How much can we trust GPS wildlife tracking? An assessment in semi-free-ranging Crested Ibis Nipponia nippon

Introduction

GPS tracking has been increasingly used for wildlife studies in recent decades, which has revolutionized our understanding of individual wildlife movement, behaviour and spatial ecology (Rodgers & Anson, 1994; Cagnacci et al., 2010; Wilmers et al., 2015). Compared to Argos satellite tracking, GPS tracking can provide abundant high-quality data on animal movement, enabling fine-scale studies and conservation even for species in extreme environments with poor accessibility (Block et al., 2011; Hawkes et al., 2013; Ramírez-Macías et al., 2017).

Despite the high quality of tracking data, a number of factors may influence GPS location accuracy and positioning success (ratio between the observed and expected number of locations), including the effects of the terrestrial atmosphere (influencing the satellite signal), satellite constellation (influencing the number of satellites contacted), environment of the transmitters (habitat, topography, and weather), and behaviour pattern and movement intensity of the tagged animal (Trimble Navigation Limited, 1998; Cargnelutti et al., 2007; Recio et al., 2011; Mattisson et al., 2016; Byrne et al., 2017). Therefore, it is crucial to assess the accuracy of GPS tracking to understand the quality of location data and how the environment and animal behaviour influence the data quality, so that the data can be exploited properly and data correction can be processed for fine-scale studies such as those of animal movements in relation to landscape (Rempel & Rodgers, 1997). Rigorous testing of positioning accuracy is an essential prerequisite, especially for newly developed GPS devices.

Tests of GPS transmitter performance have been carried out over the last two decades, mainly on large mammals due to the weight and size limitations of GPS devices (Edenius, 1997; Cargnelutti et al., 2007; Yamazaki et al., 2009; Schwartz et al., 2017). Due to the improvements in microelectronics and battery technology, lightweight GPS transmitters have been developed in recent years that make it possible to track a variety of small-sized animals (Freeman et al., 2009; Pfeiffer & Meyburg, 2015; Liu et al., 2017). Lightweight GPS transmitters are usually deployed close to the ground where they are subjected to complex micro-topographical features, likely resulting in different performance from their larger counterparts, which has not been fully assessed (Recio et al., 2011).

In recent years, a series of lightweight GPS transmitters have been developed in China to meet the need for wildlife tracking. Compared to their mainstream counterparts, these transmitters are much more affordable for researchers, and thus are in increasing need especially in developing countries. However, the performances of these transmitters have not been exploited. In this study, we provide the first performance test of eight lightweight GPS transmitters newly developed in China by attaching them to semi-free-ranging Crested Ibises Nipponia nippon. As part of a re-introduction project, captive Crested Ibises were GPS tagged and confined to two large acclimation cages where they could acquire survival skills in conditions mimicking real habitats before they were released. The tracking data would help understand the movement and habitat selection of the Crested Ibises in the acclimation cage, and thus contribute to evaluate whether these birds could survive in the wild (IUCN, 1998). The objectives of this study were to detect: (1) the positioning success of the tested GPS transmitters; (2) the 95% (i.e., 95th percentile) horizontal positioning error for each Location Class; and (3) the proportion of poor-quality locations (with >100 m positioning error) in each Location Class. We also examined and discussed the influence of surrounding vegetation on transmitter performance by comparing the two test sites, and provided suggestions on data reporting improvement by the manufacturer and the proper use of GPS tracking data.

Methods

Results

The complete dataset for the eight transmitters contained 4084 locations, with an average positioning success of 92.0% (Table 1). Positioning success differed significantly among the eight transmitters (Chi-square test, χ² = 19.073, df = 7, P = 0.008) and was higher at DZ than at TC (Mann–Whitney test, Z =  − 36.862, P < 0.001). The proportion of locations in each LC differed among the eight transmitters (Chi-square test, χ² = 1278.955, df = 21, P < 0.001). On average, the LC B locations accounted for up to 67.5%, while the LC D locations constituted the smallest proportion of 1.4% (Table 1).

The distribution of locations relative to the test cages and bound 95% positioning error at DZ and TC is shown in Fig. 2. Significant differences were detected in the positioning error among locations of the four LCs at both DZ (Kruskal Wallis test, χ² = 7.858, df = 3, P = 0.049) and TC (χ² = 65.646, df = 3, P < 0.001) (Fig. 3). For the pooled data, the 95% positioning error in the locations of LC A, B, C and D was 39 m, 41 m, 230 m and 206 m, respectively (Table 2). Therefore, considering the measurement error caused by the size of the cage, the positioning accuracy of LC A, B, C and D would be 9–39 m, 11–41 m, 200–230 m and 176–206 m, respectively. Despite the high-quality of most locations, there were up to 1.6% of locations with >100 m errors. The proportion of these poor-quality locations was much higher for LC C (8.8%) and D (6.9%) compared to LC A (0.6%) and B (0.8%) (Chi-square test, χ² = 145.162, df = 3, P < 0.001; Table 2). We also observed a higher proportion of poor-quality locations at TC (3.6%) than at DZ (0.7%)(χ² = 45.978, df = 1, P < 0.001).

