Declining survival rates in a reintroduced population of the Mauritius kestrel: evidence for non-linear density dependence and environmental stochasticity

Authors

  • M. A. C. Nicoll,

    1. School of Animal & Microbial Sciences, Reading University, Whiteknights, Reading RG6 6AJ, UK,
    2. Mauritian Wildlife Foundation, Black River, Mauritius, and
    Search for more papers by this author
  • Carl. G. Jones,

    1. Mauritian Wildlife Foundation, Black River, Mauritius, and
    2. Durrell Wildlife Conservation Trust, Les Augres Manor, Trinity, Jersey, Channel Islands, UK
    Search for more papers by this author
  • Ken Norris

    Corresponding author
    1. School of Animal & Microbial Sciences, Reading University, Whiteknights, Reading RG6 6AJ, UK,
      Ken Norris, School of Animal & Microbial Sciences, Reading University, Whiteknights, PO Box 228, Reading RG6 6AJ, UK. E-mail: k.norris@reading.ac.uk
    Search for more papers by this author

Ken Norris, School of Animal & Microbial Sciences, Reading University, Whiteknights, PO Box 228, Reading RG6 6AJ, UK. E-mail: k.norris@reading.ac.uk

Summary

  • 1We studied a reintroduced population of the formerly critically endangered Mauritius kestrel Falco punctatus Temmink from its inception in 1987 until 2002, by which time the population had attained carrying capacity for the study area. Post-1994 the population received minimal management other than the provision of nestboxes.
  • 2We analysed data collected on survival (1987–2002) using program MARK to explore the influence of density-dependent and independent processes on survival over the course of the population's development.
  • 3We found evidence for non-linear, threshold density dependence in juvenile survival rates. Juvenile survival was also strongly influenced by climate, with the temporal distribution of rainfall during the cyclone season being the most influential climatic variable. Adult survival remained constant throughout.
  • 4Our most parsimonious capture–mark–recapture statistical model, which was constrained by density and climate, explained 75·4% of the temporal variation exhibited in juvenile survival rates over the course of the population's development.
  • 5This study is an example of how data collected as part of a threatened species recovery programme can be used to explore the role and functional form of natural population regulatory processes. With the improvements in conservation management techniques and the resulting success stories, formerly threatened species offer unique opportunities to further our understanding of the fundamental principles of population ecology.

Introduction

Understanding how and to what extent density-dependent and independent processes influence demographic parameters and hence regulate population size has been fundamental within population ecology (Andrewartha & Birch 1954; Lack 1954; Newton 1998; Coulson, Milner-Gulland & Clutton-Brock 2000; Saether & Engen 2002). With demographic models being used more frequently to assess the long-term viability of populations for both commercial management (Pascual, Kareiva & Hilborn 1997; Frederiksen, Lebreton & Bregnballe 2001) and conservation purposes (Green, Pienkowski & Love 1996; Hiraldo et al. 1996) understanding the processes driving population regulation remain essential. One of the key demographic parameters examined has been survival. Density-dependent survival has been identified in long-term studies of stable and fluctuating populations for a wide range of species (Tinbergen, Vanbalen & Vaneck 1985; Francis et al. 1992; Lieske et al. 2000; Larter & Nagy 2001). Contrastingly, density-independent survival has also been found in many long-term studies (Arcese et al. 1992; Nur & Sydeman 1999; Loison et al. 2002b; Tavecchia et al. 2002). Some studies have been able to determine the relative influence of the two processes on survival (Milner, Elston & Albon 1999; Frederiksen & Bregnballe 2000; Loison et al. 2002a).

