A road in the middle of one of the last wild reindeer migration routes in Norway : crossing behaviour and threats to conservation

"e development of roads and associated infrastructure has interrupted several traditional migrations of wild reindeer (Rangifer tarandus tarandus) in Norway. "e population in the Setesdal Austhei wild reindeer area still migrates from winter to calving grounds by semi-annually crossing a cabin-lined road through a narrow corridor, in which further anthropogenic development is planned. To understand if and how infrastructure a$ected reindeer migration patterns we studied the movements of 10 female reindeer equipped with GPS collars between 2002 and 2010. First, we identi#ed the start and end of the migration period, and then we compared trajectory parameters (net displacement, step length and turning angles) recorded during migration with those recorded in proximity of the road. "e analysis of the net displacement indicated that during spring migration reindeer moved at a constant pace towards the calving ground covering a net linear distance of 25 km in 40 days. In the middle of migration, reindeer changed travel direction and roamed parallel to the road for ca. 5 days without approaching further, possibly searching for an undisturbed place and time to cross. Reindeer #nally crossed the road before daylight with highly directed movements, increasing their travel speed up to the highest values recorded during migration (4 km/3 hrs vs. 0.5 km/3 hrs). After crossing, reindeer moved quickly toward their calving ground covering the remaining 25 km net distance in less than a week. Migration patterns were markedly a$ected by disturbance during spring, as the road crossings occurred in the period characterized by the high tra%c volume and intense human activities related to Easter holidays; during autumn, on the contrary, the hampering e$ect of the road was minimal. "e results suggest that the current disturbance associated to the road hampers spring migration and might delay the arrival to the calving ground. "e planned construction of a large number of recreational cabins in the migration corridor has the potential to threaten the migration and obstruct the access to the calving ground.


Introduction
Movement is one of the most striking features of the ecology of reindeer (Rangifer tarandus tarandus).e species inhabits the highly seasonal circumpolar environment, and typically follows greening vegetation by migrating from lichen-rich winter habitat with relatively little snow cover to highly productive and insect-free calving and summer pastures.e importance of the search for the optimal seasonal habitat to tness is such that some Rangifer populations perform the longest overland movements in the world (Bergman et al., 2000).However, the ongoing expansion of human-dominated areas A road in the middle of one of the last wild reindeer migration routes in Norway: crossing behaviour and threats to conservation 1 Norwegian Institute for Nature Research (NINA), Tungasletta-2, N-7485 Trondheim, Norway (Corresponding author: manuela.panzacchi@nina.no).
Abstract: e development of roads and associated infrastructure has interrupted several traditional migrations of wild reindeer (Rangifer tarandus tarandus) in Norway.e population in the Setesdal Austhei wild reindeer area still migrates from winter to calving grounds by semi-annually crossing a cabin-lined road through a narrow corridor, in which further anthropogenic development is planned.To understand if and how infrastructure a ected reindeer migration patterns we studied the movements of 10 female reindeer equipped with GPS collars between 2002 and 2010.First, we identi ed the start and end of the migration period, and then we compared trajectory parameters (net displacement, step length and turning angles) recorded during migration with those recorded in proximity of the road.e analysis of the net displacement indicated that during spring migration reindeer moved at a constant pace towards the calving ground covering a net linear distance of 25 km in 40 days.In the middle of migration, reindeer changed travel direction and roamed parallel to the road for ca. 5 days without approaching further, possibly searching for an undisturbed place and time to cross.Reindeer nally crossed the road before daylight with highly directed movements, increasing their travel speed up to the highest values recorded during migration (4 km/3 hrs vs. 0.5 km/3 hrs).After crossing, reindeer moved quickly toward their calving ground covering the remaining 25 km net distance in less than a week.Migration patterns were markedly a ected by disturbance during spring, as the road crossings occurred in the period characterized by the high tra c volume and intense human activities related to Easter holidays; during autumn, on the contrary, the hampering e ect of the road was minimal.e results suggest that the current disturbance associated to the road hampers spring migration and might delay the arrival to the calving ground.e planned construction of a large number of recreational cabins in the migration corridor has the potential to threaten the migration and obstruct the access to the calving ground.and the rapid development of transportation infrastructure interfere with the persistence of large-scale animal movements; many of the most spectacular migrations worldwide have either disappeared, or are in steep decline (Wilcove & Wikelski, 2008;Harris et al., 2009).

