Wet season range fidelity in a tropical migratory ungulate


Correspondence author. E-mail: tmorrison80@gmail.com


1. In migratory populations, the degree of fidelity and dispersal among seasonal ranges is an important population process with consequences for demography, management, sensitivity to habitat change and adaptation to local environmental conditions.

2. Characterizing patterns of range fidelity in ungulates, however, has remained challenging because of the difficulties of following large numbers of marked individuals across multiple migratory cycles and of identifying the appropriate scale of analysis.

3. We examined fidelity to wet season (i.e. breeding) ranges in a recently declining population of wildebeest Connochaetes taurinus Burchell in northern Tanzania across 3 years. We used computer-assisted photographic identification and capture–recapture to characterize return patterns to three wet season ranges that were ecologically discrete and topographically isolated from one another.

4. Among 2557 uniquely identified adult wildebeest, we observed 150 recaptures across consecutive wet seasons. Between the two migratory subpopulations, the probability of remaining faithful to wet season areas ranged between 0·82 and 1·00. Animals from a non-migratory segment of the population (near Lake Manyara National Park) were rarely observed in other wet season ranges, despite proximity to one of the migratory pathways.

5. We found no effect of sex on an individuals’ probability of switching wet season ranges. However, the breeding status of females in year i had a strong influence on patterns of range selection in year + 1, with surviving breeders over three times as likely to switch ranges as non-breeders.

6. Social-group associations between pairs of recaptured animals were random with respect to an individual’s wet season range during the previous or forthcoming wet seasons, suggesting that an individual’s herd identity during the dry season does not predict wet season range selection.

7. Examining fidelity and dispersal in terrestrial migrations improves our understanding of the constraints that migrants experience when they face rapid habitat changes or fluctuations in environmental conditions.


Migratory wildlife often exhibits a remarkable tendency to return to the same sites, ranges and routes each year, a behaviour termed ‘fidelity’ (Greenwood 1980; Waser & Jones 1983; Sawyer et al. 2009). Examples are widespread and include many birds (Greenwood 1980), fish (Thorrold et al. 2001), amphibians (Reading, Loman & Madsen 1991) and mammals (Dobson 1982). Fidelity is thought to benefit individuals by increasing their familiarity with the location of resources and predators in specific areas (Greenwood 1980; Switzer 1997). An important consequence of fidelity is that it generates demographic and genetic substructure within populations, which promotes local adaptation and assortative mating (McNamara & Dall 2011). High fidelity, however, increases the vulnerability of migratory populations when habitat quality declines in particular ranges (Wiens, Rotenberry & Vanhorne 1986; Sutherland 1998; Cooch, Rockwell & Brault 2001). For example, high site fidelity to areas that were heavily exploited by humans likely compounded the rapid declines observed in many whale populations during the 20th century (Clapham, Aguilar & Hatch 2008). Some species exhibit a more flexible migration strategy in which some or all individuals appear capable of switching sites, possibly in response to changing environmental conditions (Sutherland 1998). Because this flexibility can have important demographic, genetic and conservation implications for migratory populations (Doligez et al. 2003; Bolger et al. 2008), there is considerable interest in first characterizing patterns of fidelity and dispersal in migratory populations, and secondly in identifying the mechanisms that generate these patterns (Shuter et al. 2011).

Characterizing ‘fidelity’, however, remains challenging for several reasons. First, it requires following a relatively large number of marked individuals across multiple migratory cycles in multiple sites or ranges (Belisle 2005). Even with large samples, separating mortality from dispersal is difficult or impossible if individuals move over large areas and if some locations remain unsampled (Webster et al. 2002). Furthermore, defining the scale and boundaries of seasonal ‘sites’ can be arbitrary, particularly in highly mobile species that have large, potentially overlapping, seasonal ranges (Schaefer, Bergman & Luttich 2000). Because non-territorial herbivores, such as many ungulates, move over relatively large areas within any given season and are unattached to nesting or denning sites, fidelity for these species may be most appropriately measured at the scale of the entire range (Hansen, Aanes & Sæther 2010). Linear displacement and range overlap across years have been used as measures of fidelity (Sweanor & Sandegren 1989) or as measures of interannual habitat selection following disturbance (Faille et al. 2010). However, many such approaches suffer from a lack of an appropriate null model (Schaefer, Bergman & Luttich 2000). Here, we quantify fidelity at the scale of the entire seasonal range and measure it along a continuum of possible inter-annual movement strategies, from active dispersal (i.e. a lower probability of returning to the same range than chance predicts), random movement (i.e. an equal proportion of individuals from a given range in 1 year move to all ranges the following year) and fidelity (i.e. a higher probability of returning to the same range than chance predicts).