Identification of the appropriate density-dependent functional form exhibited by a key demographic parameter, such as survival, is particularly relevant when populations are managed for commercial or conservation purposes around model-based population projections. The selection of an unsuitable population regulatory mechanism as part of a model's structure could result in misleading population projections and inappropriate management recommendations (Pascual et al. 1997; Runge & Johnson 2002). Recent published research has shown the existence of non-linear density-dependence in natural populations, that are characterized either by extreme fluctuations in density (Clutton-Brock et al. 1997; Leirs et al. 1997; Lima, Merritt & Bozinovic 2002) or has developed from a comparatively low density to the carrying capacity for the environment (Taper & Gogan 2002). Evidence for a density-dependent Allee effect, whereby a component of an individual's fitness benefits from the presence of conspecifics (Allee et al. 1949; Stephens, Sutherland & Freckleton 1999), remains limited (Courchamp, Clutton-Brock & Grenfell 1999; Stephens & Sutherland 1999).

Despite many recent incidental species introductions and intentional reintroductions, few documented studies have observed populations growing from extremely low densities to the carrying capacity for the environment. With the advances in conservation management techniques over the last 40 years, particularly in avian conservation, species have been saved from imminent extinction (Bell & Merton 2002). For some species the establishment of new viable populations through translocation or reintroduction has been crucial, e.g. the Chatham Island black robin Petroica traversi, Seychelles magpie-robin Copsychus sechellarum and the Mauritius kestrel (Bell & Merton 2002). In each case new populations have developed from zero and their development closely monitored. Following the successful reintroduction of a species the developing population might be expected to develop rapidly before levelling off, roughly following a logistic pattern (Newton 1998). The development of the Mauritius kestrel population reintroduced into the Bambous mountain range in 1987 (Jones et al. 1995) conforms to this. This reintroduced kestrel population therefore offers a potentially unique opportunity to study, in the wild, the role of density-dependent and stochastic processes throughout a population's development.

In this paper we use the program MARK (White & Burnham 1999) to examine how a key demographic parameter, survival, is susceptible to both density-dependent and independent processes over the course of the population's development.

Methods

study species

The Mauritius kestrel is a small accipiter-like falcon and the only extant raptor endemic to the Indian Ocean island of Mauritius. As a result of forest destruction (98%: 1753–1936) this formerly widespread raptor was restricted to the island's three main mountain chains by the late 1930s (Jones & Owadally 1985). The extensive use of organochlorine pesticides between 1948 and 1970 resulted in a catastrophic population decline (Safford & Jones 1997) and by 1960 the kestrel was found only in the Black River gorges (Fig. 1) and considered endangered (Vincent 1966; Brown & Amadon 1968). By 1974 only four individuals were known in the wild (Temple 1974). In 1973 a recovery programme was initiated for the Mauritius kestrel and ran until 1994 when the postbreeding population of the Mauritius kestrel was estimated at 222–286 birds (Jones et al. 1995). In 2000 the postbreeding population was estimated at 500–800 individuals (Jones, Groombridge & Nicoll 2002) and the Mauritius kestrel has now been downgraded from endangered to vulnerable (Stattersfield & Capper 2000). There are currently three distinct subpopulations confined to the island's three main mountain ranges (Fig. 1), outside of which kestrels are not known to persist. A detailed review of the conservation management of the kestrel including organizations involved and the techniques developed and applied is provided in Jones et al. (1991), Cade & Jones (1993) and Jones et al. (1995).

Figure 1.

The historical distribution of the Mauritius kestrel on Mauritius.

Mauritius kestrels are primarily monogamous and their breeding season spans the Southern Hemisphere summer, with the earliest eggs (clutch size two to five) being laid in early September and the latest fledglings (brood size one to four) leaving the nest in late February. Breeding seasons therefore include two calendar years and are referred to using those two years, e.g. 1991/1992. Mauritius kestrels fledge at around 35 days old, achieve independence at around 85 days old and are capable of breeding at 1 year old.

study area

This study has been conducted in the Bambous mountain range on the east coast of Mauritius (Fig. 1). It is composed primarily of a single spine, some 20 km in length, running east to west with spurs running off to the south. It rises from sea level to 680 m at its highest point. The annual rainfall (taken as September–August for the purpose of this study) ranges from 1619 to 4439 mm with up to 77% falling in the cyclone season (December–April). The core (∼50%) of the range is privately owned and managed for sport hunting. This has resulted in a habitat mosaic consisting of: grassland, invaded secondary forest and isolated pockets of remnant native forest. Areas outside of these hunting lands consist of heavily invaded secondary forest with native forest on the more remote ridges and mountain peaks. The mountain range is surrounded by agricultural land.