, 2012
Before industrial development Norwegian wild reindeer were grouped into two to three large population units that performed seasonal migrations between adjacent mountain systems, with summer ranges typically located in the west and winter ranges in the east.Nowadays, Norwegian wild reindeer are divided into 23 more or less isolated populations.Due to the development of roads, infrastructures and human disturbance, mainly along valley bottoms, several of the most important migration corridors were lost in the past centuries, and two of the most important remaining key corridors were abandoned in the 1980s (Skogland, 1986;Vistnes et al., 2004).Recently, global position-ing system (GPS) radio-collar data of reindeer from several parts of Norway show that most of the remaining movement corridors connecting di erent sub-populations are being used progressively less often or are being abandoned (e.g., Vistnes et al., 2004;Bevanger et al., 2005;Strand et al., 2006;Dahle et al., 2008).e population of Setesdal Austhei is one of the few that still perform distinct seasonal migrations between non-overlapping summer and winter ranges.However, a road surrounded by cabins lies in the middle of the migration route, and reindeer have to cross it through a narrow corridor in which further anthropogenic development is planned.We studied the e ect of this road on migration behavioral and movement patterns of GPS radio-collared females from the Setesdal Austhei population to understand if and how the proposed land development plan might a ect the persistence of this migration.

Study area
e Setesdal Austhei wild reindeer area (7° 78' E, 59° 23' N) is one of the two southernmost remaining populations of wild reindeer in Europe.e area was de ned as a wild reindeer area in 1980 and opened for hunting in 1981; earlier, the reindeer inhabiting Setesdal Austhei were semi-domesticated.e area is con ned by roads, which impede virtually any movement from or to other neighboring reindeer areas, and is bisected into a northern and southern part by a 45 km segment of the public road Rv 45, running parallel to a long dammed lake, Store Bjørnevatn, and cabins (Fig. 1).e average weekly daily tra c (AWDT) on Rv 45 is rather low but varies with time, being higher (AWDT ≈ 500 vehicles/day; Statens vegvesen, 2011) in spring during Easter holidays, which occur most often in the 8 th week of the year, thus coinciding with the spring reindeer road crossing, and lower during the road crossing in autumn (AWDT ≈ 300).e northern area (ca.770 km 2 ) is dominated by mountains up to 1537 m asl, while the southern area (ca.1600 km 2 ) is characterized by low altitude hills and is mostly covered by forests.Further information regarding the study area can be found in Strand et al. (2011).
Reindeer migrate from their winter pastures in the northern area to the calving and summer areas in south.e reindeer GPS radio-collar data (see below) indicates that there are two migration corridors (Fig. 1c, d).One is approximately 1 km wide, and it is located adjacent to the western edge of the dammed lake, between Store Bjørnevatn and Lisle Bjørnevatn; the other is ca. 4 km wide, and is located south west of Lisle Bjørnevatn.However, we suggest that to characterize the location of the migration corridor more precisely further studies and data from a larger sample of animals, collected at ner temporal scales, are needed.e land development plan for Bjørnevatn includes the construction of a total of 161 cabins for recreational use (Kommuneplan for Bjørnevatn, 2008) in the migration corridors (Fig. 1c, d).

GPS radio-collar data
We captured 10 reindeer females in Setesdal Austhei and studied their movement pattern from 2002 to 2010.e animals were darted from helicopter, anesthetized using a mix of Salopine and Ketalar, equipped with GPS collars (Vectronics) with drop-o systems, and reversed using the antidote Narketan.e collars were programmed to record 1 GPS location every 3 hours.Outliers were identi ed and removed using the script provided in Bjørneraas et al. (2010).

Trajectory analyses
To understand whether and how Rv 45 a ects reindeer migration patterns we rst identi ed the migration period for each individual, calculated trajectory parameters during migration, and compared them with trajectory parameters calculated in proximity of the road.For each individual we selected 1 location every 3 hours and calculated the following trajectory parameters: step length (SL), turning angles (TA), net squared displacement (NSD), and net displacement (ND; Kareiva & Shigesada, 1983;Calenge et al., 2009) using the package adehabitat (Calenge, 2006) in R 11.1 (R Development Core Team, 2011).ND is the mean distance between the rst relocation of the trajectory and the last relocation of the current step (NSD is the same measurement, squared; Kareiva & Shigesada, 1983); hence, it describes the position of a step relative to other steps of the trajectory, and can be viewed as illustrating the movement towards or away from a starting point.Even though the analysis of ND throughout the year well describes an animal's movement patterns, the annual NDs of di erent individuals can be compared only if a common starting point is used for its calcula-tion.As a starting point for the calculation of reindeer displacement we used calving sites, all situated in the southern part of the range.To investigate di erences in ND before and after the road crossing, for each individual we performed linear regressions between ND and the number of locations before and after crossing.We compared the slope of the regression before and after crossings with a Wilcoxon paired signed-rank test.SL and TA during migration in di erent seasons, 5 days before/after crossing, and during crossing were compared using t-tests.Finally, we calculated summary statistics of the time of crossing using circular statistics (Jammalamadaka & SenGupta, 2001), which is commonly used when analyzing directional data such as time of day.In particular, we calculated the mean crossing time, and rho, which is a measure of the spread of the data; rho ranges from 0 (all values are concentrated in one point; i.e., zero variance) to 1 (uniform distribution, with values uniformly scattered throughout the circular range; i.e., in our case the clock).