A number of hypotheses have been proposed to explain the control of fidelity and dispersal in migratory species. Extremely high fidelity is presumably caused by genetic controls, memory or cultural imprinting (Sutherland 1998) because animals fail to act adaptively when conditions change. Several migratory shrubsteppe bird species, for instance, continued to exhibit high fidelity to breeding sites despite the experimental removal of 75% of their habitat, presumably to the detriment of their fitness (Wiens, Rotenberry & Vanhorne 1986). In many populations, however, only a portion returns to the same range or site in consecutive years. While this pattern does not preclude the possibility that genetics or cultural imprinting play roles in determining range selection, it suggests that animals may follow conditional decision rules in which fidelity or dispersal is based on past experiences or environmental or social cues (Hoover 2003). Many birds (Switzer 1997; Hoover 2003) and some ungulates (Tremblay et al. 2007) consistently return to breeding sites or ranges in years following successful breeding events (although see: Paton & Edwards 1996). Switzer (1993) termed this the ‘win-stay: lose-switch’ strategy. An alternative to basing range selection decisions on past experiences is to respond to reliable environmental cues that predict the quality of habitats in the future. This strategy involves a response to either extrinsic factors, such as recent rainfall and plant growth (Holdo, Holt & Fryxell 2009), or intrinsic forces such as density-mediated habitat selection (Morris 1987). Ungulates have largely been viewed as employing this strategy, enabling them to exploit resource gradients during the growing season (‘summer’ in temperate latitudes and ‘wet’ season in tropical latitudes; Fryxell & Sinclair 1988; Albon & Langvatn 1992; Sawyer et al. 2009). This view emphasizes the importance of environmental cues in determining where and when animals migrate (Albon & Langvatn 1992).

Individual attributes are also known to play important roles in resource selection and local movement decisions (Fischhoff et al. 2007). For example, sex and age often correlate with patterns of fidelity and dispersal (Greenwood 1980; Harvey et al. 1984). In polygynous mammals, males and juveniles (of either sex) tend to have greater probabilities of dispersing than females and older individuals, presumably to reduce inbreeding, lower mate competition or retain preferred sites (Dobson 1982). Identifying causes of fidelity and dispersal becomes more complicated in gregarious species, where attraction to social groups may override an individual’s directional bias (Gueron, Levin & Rubenstein 1996; Couzin et al. 2005). For example, if an individual that selects area ‘A’ in year i joins a social group composed mostly of individuals from area ‘B’, the individual may be more likely to migrate to area ‘B’ in year + 1. This type of behaviour could be facilitated by either leadership of a few well-informed or experienced individuals or by group consensus decision-making (Conradt & Roper 2005).

Wildebeest are one of the best-studied migratory ungulates owing to long-term monitoring efforts in the Serengeti–Mara Ecosystem in East Africa (Sinclair et al. 2007). Serengeti wildebeest appear to respond to forage and nitrogen availability in new grass growth within the perceptual range of individuals (80–100 km) (Holdo, Holt & Fryxell 2009). At the large scale, migratory movements may be a strategy to maximize energetic intake (Wilmshurst et al. 1999) or the ingestion of new grass growth (Boone, Thirgood & Hopcraft 2006). Notably, these explanations focus on animal responses to environmental cues as the key to understanding local and regional scale habitat selection. Just east of the Serengeti ecosystem, wildebeest in the Tarangire–Manyara Ecosystem (TME) migrate 40 to 120 km between seasonal ranges. The TME provides a convenient location to study range fidelity because wildebeest occupy three spatially and ecologically discrete ranges during the wet season and congregate within two discrete areas in the dry season (Fig. 1) (TCP 1998). The TME population is also sufficiently small (c. 6000 individuals) that we can use photographic identification methods to individually follow animals across annual cycles (Morrison et al. 2011). We characterize range fidelity across three migration cycles and develop multistate capture–recapture models (MCR) to quantify the probability of migrating to alternative wet season ranges. MCR models are a robust method for estimating transition probabilities while accounting for potential survival and recapture differences between ranges (Brownie et al. 1993). We use these models to quantify the degree of range fidelity and test four hypotheses. (i) If environmental cues influence wet season range selection, we would expect the probability of switching ranges to be greater towards one range than others in particular years (here called ‘directionality’). We also test whether an individual’s (ii) sex or (iii) a female’s breeding success during the previous wet season influences their probability of returning to the same wet season range in consecutive years. Finally, and (iv) we examine the role of social forces in determining where individuals spend the wet season by comparing pairs of individuals recaptured within the same herd during the dry season and testing whether they come from, or move to, the same wet season ranges. If individuals associate at random during the dry season, it supports the hypothesis that social-group identity does not influence an individual’s range selection decisions in the wet season.