study population

In 1987, as part of the species’ recovery programme, the Mauritius kestrel was reintroduced into the Bambous mountain range, from which it had been extirpated for over 30 years. Some 46 captive-produced kestrels were reintroduced, using a soft release technique known as hacking, in three areas between 1987/88 and 1989/90. The developing population was augmented through the fostering of 38 chicks at established breeding pairs between 1988/89 and 1993/94. Thirty-five kestrels were also hacked out on Ile aux Aigrettes nature reserve (an island 1 km off the east coast) between 1989/90 and 1996/97. The majority of these kestrels consisted of rehabilitated individuals and those bred in captivity in the US and returned to Mauritius. For a detailed breakdown of kestrels entering the population, cf. Table 1. By 2001/02 the breeding population stood at 42 pairs.

Table 1.  Origins and numbers of Mauritius kestrels entering the population each breeding season: 1987/88–2001/02
Breeding seasonMonitored pairs (Nt)Hacked: Bambous MountainsHacked: Ile aux AigrettesChicks fosteredAll/ringed wild-bred fledglings
  • *

    Fostering post-1993/94 was incidental and consisted of movement of chicks between broods in the wild.

1987/88 012 0 0  0
1988/89 216 0 5  0
1989/90 618 914  0
1990/91 8 0 0 7  0
1991/9212 014 7 12/11
1992/9317 0 0 2 19/16
1993/9420 0 3 3 20/18
1994/9525 0 1 1* 46/43
1995/9633 0 0 2* 31/30
1996/9735 0 8 1* 42/39
1997/9835 0 0 0 50/44
1998/9934 0 0 0 41/38
1999/0040 0 0 0 49/45
2000/0141 0 0 0 39/35
2001/0242 0 0 0 54/52
Totaln/a463542403/371

data collection

Since the reintroduction of kestrels into the Bambous mountain range the population has been intensively monitored in conjunction with ongoing conservation management and continues today. Monitoring has focused on the breeding population on a seasonal basis each September–March. At the start of the breeding season all nestboxes and known nest sites were checked, pairs located, identified, their breeding status determined and breeding success monitored. Surveys were conducted to locate new pairs of kestrels. Where the nest cavities and broods were accessible chicks were ringed (aged 12–28 days). An individual colour code consisting of two plastic colour rings was put on one tarsus, allowing identification in the field, while a single numbered aluminium ring on the other tarsus provided permanent identification. Each released kestrel was similarly marked but for most closed, coloured metal rings were used. Where colour rings were lost or faded then individuals were trapped, identified and colour rings replaced. Over the course of the study period identities of kestrels could not be confirmed in only 14 instances. Re-sighting data was collected on 407 Mauritius kestrels between 1987/88 and 2001/02 with only four documented instances of adult mortality occurring within the breeding seasons. The discrete nature of this population as a result of the poor dispersal of the kestrel (Jones et al. 1991, 1995), the isolated nature of the mountain range and the distance to the nearest population (>20 km) combined with intensive monitoring and the reliance of Mauritius kestrels on nestboxes has allowed virtually all breeding pairs to be located and monitored with an estimate of only six undocumented nesting attempts by 1998 (Groombridge et al. 2001). In the last three seasons of the study the monitoring has been intensified, allowing the peripheral areas into which the kestrel has gradually expanded to be thoroughly surveyed. We believe that this coupled with the factors discussed above allows us to accurately determine on an annual basis the population size in terms of breeding pairs.

