Swallow and martin roosts detected on WSR in the Great Lakes region from 2000 to 2020

This dataset contains the resultsresults  three processing steps:
<br>
1) Rendering of Weather Surveillance Radar data and using it as input to the machine learning model described in Cheng et al. (2020) that is capable of detecting and tracking swallow and martin roost signatures.
<br>
2) Manual screening of the model's results, labeling each true positive track according to the type and degree of contamination of the roost by other sources of non-biological scattering.
<br>
3) Grouping of overlapping tracks that can be considered as the same roost dispersal event.
<br>
4) Mean shift clustering of the first detection in each track to find regions of persistent use throughout the years.
<br>
5) Calculation of number of birds per detection, assuming a Purple Martin (<em>Progne subis</em>) radar cross-section (following the procedure proposed by Chilson et al 2012).
<br>
These procedures have been described in detail in the following paper submitted for publication<em>: Long-term analysis of persistence and size of swallow and martin roosts in the US Great Lakes, </em>which has the same authors as this dataset (full citation will be added here once paper is accepted for publication).

<br>
<strong>References:</strong>
<br>
Cheng, Z., Gabriel, S., Bhambhani, P., Sheldon, D., Maji, S., Laughlin, A., &amp; Winkler, D. (2020). Detecting and Tracking Communal Bird Roosts in Weather Radar Data. <em>Proceedings of the AAAI Conference on Artificial Intelligence</em>, 34(01), 378–385. https://doi.org/10.1609/aaai.v34i01.5373
<br>
Chilson, P. B., W. F. Frick, P. M. Stepanian, J. R. Shipley, T. H. Kunz, and J. F. Kelly. (2012). Estimating animal densities in the aerosphere using weather radar: To Z or not to Z? <em>Ecosphere</em>, 3(8):72. http://dx.doi.org/10.1890/ES12-00027.1
<br>

本数据集包含经如下五步处理流程所得的结果：
1) 对天气监视雷达（Weather Surveillance Radar）数据进行可视化，并将其作为输入，送入Cheng等人（2020）提出的可检测并追踪燕类与马丁燕栖息地标定特征的机器学习模型中。
2) 对模型输出结果开展人工筛查，根据非生物散射源对栖息区的干扰类型与程度，为每条真阳性轨迹标注对应标签。
3) 将可视为同一栖息区扩散事件的重叠轨迹进行归组。
4) 对每条轨迹的首次探测点开展均值漂移聚类（Mean Shift Clustering），以识别多年间持续被使用的区域。
5) 假设紫崖燕（<em>Progne subis</em>）的雷达散射截面参数，计算单次探测对应的鸟类数量（计算流程遵循Chilson等人2012年提出的方法）。

上述处理流程已在如下已投稿待发表论文中详细阐述：《美国五大湖区域燕类与马丁燕栖息区的持续性与规模长期分析》，本数据集的作者团队与该论文一致（论文正式录用后将补充完整引用信息）。

**参考文献：**
Cheng, Z., Gabriel, S., Bhambhani, P., Sheldon, D., Maji, S., Laughlin, A., & Winkler, D. (2020). 利用天气雷达数据检测并追踪集群鸟类栖息区. 《美国人工智能协会年会论文集》（*Proceedings of the AAAI Conference on Artificial Intelligence*）, 34(01), 378–385. https://doi.org/10.1609/aaai.v34i01.5373

Chilson, P. B., W. F. Frick, P. M. Stepanian, J. R. Shipley, T. H. Kunz, & J. F. Kelly. (2012). 利用天气雷达估算大气圈动物密度：Z值取舍之辨. 《生态圈》（*Ecosphere*）, 3(8):72. http://dx.doi.org/10.1890/ES12-00027.1