerial Photogrammetry (2018-2024): Orthomosaics, Digital Surface Models, Point Clouds, Raw Images, and Globally/Locally Aligned Timeseries

Data are available for download at: https://smithsonian.dataone.org/datasets/BCI_50ha_drone_productsMetadata of the products comes as comma separated value file named metadata_all.csv , variables_description.csv and a README.txtThis dataset is part of a larger initiative monitoring forests in Panama using drones (unoccupied aerial vehicles), an initiative led by Dr. Helene Muller-Landau at the Smithsonian Tropical Research Institute. As part of this initiative, we have been collecting repeat imagery of the 50-ha forest dynamics plot on Barro Colorado Island (BCI), Panama, since October 2014 (see Garcia et al. 2021a, b for data products for 2014-2019).The current dataset encompasses a total of 121 flight dates between February 2, 2018, and March 18, 2024, all conducted using a DJI Phantom 4 Pro drone and a FC6310 camera. The flight interval was near monthly through January 2023, and near weekly thereafter.The drone operated at a flight altitude of 180 meters above the take-off point. The imagery was captured with a front and side overlap of 79%. The drone maintained a speed of 5.5 meters per second, with the camera set to a shutter interval of 7 seconds.The data products available within this dataset include orthomosaics, digital surface models (DSMs), point clouds, raw images, and detailed processing reports. All these products are provided in the UTM Zone 17N coordinate reference system (EPSG: 32617).Orthomosaics and DSMs are available in GeoTIFF format, while point clouds are provided as LAS files in versions 1.2 and 1.4. The original drone imagery was independently processed for each date using the Agisoft Metashape Pro 2.0 Python API (Agisoft LLC) following a standardized workflow. Key processing parameters included the highest setting for photo alignment, a medium setting for point cloud construction, and aggressive point filtering. For more details, you can refer to the functions in the script available at this Github link: https://github.com/VasquezVicente/ForestLandscapes/blob/main/LandscapeScripts/UAV_photogrametry.py. This workflow produced the initial point clouds,orthomosaics, and digital surface models, which are provided as-is.Globally and locally aligned timeseries is a selected group of 106 orthomosaics with the DSM in the fourth band. These were horizontally and vertically aligned in two alternative ways, producing two time series: one based on local alignment, and one based on global alignment.(No fixed ground control points were available: the focal area is entirely forested and in a protected area, with no possibility of establishing permanent clearings.) Depending on the application, the globally or locally aligned products may be preferred. The local alignment algorithms at times produced warping which have the potential to remove some of the signals of interest (e.g., obscuring tree crown growth or damage, as the edges of crowns are aligned across dates). The global aligned time series is free of such warping but due to photo alignment and matching during photogrammetry processing, it can display noticeable systematic alignment