A comparison of turn identification methods on high-frequency movement trajectories reveals potential comparability issues between studies of movement ecology

Abstract

High-frequency animal tracks must often be subsampled to allow a simple analysis of the movement on the most meaningful scale for the respective study. One way of achieving this is to identify biologically significant turns, compared to heading changings caused by noise. Many turn identification methods have been developed, but the accuracy and consistency of such methods have rarely been validated against ground truth trajectories with known true turns and noise. We analyze simulated tracks with known parameters as well as two empirical tracks and identify turns with 10 different frequently used resampling methods. We assess the specificity and sensitivity of identifying the location of turns and compare the known mean step length and turn angle of the paths with the resampled trajectories. We found great accuracy differences between, and sometimes within, methods, even on simulated tracks of the same characteristics. Results of some methods were also highly sensitive to the user-set threshold the method requires (e.g. max angle). Overall, the best-performing methods in this study were DP and MRPA, methods used in human mobility research, and TPA, which is mostly used in primate research. We thus advise caution when comparing results of studies using different resampling methods and recommend justifying the use of the resampling method in addition to quantifying the sensitivity of results to the threshold value. This study is also an appeal to authors of novel turn identification methods to consider thorough comparisons in different scenarios with a wide range of previous methods, including those developed outside the movement ecology discipline.