Identifying calving sites
We did not know the exact calving dates of the GPS-collared females in the area.Hence, we estimated their calving dates by analyzing variations in trajectory parameters during the calving season (Van Moorter et al., unpublished data.).As calving occurs after reindeer reach the calving area south from Rv 45, for each individual we analyzed movement patterns during the period starting after the road crossing and terminating on July 15 th .In particular, for each female we calculated the residence time (RT) which quanti es the time (i.e., number of locations) spent within a patch of a given radius, in our case the estimated calving site, before moving further (Barraquand & Benhamou, 2008).RT was calculated within 8 bu ers of di erent radii (10, 25, 50, 75, 100, 150, 200, 500 m), chosen based on information on the specie's calving behavior.We assumed that calving oc-curs on the day with the highest peak in RT.As the mean observed calving dates of wild reindeer in mainland Norway range between May 7 and May 28, and calving typically occurs very synchronously within each wild reindeer area (Reimers et al. 1983;Skogland, 1984;Reimers, 2002), we assumed that the best calving model was the one providing the most synchronous calving season (i.e., the one providing the lowest variance in calving dates).e location of each individual during the average calving date was taken as a starting point for the calculation of its annual NSD, to investigate migration patterns.

Identifying the migration period
We calculated the average start and end date of the migration period by investigating the NSD of those individuals for which a full year of data, from one calving season to the next, was available (n = 21 reindeer-years).When animals move from a summer area to a winter area and back, the plot of their annual NSD appears like a double sigmoid: NSD increases during the autumn migration, reaches an asymptote when the animals reach the winter ground, and decreases again during the spring migration approaching zero if the animal comes back to the starting point (i.e., the calving site).Hence, following Bunnefeld et al. (2011), we tted a double sigmoid to the annual NSD of each individual: (Equation 1) where ␦ is the asymptotic height, s and a are the timing at which the migration reaches half its asymptotic height in spring and autumn, respectively, s and a models the timing elapsed between reaching half and of migration in spring and autumn, respectively, and t as time.Hence, the rst sigmoid represents the autumn migration and the second the return to the calving ground.e in ection point of each sigmoid indicates the time at which an individual has reached the middle distance between summer and winter areas, while the distance between the upper and lower asymptotes re ects the distance between the start and the end point of migration.Equation 1 was tted by non-linear least squares using the nls function in R. For each individual, we de ned the start and the end of the seasonal migration as the number of days required to move from 5% to 95% of each sigmoid (Fig. 2).All trajectory parameters (SL, TA, and NSD) were calculated within the migration period.

Results
We monitored 10 reindeer for a total of 21 reindeer-years; for each reindeer-year we obtained a minimum of 1085 locations, and a maximum of 2921 locations, for a total of 48 256 GPS positions, which can be viewed at http://www.dyreposisjoner.no.

Road crossing
All radio-monitored reindeer crossed Rv 45 twice a year during the seasonal migration.All crossings occurred during one single time bout, except for one case in autumn.All spring crossings occurred between April 11 th and May 8 th (average ± SD: April 23 rd ± 6 days, n =18) and during autumn/winter between Sept. 9 th and Jan. 2 nd (Oct.9 th ± 23 days, n = 22; note: in only two cases reindeer crossed after Oct. 13th).Circular statistics (Jammalamadaka &  , 2012 SenGupta, 2001), required for directional data, such as time of day, showed that most crossings occurred around 8:00 am (rho = 0.98).Seasonal di erences in crossing time were detected (Fig. 3), with a higher frequency of crossings in the rst hours of the day in spring (mean: 7:37 am) with very little variability (rho = 2.9), and no di erences in autumn (mean: 6:18 pm, large variability: rho = 0.7); the di erence between spring and autumn in the observed number of crossings in each of the eight 3-hour time bins was signi cant (X 2 = 27.943,df = 7, P < 0.001).

