Finding turning-points in ultra-high-resolution animal movement data
Finding turning-points in ultra-high-resolution animal movement data
复制标题
寻找超高分辨率动物运动数据的转折点
DOI：
10.1111/2041-210x.13056
复制
发表时间：
2018
影响因子：
6.6
通讯作者：
Potts J
中科院分区：
文献类型：
--
作者：
Potts J
生物记录的最新进展导致动物位置数据达到前所未有的高时间分辨率，有时每秒多次。然而，目前许多用于分析动物运动的方法(例如，步骤选择分析或状态空间建模)都是在考虑到较低分辨率的数据的情况下开发的。为了使这种方法适用于高分辨率数据，我们需要技术来识别轨迹中运动偏离直线的特征。我们认为，运动路径的复杂性，特别是转弯，反映了动物做出的决定，因此转折点与行为生态学家特别相关。因此，我们引入了一种快速、准确的算法来推断高分辨率数据中的转折点。对于分析大数据，速度和可扩展性至关重要。我们在模拟数据上测试了我们的算法，其中不同数量的噪声被添加到散布着转弯的直线段的路径上。我们还在自由放养大麦的数据上演示了我们的算法。我们将我们的算法与现有的断点推断统计技术进行了比较，该算法是线性扩展的，在中端台式计算机上可以在几秒钟内分析数十万个数据点。当标题中的噪声具有±8∘的标准偏差时，它可以完全准确地识别模拟数据中的转折点，这完全在许多现代生物学家的容忍范围内。它具有与现有测试算法相当的精度，并且速度高达三个数量级。我们的算法在RandPython中免费提供，作为处理超高分辨率动物运动数据的第一步，产生了一条稀薄的路径，可用作许多现有步进和转弯分析方法的输入。由此产生的路径由动物明确转弯的点组成，从而为决定运动模式提供了有价值的数据。因此，它提供了作为分析亚秒分辨率数据的起点所需的重要突破。
Recent advances in biologging have resulted in animal location data at unprecedentedly high temporal resolutions, sometimes many times per second. However, many current methods for analysing animal movement (e.g. step selection analysis or state‐space modelling) were developed with lower‐resolution data in mind. To make such methods usable with high‐resolution data, we require techniques to identify features within the trajectory where movement deviates from a straight line.We propose that the intricacies of movement paths, and particularly turns, reflect decisions made by animals so that turn points are particularly relevant to behavioural ecologists. As such, we introduce a fast, accurate algorithm for inferring turning‐points in high‐resolution data. For analysing big data, speed and scalability are vitally important. We test our algorithm on simulated data, where varying amounts of noise were added to paths of straight‐line segments interspersed with turns. We also demonstrate our algorithm on data of free‐ranging oryxOryx leucoryx. We compare our algorithm to existing statistical techniques for break‐point inference.The algorithm scales linearly and can analyse several hundred‐thousand data points in a few seconds on a mid‐range desktop computer. It identified turnpoints in simulated data with complete accuracy when the noise in the headings had a standard deviation of ±8∘, well within the tolerance of many modern biologgers. It has comparable accuracy to the existing algorithms tested, and is up to three orders of magnitude faster.Our algorithm, freely available inRandPython, serves as an initial step in processing ultra high‐resolution animal movement data, resulting in a rarefied path that can be used as an input into many existing step‐and‐turn methods of analysis. The resulting path consists of points where the animal makes a clear turn, and thereby provides valuable data on decisions underlying movement patterns. As such, it provides an important breakthrough required as a starting point for analysing subsecond resolution data.