Figure 1.

 Map of the Tarangire–Manyara Ecosystem, Tanzania. Solid arrows denote the two primary pathways used by wildebeest as they migrate between Tarangire National Park in the dry season and the Simanjiro Plains and Northern Plains during the wet season. Approximate seasonal ranges are outlined with dotted lines. The non-migratory population inhabits Lake Manyara National Park.

Materials and methods

Study area

The TME lies in the eastern branch of the Great Rift Valley in northern Tanzania and encompasses roughly 20 000 km2 (Fig. 1). Precipitation is highly variable across time and space (mean, 656 mm year−1; coefficient of variation, 36.4%) and largely falls between November and May (Foley & Faust 2010). Both calving and mating are highly synchronous in wildebeest and occur within short periods during the wet season (Fig. 2; Estes 1976). In most years, calving and breeding occur in three distinct wet season ranges: the Northern Plains (NP), the Simanjiro Plains (SP) and along the northern shore of Lake Manyara (LM) (TCP 1998). Once surface water dries out (c. July), animals migrate to Tarangire and Lake Manyara National Parks, where they spend the dry season. SP and NP lie c. 140 km apart and are separated by a chain of forested volcanic mountains (Losiminguri, Burko and Mondouli mountains). Thus, direct movements between these ranges within the same wet season are unlikely and would require passing through Tarangire National Park (TNP) (Fig. 1). The LM subpopulation is thought to be non-migratory (Borner 1985; Prins & Douglas-Hamilton 1990); though, no quantitative data exists about mixing with animals from Tarangire. On the western edge of TME, the Gregory Rift Wall forms a major geographic barrier between TME wildebeest and adjacent Ngorongoro–Loliondo wildebeest, preventing any significant recent gene flow between populations (Georgiadis 1995). Connectivity with other nearby ecosystems (e.g. Amboseli and Shompole) is possible, although any movement that does occurs is likely at low levels because of the considerable distance and lack of suitable habitat (open grassland) between the areas.

Figure 2.

 Generalized annual cycle for wildebeest in the Tarangire–Manyara Ecosystem, Tanzania. Calving and mating occur during brief periods in the wet (i.e. breeding) season and are thought to be highly synchronous across space (Estes 1976).

Historically, the TME wildebeest inhabited four or five distinct wet season ranges (Lamprey 1964; Borner 1985). However, since the 1940s, human population and agricultural expansion outside of Tarangire and Lake Manyara National Parks have increased four to sixfold (Mwalyosi 1991), reducing the connectivity in the ecosystem and causing substantial habitat loss (TCP 1998). Between 1988 and 2001, wildebeest in TME have experienced an estimated sixfold decline, from roughly 40 000 to 6 000 individuals (TAWIRI 2001).

Study design and data

The presence of natural variation in shoulder stripe patterns of adult (>2 years old) wildebeest allowed us to use computer-assisted photographic identification to compile encounter histories across three wet and dry seasons of 2005–2007 (Fig. 2). In each study year, we initiated photo sampling in May (i.e. c. 2·5 months after the calving pulse) when wildebeest were on their wet season ranges. Sampling of each range took between 2 and 14 days, depending on conditions and local animal densities. We sampled each wet season range twice per year (except in 2005, which had only one sample) within a robust design framework (Table S1; Pollock 1982). We also collected photographs in Tarangire and Lake Manyara National Parks during the dry season (October–November). These dry season photographs were used in the analysis of social-group associations and to establish connectivity between dry season ranges.

To collect photographs, we drove all roads and main tracks within the SP, NP and LM ranges once per secondary period and photographed individuals within herds that we encountered. We aged and sexed animals on the basis of horn morphology and body size (Watson 1967) and collected GPS locations of each herd. Herds were defined as groups of wildebeest in which no individual was >100 m from the next closest individual. We photographed individuals on their right sides (stripe patterns were not symmetrical on both sides), perpendicular to the length of the animal. Photographs were collected from a stationary vehicle at a distance of ∼10–100 metres during daylight hours using a 6·1 megapixel Pentax istD camera (Pentax Corporation, Denver, CO, USA) with a 400-mm Sigma telephoto zoom lens. For each herd, we attempted to collect as many photographs as there were adults in the group. In some cases, herds moved away or joined other herds before we had collected the target number of photographs. Because the identity of individuals was not known at the time of photographing, some herds were unknowingly photographed multiple times while others were not photographed at all. Overall, we aimed to photograph 40–50% of all adults within each range to balance sample size and coverage. Actual capture rates (i.e. the percentage of the population identified) were much lower (capture probabilities, P, ranged from 0.02 to 0.22) because some individuals were photographed multiple times and c. 30% of images were too poor in quality to be used for matching (Table S1).

