Detours in long-distance migration across the Qinghai-Tibetan Plateau: individual consistency and habitat associations

Satellite tracking and location data processing

We tracked four Pallas’s Gulls (denoted G1 to G4) breeding at Qinghai Lake (99.36–100.77°E, 36.32–37.25°N; altitude = 3,200 m) in northwestern China by satellite telemetry (12 g PTT-100; Microwave Telemetry Inc., Columbia, MD, USA; duty cycle 8 h on/15 h off) during 2006–2009 (Zhang et al., 2014). We received the Doppler-derived PTT location information from the CLS/Service Argos satellite tracking system that reports 1-sigma error radii of 150–1,000 m to reflect the location accuracy of location classes (LC) 1, 2 and 3 (Argos, 1996); LC 0, A, B and Z are not assigned accuracy estimates but provide a large amount of relatively accurate data that can be used to analyse gross movements after being filtered (Soutullo et al., 2007). We used the Douglas Argos-Filter Algorithm (v. 8.50) to identify and remove implausible auxiliary Doppler locations based on the distance moved, movement rate, and turning angle (Douglas et al., 2012). Only locations recorded during complete migratory journeys (those that covered the entire distance between the breeding and wintering grounds) were included in this study.

Parameters for migration timing and routes

For each bird and migratory season, we recorded a set of parameters to describe migration timing and routes that included departure date, arrival date, duration, shortest distance (the great-circle distance between the breeding and wintering site), cumulative distance (the total great-circle distance of the migration segments) and the average migration speed (calculated as the cumulative distance/duration). We plotted telemetry locations in Google Earth Pro (version 7.1.8.3036; Google Inc., Mountain View, CA, USA) to determine migration timing parameters by visually inspecting the raw data and determining the habitat of the stopover sites. If the PTTs did not transmit continuously, we used the median date between the last point at the previous location and the first point of the new location to calculate timing. We plotted the telemetry locations in ArcView GIS (version 3.3; Environmental Systems Research, Redlands, CA, USA) and used the XTools extension to compute route distance. We also calculated the straightness index as the ratio between the shortest distance and the cumulative distance to roughly measure the degree of the migration detours.

To explore the spatial and temporal patterns of the migration of Pallas’s Gulls across the QTP, we recorded the parameters of trans-QTP duration, trans-QTP distance and trans-QTP speed (calculated as the trans-QTP distance/trans-QTP duration). For this study, the QTP was defined as the areas above 3,000 m in altitude, and we contoured it in ArcView GIS and computed the trans-QTP distance of each migration route using the XTools extension. To assess the topographical preference for migration, we determined the mean altitude of each route, which was automatically generated in Google Earth by plotting the migration routes in the software. The southern and the northern QTP is crossed by two representative latitudes, 29°N and 35°N, so we recorded the longitudes crossing these two latitudes to test the spatial flexibility of the repeated migration routes across the QTP.

Longitudinal detours in migration routes

To assess the degree of longitudinal detours in migration routes by individuals within each season, we compared each observed route with 1,000 randomized routes that were automatically generated by rearranging each segment of the observed route using the Alternate Animal Movement Route extension for ArcView GIS (Jenness, 2005; López-López, García-Ripollés & Urios, 2014). Each randomized route had the same number and sequence of segments but a different segment orientation from the observed route. We recorded the longitudes crossing each 3° latitude interval from 23°N to 44°N for each observed route and the 1,000 simulated routes (i.e., the randomly generated routes). If a bird doubled back during the migration, resulting in two or more longitude values crossing a specific latitude, the most longitudinally deviated value was used. If the observed longitudes for a bird fell below the 2.5th percentile or above the 97.5th percentile (equivalent to a two-tailed test with P = 0.05) of the 1,000 randomly generated longitudes, the observed longitudes were considered to have significantly deviated from the possible routes (López-López, García-Ripollés & Urios, 2014) (Fig. S1).