errors towards the edges of the plot (the center tends to be well-aligned). The workflow performs several tasks such as combining digital surface models into the 4th band and cropping to the area of interest. Horizontal alignment was based on the arosics.CoReg module (Scheffler, 2017) applied to the 4-band raster (including RGB and the digital surface model). This was followed by global vertical alignment, implemented simply by subtracting the median elevation difference. We took advantage of airborne LiDAR data(doi:10.60635/C34W2W) and associated RGB photogrammetry collected on May 26, 2023, as a basis for the alignment. The drone-acquired raster for the closest date (May 23, 2023) served as the main reference. Subsequently, an iterative procedure was implemented in which each next closest date was aligned to the preceding one, creating a chain of alignments. This process continued backward and forward in time, creating a series of aligned datasets in which each date served as the reference for the next and previous dates. We provide with the grids used for the local alignment in the form of comma separated value files, it contains the x and y position in the form of spatial points. Careful attention was maintained throughout these processes to uphold the integrity and quality of the data. For a comprehensive understanding of the alignment methodology, please refer to the GitHub repository housing the complete code: https://github.com/VasquezVicente/ForestLandscapes/blob/main/LandscapeScripts/50ha_aligment_v2.py

数据可从以下链接下载：https://smithsonian.dataone.org/datasets/BCI_50ha_drone_products。该数据集的元数据包含名为metadata_all.csv、variables_description.csv及README.txt的逗号分隔值文件。
本数据集隶属于史密森尼热带研究所Helene Muller-Landau博士主导的巴拿马森林无人机监测项目。该项目自2014年10月起，对巴拿马巴罗科罗拉多岛（Barro Colorado Island，简称BCI）的50公顷森林动态监测样地开展重复影像采集工作（相关2014-2019年的数据产品详情可参阅Garcia等人2021a、b的研究）。
本次发布的数据集涵盖2018年2月2日至2024年3月18日期间共121个飞行日期，所有飞行任务均采用DJI Phantom 4 Pro无人机及FC6310相机完成。2023年1月前，飞行间隔接近每月一次，此后调整为接近每周一次。
无人机飞行高度设置为起飞点上方180米，影像的航向与旁向重叠率均为79%，飞行速度为5.5米/秒，相机快门间隔设置为7秒。
本数据集提供的数据产品包括正射影像（orthomosaics）、数字表面模型（digital surface models, DSM）、点云、原始影像及详细处理报告，所有产品均采用UTM Zone 17N坐标系（EPSG: 32617）。其中正射影像与数字表面模型采用GeoTIFF格式存储，点云则以LAS 1.2及1.4版本格式提供。
所有日期的原始无人机影像均采用Agisoft Metashape Pro 2.0 Python API（Agisoft LLC）按照标准化工作流独立处理，关键处理参数包括最高精度的照片对齐、中等精度的点云构建，以及激进的点云滤波。详细处理流程可参阅GitHub脚本：https://github.com/VasquezVicente/ForestLandscapes/blob/main/LandscapeScripts/UAV_photogrametry.py中的相关函数。本工作流生成的初始点云、正射影像及数字表面模型均按原始状态提供。
全局与局部对齐时间序列：我们选取了106幅正射影像，并将数字表面模型作为第四波段，通过两种对齐方式生成两类时间序列：一类基于局部对齐，另一类基于全局对齐。（注：本研究区域为全森林覆盖的保护区，无法设立固定地面控制点。）根据具体应用场景，可选择全局对齐或局部对齐产品。局部对齐算法有时会产生扭曲，可能会移除部分感兴趣的信号（例如，对齐树冠边缘时会掩盖树冠生长或损伤的信号）。全局对齐的时间序列无此类扭曲，但由于摄影测量处理中的照片对齐与匹配步骤，在样地边缘会出现明显的系统性对齐误差（样地中心对齐效果较好）。
该工作流包含多项任务，例如将数字表面模型合并至第四波段，并裁剪至研究区域范围。水平对齐基于arosics.CoReg模块（Scheffler, 2017），应用于包含RGB波段与数字表面模型的四波段栅格数据。随后通过减去中值高程差实现全局垂直对齐。我们以2023年5月26日采集的机载激光雷达（LiDAR）数据（DOI:10.60635/C34W2W）及同期RGB摄影测量数据作为对齐基准，以最接近该日期的2023年5月23日无人机采集的栅格数据作为主要参考。随后采用迭代对齐流程：将每个后续日期的影像对齐至前一个日期的影像，形成对齐链；该流程沿时间正反两个方向执行，生成一系列对齐数据集，其中每个日期均可作为相邻日期的对齐参考。
我们还提供了用于局部对齐的CSV格式栅格文件，其中包含以空间点形式存储的x、y坐标。整个处理过程始终注重数据的完整性与质量。如需了解完整的对齐方法学细节，请参阅包含全部代码的GitHub仓库：https://github.com/VasquezVicente/ForestLandscapes/blob/main/LandscapeScripts/50ha_aligment_v2.py