Computer-assisted photograph identification

We used two software platforms to identify individuals based on stripe patterns: one for adult males and one for adult females. The first program matched all males; however, the availability of a more time-efficient program led us to switch platforms prior to identifying females. Both platforms yielded similar probabilities of identifying individuals. The ‘male’ platform was developed by Conservation Research Ltd. (Hastings, Hiby & Small 2008) and involved three preprocessing steps: (i) users digitally outlined the margins of each individual within an image and placed reference markers on several key features, such as the nose and base of the tail, (ii) The software used these markers to fit a three-dimensional surface model of the animal, which helped compensate for variation in viewpoint, posture changes and body shape of the animal across photographs, and (iii) The software extracted a standard region of the shoulder stripes and created a planar black-and-white image, which was then used for pattern-recognition. For female images, we switched to a simpler identification program (Wild-ID; http://dartmouth.edu/~envs/faculty/bolger.html) that required only one preprocessing step: cropping a rectangular region of the torso of each animal (Fig. 3). In both programs, the main region of interest for pattern matching was along the torso between the mid-neck and rump.

Figure 3.

 Example of an adult female wildebeest photograph captured on two occasions in different wet season ranges: (a) in the Simanjiro Plains in June 2006 and (b) in the Northern Plains in June 2007. Female was a breeder in both years. Dashed lines show the cropped region of the torso used for pattern analysis and image matching in adult females.

Both software programs matched and scored images using four similar steps: (i) distinctive features within each processed image were located using the SIFT operator (Scale Invariant Feature Transform; Lowe 2004). These ‘SIFT features’ were invariant to scale and rotation, (ii) The program identified candidate pairs of SIFT features from each pair of images in the data base, (iii) A subset of geometrically self-consistent matched image pairs obtained in step 2 was selected, from which the program calculated a 2D affine transform, mapping the first image to the second image, (iv). The program assigned a standardized score between 0 and 1, describing the strength of match between the two images, and (v) Images were ranked based on the standardized score. For each photograph, the user (T. Morrison) visually compared the top twenty ranking photographs and recorded any matches. We then compiled the resulting set of matched photographs into encounter histories that denoted whether individuals were seen or not seen (1 or 0) during each sampling period.

Photographic data often violate the capture–recapture assumption that all marks (i.e. photographs) are correctly identified. False acceptances (i.e. falsely matching two photographs of different animals) are relatively rare in the wildebeest encounter history data sets (estimated false acceptance rate was 8·1 × 10−4, based on 100 test images; Morrison et al. 2011), and we assume these errors did not have a significant impact on data structure. However, encounter histories likely contained moderate numbers of false rejections (i.e. failures to match two photographs of the same individual), which inflate the number of observed encounter histories (Yoshizaki et al. 2009; Morrison et al. 2011). We estimated the false rejection rate (FRR) for both male and female identification programs using a test set of 198 images of known-identity animals collected in both the dry and the wet seasons (Morrison et al. 2011). The two software programs yielded similar false rejection rates (‘FRR’; FRR: 0·06–0·08; Fig. S1). This was unsurprising, given that both programs used the same pattern-characterization algorithm (SIFT) and scored images in a similar manner. Any slight differences between male and female data sets because of the software would be reflected in recapture probabilities and not in transition probabilities (i.e. the probability of migrating to alternative ranges the following year) of the capture–recapture models because transition probabilities are already conditioned on individuals being available for capture at least twice. Thus, we combined both male and female data sets and using them in a single ‘all-adults’ model.

Range fidelity models

We fit two sets of wet season encounter history data to multistate robust design capture–recapture (MSRD) models (Brownie et al. 1993): the ‘all-adults’ model and the ‘females-only’ model. MSRD models provide estimates of transition probabilities among and between different states across some sampling interval. The interval between primary sampling periods was 1 year and ‘state’ in our models corresponded to the three wet season ranges that individuals occupied at the time of sampling: Simanjiro (SP), NP or Lake Manyara National Park (LM). However, because we observed very few switches to, or from, LM, we excluded LM data from the model. We report the observed transitions involving LM animals and discuss this smaller population separately.

The ‘all-adults’ model examined the effect of sex and range in year i on the probability of transition to an alternative wet season range (SP or NP) in year i + 1. We define inline image and inline image as the probabilities that an animal present in SP and NP, respectively, in the wet season of year i and alive in year + 1, selects the same wet season range in + 1. We developed various parameterizations of these transition probabilities. First, ‘non-Markovian transitions’ occurred when range in year i was random with respect to the range selected the previous year, following constraints: ψSP−SP = ψSP−NP, and ψNP−NP = ψNP−SP (Nichols et al. 1994). These constraints implied that individuals first captured in SP or NP have the same probability of returning to those same ranges as they do of migrating to the alternative range (i.e. that movement between ranges is random). Models lacking this constraint indicate that individuals exhibit either fidelity (e.g. ψSP−SP > ψSP−NP) or dispersal (e.g. ψSP−SP < ψNP−SP). ‘Directionality’ occurs when ψSP−NP > ψNP−SP or ψSP−NP < ψNP−SP.