To measure the extent of longitudinal detours, we calculated the linear distance of the maximum longitudinal deviation of each observed route from the expected shortest route at 3° latitude intervals from 23°N to 44°N. When the gulls overshot the breeding site or wintering site, we calculated the linear distance of the maximum longitudinal deviation of the overshot segment from the breeding site or wintering site. In this case, we also calculated the linear distance of the maximum latitudinal deviation of overshot stopover sites from the breeding site or wintering site as a measurement of latitudinal detour (Fig. S2).

Detours and habitat association

Pallas’s Gulls exclusively use open water as stopover sites for refuelling, so the availability of open water may have profound influence on migration route selection (Zhang et al., 2011a). We analysed the proximity to water bodies based on the China National River System Data Set (http://www.pudn.com/downloads608/sourcecode/graph/detail2478372.html), which comprises information on third-order rivers and lakes with areas ≥0.01 km² in China. The composition of lakes and rivers was extracted from zones with 100 km buffers around the observed and expected shortest routes in ArcView GIS, and the differences between the zones were then compared to assess migration detours and habitat associations. We did not analyse detours and habitat associations for the southern stretches of the route outside China because all individuals were funnelled through a narrow corridor to the wintering site, which was spatially near the expected shortest routes.

Statistical analyses

We checked the data for normality to determine whether the use of parametric or nonparametric tests. We used paired t-tests or Wilcoxon signed-rank tests to compare the differences between average migration speed and trans-QTP speed for each route and determine the differences in proximity to water bodies between the observed and expected shortest routes. We compared the longitudinal variation across the QTP by ANOVA, with individual and season as factors, and we analysed the seasonal differences in migration route and timing parameters using GLM mixed models, with individual as a random factor. We also used GLM mixed models to compare the between- and within-individual variation in the timing and route of repeated migration tracks. Based on the derived variance components, we calculated the repeatability (i.e., intra-individual correlation coefficient; Lessells & Boag, 1987) of migration timing and route to measure how consistently individuals differed from each other. All statistical analyses were performed using the SPSS statistical software package (version 22.0, IBM 2013), and the results were calculated as a mean ± SD, with a 0.05 level of significance based on two-tailed tests.

Ethical note

All data collected as part of this study were approved by the National Bird Banding Center of China (No. NBBC20060407). The field work was approved by the State Forestry Administration (No. 33 Forestry Protection (2002)).

Spatial patterns of detours

Notable detours were observed along the migration routes in both seasons (Fig. 1). Gulls exhibited increasing longitudinal detour extents from south to north across the migration routes in both autumn (chi-square test for trend, X² = 6.627, df = 1, P = 0.010) and spring (X² = 4.151, df = 1, P = 0.042) (Fig. 2), with the largest detours being 1,068 km at 38°N in autumn and 685 km at 32°N in spring. Similarly, the percentage of the observed routes that significantly deviated from the simulated routes increased from south to north in autumn (X² = 5.955, df = 1, P = 0.015).

After departure from Qinghai Lake in seven (n = 12) autumn migration seasons, three of the four gulls exhibited notable northward migration outside of the QTP to stopover sites at lower altitudes (Δ altitude = 2,337 ± 385 m compared with Qinghai Lake, n = 7; Fig. 1), with an average latitudinal detour of 627 ± 275 km (range = 133–799 km, n = 7). Similarly, in five (n = 10) spring migration seasons, two of the four gulls overshot the breeding grounds to farther northern stopover sites at lower altitudes (Δ altitude = 2,029 ± 139 m compared with Qinghai Lake, n = 5) and then reverse migrated back to Qinghai Lake, with an average latitudinal detour of 168 ± 24 km (range = 140–193 km, n = 5).