Calving dates
e method which performed best was the RT calculated within a radius of 150 m.Based on this method we estimated 20 calving dates (i.e., for 10 adult females followed in 1 to 3 years).Calving occurred from May 6 th to May 23 rd (average: May 13 th ± 4.6 days, SD).Calving occurred on average 21.6 ± 7.1 days (n = 17; range: 9 -32 days) after the spring road crossing.In two cases the precision of our estimate was low, and did not allow us to estimate calving dates. is means that either the animals did not calve, or that the mother did not change movement patterns markedly during calving days.

Migration time
e double sigmoid model described well the reindeer migratory movements, as it explained on average 94.2 % (± 1.7, n = 10) of the variance in individual annual NSD.However, the method did not always perform well in identifying with high precision the start and end day of migration.e reason is that NSD is strongly dependent on the location of the starting point, and as common starting point we chose the calving sites, which are not too spatially concentrated within the summer range.Hence, calving sites located closer to the road led to a slightly delayed estimated start of migration, while calving sites located at the extreme south of the distribution range led to a slightly early start of migration.However, our results indicate that autumn and spring migration started on average on Sept. 20 th (± 15 days) and March 15 th (± 21 days), lasted 49 and 69 days, respec-tively, and covered 56.4 ± 5.4 km.A summary of the estimated migration periods, calving periods, and of the observed crossing dates is presented in Table 1.

Net displacement
If the presence of the road had no e ect on the , 2012 spatial behavior during migration, we would expect the ND calculated from the crossing point to decrease linearly while the reindeer were approaching the road, and increase linearly after the road crossing (Fig. 5a). is is what we observed in autumn (Fig. 5b), as the slopes of the regression lines calculated for each individual between ND and the number of locations did not di er before and after crossing (V = 87, n = 18, P = 0.644).On the contrary, during spring the slopes of the regression lines were much higher after crossing compared to when approaching the road (V = 165, n = 18, P < 0.001).Indeed, during spring ND decreased slowly as the reindeer approached the road, it suddenly dropped before the road cross, and increased sharply after the crossing, with very little individual variability (Fig. 5c), due to the fact that after crossing the road movements became much more directional, with reindeer heading straight and quickly to their calving ground.While during spring reindeer migrated slowly towards the road covering a net linear distance of ca. 25 km in ca.40 days, after the crossing they quickly drifted away from the road covering the same net distance in a week, and after 20 days they were displaced 35 km south of the road (Fig. 5b).