The second model (‘female-only’) used female captures across two wet seasons (2006 and 2007) to examine whether a female’s breeding status [breeder (B) or non-breeder (N)] in year i influenced the probability of switching ranges in year + 1. Breeding status was recorded at the time of photo capture (i.e. May–July, 2·5–4·5 months post-calving), and females occupied four possible states (SPB, SPN, NPB, NPN). Breeding status was based on whether or not females had visible mammary glands (i.e. teats), indicating that they were nursing a calf. Any adult female that had lost their calf within c. 10 days of being photographed would likely still have visible teats and would thus be recorded as a breeder (Watson 1967). Similarly, ‘non-breeders’ included both females that had failed to breed and females that had reproduced during that current breeding cycle but had lost their calf or foetus c. 10 or more days before being photographed. In 24·5% of captures, we were unable to discern breeding status because females moved away too quickly, so these females have unclassified breeding statuses. All females that had unclassified states (always because of unknown breeding status) were censored from the data set (n = 41 individuals). We assumed that all recorded breeding statuses were classified correctly and that unclassified females were random with respect to transition probability and survival.

Other assumptions were similar for both the ‘all-adults’ and the ‘females-only’ models: (i) there was no heterogeneity in capture or survival probabilities within ranges and sexes, (ii) within primary periods, survival probability was 1·0 and individuals could not transition between states, and (iii) that the population was open to transitions between states, mortality and recruitment between primary periods (Brownie et al. 1993).

Model selection

In the ‘all-adult model’, we compared 20 candidate models where survival (ϕ), transition (ψ), and capture probabilities (P) varied by sex (g), state (s), primary periods (T) and secondary periods (t). In the ‘females-only’ model, 18 candidate models were developed in a similar fashion; though, in these we varied all model parameters by breeding status (b) rather than sex. We developed parameterizations of ϕ, ψ and P based on a priori model sets to reduce the number of potential models to a manageable figure (Tables S2 and S3). The ‘all-adults’ model used a global model of {ϕ(T,g,s), ψ(T,g,s), P(T,t,g,s)}, indicating variation in survival, transition probability, and capture probability across sex, breeding states, primary periods and secondary periods. The ‘females-only’ model used a global model of {ϕ(s,b), ψ(s,b), P(s,b,t)}. We assessed Goodness-of-fit tests on the global models using the program MSSURVIV (Hines 1994). This program estimates a pooled G2 Goodness-of-fit test statistic, which can be used to assess the amount of dispersion in the data (c-hat) by dividing G2 by model degrees of freedom (Lebreton et al. 1992). We compared competing models using the quasi-Akaike Information Criteria corrected for small sample sizes (‘QAICc’; Burnham & Anderson 2002). QAICc weights determined the strength of support for a particular model within a model set (Burnham & Anderson 2002). All model selection steps and estimation procedures were conducted using the ‘Open Robust Design MultiState’ model with Huggins Closed Capture data structure in Program MARK, ver 5.1 (White & Burnham 1999).

Social-group associations in the dry season

We identified all pairs of individuals captured in the same herd in TNP during the dry season that were both also photographed in either the preceding or subsequent wet season. We classified each of these pairs into one of three categories: (i) SP pair (i.e. both individuals used Simanjiro in the wet season), and (ii) NP pair (i.e. both individuals used the NP in the wet season) or 3) mixed pair (i.e. one individual from SP and one from the NP). In all cases, herds contained other unidentified individuals whose wet season range affiliations were unknown. If social-group associations during the dry season were random with respect to their wet season range, we expected the number of herds in each of the three pair categories to approximate a binomial distribution. We tested this hypothesis in each of three transition periods using a chi-squared test. We generated the expected frequency of herd category in each transition using the relative frequency of individuals from either wet season range. All estimates are reported as mean ± SE.