Relative to the expected shortest routes, gulls travelled 1,210 ± 1,002 km (range = 81–2,240 km, n = 12) or 61.3 ± 50.4% (range = 4.5%–123.2%, n = 12) of additional detour distance in autumn and 860 ± 648 km (range = 63–1,762 km, n = 10) or 45.3 ± 34.6% (range = 2.8%–95.0%, n = 10) in spring. Gulls covered significantly longer trans-QTP distances along the observed routes than along the shortest routes both in autumn (paired t-test, t = 6.52, df = 11, P < 0.001) and in spring (t = 3.97, df =9, P = 0.003).

Temporal patterns of detours

Gulls used different routes in different seasons and followed a loop migration pattern (Fig. 1). On average, gulls followed larger longitudinal detours in autumn than in spring at all eight measured latitudes from 23°N to 44°N (Wilcoxon signed-rank test, Z =  − 2.521, P = 0.008) (Fig. 2). The effect of season on longitudinal variation was strong at 35°N (ANOVA, F = 16.46, df = 1, P = 0.001, h² = 0.540) but weak at 29°N (F = 0.68, df = 1, P = 0.424), reflecting that the loop migration primarily occurred across the northern part of the QTP. A significant individual × season interaction was detected both at 35°N (F = 9.47, df = 3, P = 0.001, h² = 0.670) and 29°N (F = 18.33, df = 3, P < 0.001, h² = 0.797), indicating that different individuals migrated in different loops or even followed loops in different directions (i.e., clockwise or counter clockwise; Fig. 1).

No significant seasonal differences were observed in the migration route parameters, including cumulative distances, trans-QTP distances, straightness and mean altitude (GLM mixed models, F_1,20 = 0.10–0.31, P = 0.618–0.778). However, the spring migration duration was significantly shorter than that in autumn (GLM mixed models, F_1,20 = 22.38 P = 0.018). Additionally, no significant differences were detected in trans-QTP duration, average speed and trans-QTP speed (GLM mixed models, F_1,20 = 0.23–2.88, P = 0.188–0.424) between autumn and spring (Table 1).

Gulls migrated faster across the QTP than during the entire journey both in autumn and spring (Table 1), but the differences were not significant (Wilcoxon signed-rank test, autumn: Z = 1.413, P = 0.170; spring: Z = 0.867, P = 0.418).

Detours and habitat associations

During the autumn migration, we found significantly increased (paired t-test, t = 2.90, df = 11, P = 0.015; Fig. 3) proximity to water bodies along the observed detour routes (1.3 ± 0.7%, n = 12) than on the expected shortest routes (0.7 ± 0.2%, n = 12). However, we did not detect a significant difference in proximity to water bodies (paired t-test, t = 0.49, df = 9, P = 0.635) between the observed detour routes (0.8 ± 0.7%, n = 10) and the expected shortest routes (0.7 ± 0.1%, n = 10) in spring.

Individual consistency in detours

The effect of the individual was significant for almost all migration variables, except for several timing variables in autumn (Table 2). Tests of within-individual variation showed that migration variables were more repeatable in spring than in autumn and in route than in timing. In autumn, the departure date was highly flexible (r =  − 0.07), but the arrival date was highly repeatable (r = 0.85); however, both the departure and arrival date were consistent, for which individuals differed between years by an average of ± 6.1 d and ± 3.5 d, respectively. In spring, both the departure date (r = 0.84) and arrival date (r = 0.78) were highly repeatable; however, gulls showed consistent departure date and somewhat flexible arrival date, for which individuals differed between years by an average of ± 4.5 d and ± 14.0 d, respectively. The longitude crossing 35°N (r_autumn = 0.82, r_spring = 0.91) and 29°N (r_autumn = 0.75, r_spring = 0.89) in both seasons had high repeatability. In autumn, the within-individual variation of the longitude crossing 35°N and 29°N between years averaged 201 km (6.4% of the accumulated distance) and 224 km (7.1% of the accumulated distance), respectively. In spring, the variation averaged 84 km (3.0% of the accumulated distance) and 138 km (4.9% of the accumulated distance), respectively. Therefore, gulls showed higher consistency in routes across the QTP in spring than in autumn.