Discussion
Several wild reindeer populations in Norway have abandoned their historical migration routes due to the expansion of human activities, namely transportation infrastructure and recreational facilities.We present the case of the population in Setesdal Austhei, which still performs distinct seasonal migrations between winter and calving grounds by crossing a road located in the middle of the migration corridor.Our results, however, show that all trajectory parameters are markedly altered in proximity of the road during both spring and autumn, as while crossing the road reindeer increase their travel speed to reach the highest values record-ed during migration, perform highly directional movements, and cross during hours when the tra c is minimal.Most importantly, as the spring road crossings occur in the period characterized by the high tra c volume and intense use of private cabins related to Easter holidays, the overall spring migration patterns change as reindeer approach the road, and remain altered up to the arrival to the calving ground.During spring reindeer migrate at a constant pace towards the calving ground covering a net linear distance of 25 km in 40 days.As rein- deer approach the road, they change travel direction and roam parallel to the road for ca. 5 days without approaching further, possibly searching for an undisturbed place and time to cross.As the pregnant reindeer cross the road, they speed up and quickly head toward calving ground covering the remaining 25 km net distance in less than a week. is suggests that the current disturbance associated with the road hampers spring migration and may delay the arrival to the calving ground.Interestingly, during autumn we could only identify short-term e ects of the road (i.e., travel speed increased and movements were highly directional during crossing), but no longer-term alteration of the migration patterns could be detected.is is probably due both to the lower tra c volume during autumn and to the lower physiological urgency of reaching the winter range.
e impact of infrastructure on Rangifer distribution and habitat use has been extensively documented worldwide during the last decades (Mahoney & Schaefer, 2002;Vistnes et al., 2004;Johnson et al., 2005;Nellemann et al., 2010).Several studies reported range shifts in response to the construction of infrastructure (e.g., Nellemann et al., 2001;Dyer et al., 2002), or documented the avoidance of potentially suitable habitat for several kilometers around disturbed areas, with little or no sign of habituation in the following decades (Vistnes & Nellemann, 2008;Nelleman et al., 2010) or centuries (Schaefer, 2003).Not all studies, however, reached similar conclusions, and it has become clear that the responses to disturbance can vary greatly spatially and temporally depending on the type and size of the infrastructure, its location with respect to the core area of reindeer distribution range, variations in the intensity of disturbance, winter harshness, food availability and population density, and the degree of wildness of the population (Skogland, 1986;Carruthers & Jakimchuk, 1987;Duchesne et al., 2000;Mahoney & Schaefer, 2002;Reimers et al., 2007).
e list of confounding factors complicating the assessment of reindeer responses to human disturbance likely increases if the infrastructure is located in the middle of a migration corridor, as during migration reindeer may be forced to tolerate, up to a certain degree, higher levels of disturbance to avoid potentially serious population consequences.Dahle et al. (2008), for example, documented a clear avoidance of a bu er area surrounding a road, but no aversion could be detected along the same road in proximity of a traditional migration corridor.Hence, the drive for migration is such that it is often di cult to recognize the tipping point turning a permeable barrier into an e ective barrier to migration in time useful for implementing mitigation measures.e construction of a railway and a parallel road in Norway, for example, obstructed a key migration corridor for decades.As the population, trapped in one of the seasonal ranges, reached high densities and faced overgrazing and starvation, reindeer started crossing the infrastructure occasionally during harsh winters.However, coincident with the increase in tra c and a reduction in deer numbers, all crossings nally stopped during the 1980s (Skogland, 1986).Hence, if the drive for migration is high, a given infrastructure can represent a barrier under some circumstances but not others, and the long-term population response may become evident only when the implementation of mitigation measures would be more complex and costly.
Previous analyses (Panzacchi et al., 2011) showed that the motivation pushing the Setesdal Austhei population to continue migrating is high, as the optimal calving and winter grounds are located on the opposite sides of the road, and none of seasonal ranges seem to contain preferred resources year-round.Our study shows that even though reindeer still migrate through a narrow bottleneck on the west side of the road, migratory patterns are markedly , 2012 altered by the disturbance associated with the road, especially during the spring migration to the calving ground.Considering the high drive for migration, our results suggest that the Setesdal Austhei population is already forced to tolerate relatively high levels of disturbance during spring migration to reach the preferred calving ground.
e long-term consequences of the present intensity of disturbance associated with the road, cabins and dam are di cult to predict.However, based on previous experiences (e.g., Skogland, 1986), it is plausible that the planned construction of additional 161 private cabins in the migration bottleneck (Strand et al., 2011;Fig. 1c, d) will have critical consequences for the persistence of migration and for population viability.e data for the 10 GPS radio-collared females indicate that the migration corridors are very small: one is located in a very small geographic bottleneck of ca 1 km between two lakes, and the other covers an area of ca. 4 km (Fig. 1c, d). is implies that the implementation of mitigation measures at this stage might still be relatively cost-e cient.We are currently collecting GPS data for more animals on a ner temporal scale (i.e., 15 min) to be able to assess more precisely the location of the migration corridor and analyze the relative e ect of factors a ecting the time and location of crossing, in order to assist the development of e cient mitigation measures.We recommend developing urgent, targeted and sound conservation plans to secure a future for the migration of the Setesdal Aushtei population.

Fig. 1 .
Fig. 1.The Setesdal Austhei study area.The location of a) winter (February) and b) calving range May15th -June 30th are represented by a 95% kernel density of the GPS radio-collar locations of 10 female reindeer monitored from 2002 to 2010.In (b) the yellow dots represent the locations of the calving events (n = 20) estimated for each individual each year.The dashed rectangle in b depicts the location of the area illustrated in (c), reindeer GPS radiocollar locations indicating the location of the two migration corridors, and (d), details of the land development plan for Bjørnevatn, indicating the planned location of the construction of several cabins for recreational use.

Fig. 2 .
Fig. 2. Example of the migration parameters estimated for one focal individual based on its annual Net Square Displacement (NSD).The smooth line represents the gratory movement from the calving range to the winter range and back(Bunnefeld et al., 2011).Grey lines iden-

Fig. 3 .
Fig. 3. Reindeer crossing frequency of road Rv 45 in Setesdal Austhei with respect to time of day during spring and autumn migration.

Fig. 5 .
Fig. 5. Average Net Displacement (ND, km), which illustrates the speed of the movements towards and away from the road, calculated from the crossing point (coded 0) during 350 locations (ca.45 days) before and after the crossing of road Rv 45 during both spring and autumn migration (a), and separately during spring (b) and autumn (c); dashed bars indicate standard errors.The dotted line in (a) indicates a theoretical ND if reindeer moved between the summer and winter areas at a constant speed (i.e., if the presence of the road had no effect on the spatial behavior during migration).The shaded area and the dashed line in (b) indicate, respectively, the estimated calving range and average estimated calving date.