Overall, we collected 5657 high-quality images of 2557 unique wildebeest on wet season ranges between 2005 and 2007 (Table S1). We observed 150 recaptures (involving 136 unique individuals) among the wet season ranges in consecutive years (Fig. 4). Wildebeest exhibited high but variable fidelity to migratory wet season ranges. The most parsimonious model in the ‘all-adult’ data set {ϕ(T,g,.)ψ(T,.,.) P(T,t,g,s)} estimated annual range fidelity as 1.0 from 2005 to 2006 (no SE because of estimates lying at the edge of parameter space) and 0·82 ± 0·06 from 2006 to 2007 (Table 1). We found a strong effect of year on the probability of switching wet season ranges between year i and year i + 1, with switching more likely between 2006 and 2007, but no effect of sex nor of directionality in year i [summed QAICc weights for models with an effect of year = 0·94, for models with an effect of sex = 0·28, and for models with directionality (i.e. a state effect) = 0·08; Table S2]. The effect of year may have been partially an artefact of lower power to detect transitions in 2005 (unique captures: N2005 = 384, N2006 = 1178 and N2007 = 1230; Table S1). The ‘all-adult’ model did not suffer from a significant lack of fit (χ2 = 11·60, d.f.=10, = 0·37).

Figure 4.

 Observed fidelity patterns to wet season ranges in Tarangire–Manyara wildebeest, summed across two wet–wet season transitions (2005–2006, and 2006–2007). Black squares indicate instances where individuals returned to the same seasonal range, while white squares are instances of switching seasonal ranges. Ranges include: Simanjiro Plains, Northern Plains and Lake Manyara National Park .

Table 1.   Estimates of transition probabilities between Simanjiro Plains (SP) and the Northern Plains (NP), Tanzania between 2005–2006 and 2006–2007. These estimates are derived from a time-varying model of transition probability (Model 1, Table S2). Note that we could not estimate SE for transition probabilities that fell near the boundary of parameter space (0.0 and 1.0)
2006SwitchψSP−NP, ψNP−SP0·000·00
StayψSP−SP, ψNP−NP1·000·00
2007SwitchψSP−NP, ψNP−SP0·180·06
StayψSP−SP, ψNP−NP0·820·06

In the ‘females-only’ data set, the probability of returning to, or of switching, wet season ranges from years i to i + 1 depended on breeding status in year i (Table 2). The top eight models ranked by QAICc included an effect of breeding status on the transition probability (Table S3; summed QAICc weights for models with an effect of breeding status = 0·99). Breeders in year i were over three times more likely than non-breeders to switch wet season ranges between year i and i + 1. Total switching probability among breeders in year i was 0·20 (i.e. 0·10 + 0·10), while non-breeders was only 0.06 (i.e. 0·03 + 0·03; Table 2). Overall, we observed 9 of 32 breeders in year i switching wet season ranges in year i + 1, while 0 of 10 non-breeders switched between years (Fig. 5). While eight of nine observed range switches involved breeders moving from SP to NP (all between 2006 and 2007), the top model indicated that range ‘switching’ probabilities were equal in both directions (i.e. ψSP−NP = ψNP−SP; Table 2). The ‘female-only model’ did not suffer from lack of fit (χ2 = 38·54, d.f. = 28, P = 0·09) and had a c-hat value of 1·38.

Table 2.   Transition probabilities for the ‘Female-only’ model for breeders (B) and non-breeders (N) in the Simanjiro Plains (SP) and Northern Plains (NP), Tanzania from June 2006 to June 2007
Breeding StatusTransition typeParametersEstimateSELCIUCI
  1. Estimates were derived from Model I (Table S3), which assumed transitions across wet season ranges from year i to i + 1 were dependent upon breeding status of females at year i and that ‘switching’ was equal for breeders and non-breeders at year + 1. Note that transition probabilities of non-breeders and breeders at year i each sum to 1·0. Estimates with asterisks were calculated by subtraction; no SE or CI could be estimated.

Non-breeder at 1st captureStay-do not breedinline image0·10*   
Stay-breedinline image0·840·110·510·97
Switch-do not breedinline image0·030·03<0·010·17
Switch-breedinline image0·030·03<0·010·17
Breeder at 1st captureStay-do not breedinline image0·050·040·010·24
Stay-breedinline image0·75*   
Switch-do not breedinline image0·100·040·050·21
Switch-breedinline image0·100·040·050·21
Figure 5.

 Observed fidelity patterns among adult females only, summed across three consecutive wet seasons (2005–2007). Breeding status was classified at year i. Black squares indicate instances where individuals returned to the same wet season range, while white squares are instances of switching wet season ranges. Note these data are a subset from Fig. 4A (i.e. females in which breeding status was recorded in year i).

Wildebeest around Lake Manyara National Park were largely isolated from the migratory portion of the population and exhibited near-absolute fidelity. We observed three transitions to, or away from, LM in consecutive wet seasons. Across all encounters, including non-consecutive wet and dry seasons, we observed 12 total transitions between LM and other ranges (9 females and 3 males). This involved movement to, or from, the NP (= 7 transitions), TNP (= 4 transitions) and SP (= 1 transition), demonstrating an underlying degree of connectivity between the migratory and resident populations.