Discussion

A series of lightweight GPS transmitters have been newly developed in China in recent years to meet the increasing need for wildlife tracking. To the best of our knowledge, our study presented the first performance assessment of GPS transmitters from HQXS, one of the major brands in China, which will provide an essential prerequisite for the application of these transmitters and contribute to the proper use of GPS wildlife tracking data. Since animal behaviours such as locomotion, hiding and diving may influence GPS location accuracy and positioning success (Recio et al., 2011; Mattisson et al., 2016), it is more reasonable to assess transmitter performance on free-ranging animals compared with using a stationary test. The acclimation cages provided ideal test sites for transmitter performance on semi-free-ranging Crested Ibises in conditions mimicking real habitats, which allowed animals to behave more naturally compared with those in tests on controlled animals moving in selected routes (Cargnelutti et al., 2007; Forin-Wiart et al., 2015; De Weerd et al., 2015). However, our results should be interpreted cautiously because the behaviour pattern of the tagged Crested Ibises in the acclimation cage might differ from that of their wild counterparts and consequently affect the positioning accuracy. For instance, the birds would fly relatively less, and be subjected to denser vegetation due to the less open foraging wetlands in the cage. Depending on the frequency and strategy of food supplement, the time budget on foraging and roosting of the Crested Ibises in the cages might also differ significantly from that in the wild (Liu et al., 2008). In addition, because the tagged birds were free-ranging in the acclimation cages, the use of central column of the cages as the reference point might incur an estimate error for the positioning accuracy. To take this into account, we reported the positioning accuracy in a range to provided the basic prerequisite for the proper use of the tracking data.

We detected an average positioning success of 92.0%, which was much higher than that of GPS transmitters deployed on animals (69.3%) and similar to that in stationary tests (94.8%) based on a review of 35 journal articles (Cain et al., 2005). This indicated quite good positioning sensitivity of our tested transmitters. The observed 95% positioning error of the LC A and B locations was estimated to be 9–39 m and 11–41 m, respectively, which was in line with the advertised accuracy (5 m for LC A and 10 m for LC B). The locations in these two LCs constituted 88.7% of the total locations, guaranteeing quite high quality of the tracking data overall. For the LC C and D locations, however, the observed positioning error (200–230 m for LC C and 176–206 m for LC D) was much higher than the advertised accuracy (20 m and 100 m, respectively). Moreover, 6.9–8.8% of the LC C and D locations were of poor-quality with >100 m or even >1,000 m positioning errors. Therefore, although most of the locations are supposed to be quite accurate, we should bear in mind that GPS tracking data may also include a few implausible locations and will need to be properly processed before use. Especially for fine-scale studies such as those of animal movements in relation to landscape, ornithologists are suggested to present the expected location error and its probability, and to make sure that the resolution of habitat data is not greater than the expected location error (Byrne et al., 2017).

Location accuracy is calculated by HDOP or PDOP and classified as four LCs for our tested transmitters (HQXS, 2017). This classification causes considerable information loss, and consequently, we cannot determine the exact accuracy of each location. We suggest the manufacturer reports the HDOP or the PDOP instead of the LCs for each location, so that we can identify and filter implausible locations for fine-scale studies. To compromise between the quantity and quality of the retained data, a HDOP value of 5 or a PDOP value of 10 was suggested as a threshold for gross data (Dussault et al., 2001; D’Eon & Delparte, 2005).

Our results revealed that there was lower positioning success and relatively more poor-quality locations at TC than at DZ, probably due to the relatively smaller cage and dense surrounding trees at TC. This was similar with many previous studies (Cargnelutti et al., 2007; Frair et al., 2010; Recio et al., 2011). Recio et al. (2011) demonstrated that vegetation structure played a critical role in the performance of GPS transmitters, i.e., low vegetation could only incur negligible impact, while dense forest might result in significant influence. This has an important implication in the appropriate use of GPS tracking data. Take Crested Ibises as an example; their daytime locations could be quite accurate since the species usually feeds in open areas with low vegetation. The night-time locations, however, might be subject to roost trees and will need to be carefully filtered for appropriate use.

Conclusions

In conclusion, our study provided the first performance assessment of lightweight transmitters newly developed by HQXS in China, which could provide a large proportion of high-quality data. However, a number of factors, including vegetation structure, might have impacts on the positioning success and accuracy of these GPS transmitters, and result in a number of poor-quality locations. Thus, the tracking data need to be exploited properly and corrected carefully for fine-scale studies. We suggest the HPOD or the PDOP be reported instead of the LC as a measurement of location accuracy for each location to ensure the identification and filtering of implausible locations.

Supplemental Information