Individuals coming from, or going to, the two wet season ranges appear to associate at random within dry season herds in TNP (Table 3). The distribution of pair-wise within-herd associations in the dry season did not differ significantly from a random null model of associations in all years, except in the early dry season of 2007. During this early 2007 wet-to-dry transition (‘3a’ in Table 3), pairs of animals in TNP were significantly segregated by the identity of their wet season ranges during the previous wet season. However, by the late dry season sample (i.e. immediately prior to migrating to wet season ranges), associations were random with respect to the identity of their previous wet season (‘3b’ in Table 3).

Table 3.   Social-group associations within dry season herds are random with respect to wet season range in the prior or forthcoming wet season
 Transition typeDate rangeNo. of associationsχ2d.f.P-value
  1. aTransition 3a included 25 associations that were observed within a month of animals migrating back to the dry season range. If these associations are excluded so that we only included mid or late dry season observations (3b), the relationship is not significant.

1Dry to wetOctober 2006–July 200790·4420·80
2Wet to dryJune 2006–November 2006280·5020·97
3aaWet to dryMay 2007–November 2007399·6820·01
3baWet to dryMay 2007–November 2007142·1220·35


Range fidelity in migratory ungulates

Wet season ranges provide tropical ungulates with seasonally available, high-quality forage that is critical for reproductive activities and play central roles in adaptive explanations of the causes or the timing of migration (Fryxell & Sinclair 1988; McNaughton 1990; Holdo, Holt & Fryxell 2009). Fidelity to these ranges constrains the ability of individuals to respond to resource heterogeneity across the entire landscape, which increases sensitivity to habitat degradation or loss (Owen-Smith 2004). High fidelity also promotes genetic differentiation among population segments, which furthers the importance of managing each segment independently.

Adult wildebeest in the TME exhibited high but variable fidelity to wet season ranges. While patterns of dispersal and fidelity have not been well-documented in other tropical migratory ungulates (Bolger et al. 2008), many temperate ungulates exhibit similarly high fidelity to seasonal ranges. For example, in Yellowstone National Park, 96% of migratory elk Cervus canadensis Erxleben (= 52 individuals followed for 2–4 breeding cycles) returned to the same summer grounds in consecutive years across 12 possible summer ranges (White et al. 2010). Additionally, pronghorn Antilocapra americana Gray (White et al. 2007), sika deer Cervus nippon Temminck (Sakuragi et al. 2004) and barren-ground caribou Rangifer tarandus groenlandicus Borowski (Cameron, Whitten & Smith 1986) exhibit high fidelity to summer ranges, while woodland caribou R. tarandus caribou Gmelin (Schaefer, Bergman & Luttich 2000) and bighorn sheep (Festa-Bianchet 1986) exhibit high winter site fidelity. Other ungulate populations exhibit much lower fidelity to all or portions of their ranges (Faille et al. 2010). In the Porcupine caribou herd that migrates hundreds of kilometres each year in northern Canada and Alaska, Fancy & Whitten (1991) found low fidelity to calving sites when comparing occupied calving sites to randomly selected calving sites in consecutive years (= 245 transitions across years). This variation in strategies across seasons, populations and species suggests an underlying degree of plasticity in range selection in ungulates that is shaped by local environmental conditions. An important aspect of future research, with high relevance to conservation of migratory ungulates, will be to determine if and how fidelity patterns change as local conditions change (Faille et al. 2010).

Determinants of wet season range selection

Surprisingly, breeding female wildebeest were responsible for over three times as many range switches as non-breeders within the two migratory ranges (Table 2). This result runs contrary to past research that has found either no correlation or a positive correlation between reproductive success and fidelity (Schaefer, Bergman & Luttich 2000; Hoover 2003; Tremblay et al. 2007). One possibility is that breeders were more nutritionally stressed than other animals, making them particularly sensitive to variation in food quality across the landscape and more inclined to explore alternative ranges. While we could not measure nutritional state or body condition, annual rainfall was below average in TME during the course of our study (2005–2007) and was particularly low during April in both 2005 and 2006 (Fig. S2), the period of year when lactating wildebeest are under their greatest energetic demands (Sinclair 1977). Given the rapid decline in wildebeest abundance in this ecosystem (TAWIRI 2001), we suspect food limitation and resource competition are relatively low on wet season ranges and that these factors could not adequately explain the pattern of range switching among breeding females, except perhaps at the very beginning of the rainy season when resource patches are highly variable across space. However, even if range switching by lactating females is a response to spatial heterogeneity in resources at the beginning of the rainy season, we would also have expected directionality in range shifts (i.e. a greater proportion of individuals moving one direction than the other) among these females. We found no statistical support for directionality in range selection; though, a greater number of animals shifted ranges from the SP to the NP. This lack of directionality makes it difficult to implicate resource-driven explanations for range switching above other possible explanations, such as human disturbance (Faille et al. 2010) or predation risk (Wittmer, McLellan & Hovey 2006).

Social-group associations during the dry season did not appear to relate to wet season range selection (Table 3). Individually based movement models in group-forming animals have assumed that individual movement is a consequence of social forces and directional biases of one or more leaders in a herd (Gueron, Levin & Rubenstein 1996; Couzin et al. 2005). While our results cannot rule out the influence of social forces in influencing wet season range selection, they do suggest that wildebeest herds are not stable over the dry season and that herd identity does not influence wet season range selection. Early in the 2007 dry season, individuals in TNP were still segregated according to their wet season affiliations. However, this pattern disappeared by the late dry season (October–November) and pair-wise associations within herds become random with respect to the identity of an individual’s previous or forthcoming wet season range.

Migratory vs. residency strategies

Partial migration occurs when only a portion of a population migrates in any given year (Kaitala, Kaitala & Lundberg 1993). Similar to fidelity, an individual’s migration strategy (residency vs. migratory) may be inflexible such that they either always migrate or always remain resident (Anderson 1991). Alternatively, decisions to migrate may vary from year to year and be conditioned on a combination of environmental, internal or social cues (White et al. 2007). In the LM subpopulation, only a small portion of individuals were observed in other wet season ranges (c. 6% of observed consecutive-year transitions). This confirms earlier speculations that the LM wildebeest form a relatively isolated resident subpopulation (Borner 1985; Prins & Douglas-Hamilton 1990; Georgiadis 1995). Prins & Douglas-Hamilton (1990) report that wildebeest went locally extinct in the LM basin in the mid-1960s as a result of rising lake levels, but that a small number of animals (n = 80) repatriated the area in the mid-1970s. Their residency near the northern boundary of Lake Manyara National Park is intriguing in the light of the fact that the NP migration route lies within <5 km of the LM population, near the vicinity of Manyara Ranch. While land between the migration corridor and LM is inhabited by an increasingly sedentarized pastoralist community, much of the area remains open rangeland through which other large herbivores are known to move (e.g. elephants). The availability of year-round water and perennial grasses with long growing seasons near the lake edge (Prins 1988) likely allow animals to move locally, rather than regionally, in search of resources (Owen-Smith 2004). Nonetheless, dispersal events to and away from LM demonstrate that this subpopulation is, at least nominally, still connected to the large TME population. The fact that the majority of their range is inside Lake Manyara National Park may decrease their long-term vulnerability to the changes happening elsewhere in the ecosystem. Nonetheless, the dichotomy in movement strategies between LM and Tarangire animals highlights the need for comparative analyses that quantify the costs and benefits of migratory vs. residency strategies in animal populations (Bolger et al. 2008; Hebblewhite & Merrill 2009), as well as longitudinal studies that track migratory strategies over the course of individuals’ lifetimes.

Fidelity and conservation

The TME ecosystem is undergoing rapid conversion of rangelands for agricultural use (TCP 1998). Most historical migratory pathways first described by Lamprey (1964) are no longer available for use, and the population has declined roughly eightfold between 1988 and 2001 (TAWIRI 2001). The majority of wet season habitat of the Eastern white-bearded wildebeest subspecies (C. taurinus albojubotus) in southern Kenya and northern Tanzania lies outside of formally protected areas (Estes & East 2009). If range fidelity remains consistently high over many years, these wildebeest will have a limited capacity to overcome rapid habitat changes, severe fluctuations in environmental conditions or barriers to migration such as high traffic roads. Wildlife managers, therefore, should not assume that wildebeest can easily switch to new wet season ranges if previously inhabited ranges deteriorate. Overall, high fidelity to wet season ranges constrains migration patterns at the seasonal scale, promotes genetic differentiation among population subunits (assuming that breeding occurs on these ranges) and furthers the importance of managing each subunit independently (White et al. 2010). Understanding variation in fidelity and dispersal in migratory ungulates is central to developing effective conservation strategies in the face of habitat changes to seasonal ranges.


We thank the Tanzania Wildlife Research Institute, the Commission for Science and Technology and the Tanzania National Parks for permission to conduct research in Tanzania. We are grateful to R. Mollel, J. McGrew and N. Brown for help collecting and analysing photographs, to J. Nichols and J. Hines for input about the capture–recapture modelling, and to G. Hopcraft and an anonymous reviewer for constructive comments on the manuscript. The work was funded by the Wildlife Conservation Society, Dartmouth College, the Marion and Jasper Whiting Foundation, the Nelson A. Rockefeller Center and NSF grant DBI-0754773.