Daily rainfall (mm) was collected throughout the study period near to the site of the first release.

survival analysis

Re-sighting histories for 28 hacked, 28 fostered and 319 wild-bred Mauritius kestrels were constructed from survival data collected between 1987/88 and 2001/02. Kestrels hacked on Ile aux Aigrettes or fledged in 2001/02 were excluded from the analysis. Analysis conducted by M. A. C. Nicoll, C. G. Jones & K. Norris (unpublished data) showed that origin (hacked, fostered or wild-bred) had a limited influence on a Mauritius kestrel's subsequent survival but that it was cohort specific. Reduced first year survival was evident only in the 1989/90 cohort (cf. Fig. 2), for which we could find no plausible explanation. We therefore excluded this cohort from the analysis, on the grounds that this anomaly in the data series could potentially lead to identification of an inappropriate functional form of density-dependent Φ. The 375 re-sighting histories were pooled. We employed an a priori, stepwise approach using a set of candidate models to examine the influence of a particular variable, or set of variables, on survival. This approach allowed us (i) to refine the number of models in each candidate set based on biological reasoning and (ii) to logically work towards the development of models incorporating two or more variables that might explain any temporal variation in survival. The analysis was conducted using the program MARK (White & Burnham 1999).

Figure 2.

Juvenile Mauritius kestrel survival probabilities: 1987/88–2000/01. Survival estimates are generated from the most parsimonious time and age model but the re-sighting histories include the 1989/90 cohort +. Vertical bars are 95% confidence intervals. Adult survival probability remained constant at 0·783 (95% LCI 0·744, UCI 0·817).

time and age

A standard Cormack–Jolly–Seber (CJS) model fully time-dependent for survival (Φ) and re-sighting probability (P) was taken as the starting model. Age structure was introduced into Φ by distinguishing between juveniles and adults. A two-age class structure, juveniles and adults, was selected since for many raptors, including Mauritius kestrels, juvenile survival is known to be less than adult survival (Brown & Amadon 1968; Newton 1979, 1986; Village 1991; Watson 1997). Mauritius kestrels are known to breed as 1 year olds (Jones et al. 1995) and were therefore considered to be adults at this age. First year birds were treated as juveniles. We then examined the influence of time-dependent Φ on both age classes, constant Φ on both age classes and time-dependent Φ in juveniles only. This later model was based on the grounds that juvenile Φ < adult Φ in raptors and more sensitive to annual variation in environmental conditions (Village 1991). P remained time dependent. The most parsimonious model from the candidate set of age-structured Φ models was then selected and age structure introduced into P. We then examined the influence of time-dependent P on both age classes, constant P on both age classes and time-dependent P in juveniles only.

density

The most parsimonious age structured model was constrained by the following two measures of population density:

  • 1Breeding pair density at time t(Nt).
  • 2Breeding pair density at time t + 1(Nt+1).

These were selected as they could potentially influence either Φ through competition for resources (primarily food), or P through competition for breeding territories. Their relative influence on Φ was examined using three documented functional forms of density dependence: linear, non-linear (threshold) and Allee effect, cf. Fig. 3. In order to explain any time dependence in Φ we constrained Φ rates to be a function of density, i.e. density was included in the model at the expense of time. Initially the age-structured model was constrained in a linear fashion associated with increasing population density. To test for non-linear density dependence we assumed that Φ was constant before declining in response to density exceeding a critical threshold. A threshold level (taken as the average density for that period) was applied through the time series starting in 1987/88 with an initial endpoint of the 1991/92 breeding season. This season was selected as from this point on wild-bred fledglings were entering the population and rapid expansion began. In successive models the threshold was then extended forwards in time on an annual basis. The Allee effect was tested for by constraining the Φ rates to be 2nd order polynomial functions of density.

Figure 3.

Functional forms of density-dependent survival.

climate

Daily rainfall (mm) was used as the climatic variable. Mauritius kestrels fledge between 32 and 38 days and usually achieve independence between 70 and 100 days of age (Jones et al. 1991). For the purpose of this study fledging was taken as 35 days and independence as 85 days (50 days postfledging). For each cohort the number of rain days were calculated for the dependency period. In Mauritius the cyclone season runs from the start of December to the end of April (defined from data from the Meteorological Services in Mauritius, 1945–2001). For each cyclone season the number of rain days was calculated.

The number of rain days was converted into a fraction of the specific time period for use in program MARK. For each period the average proportion of rain days was calculated for the study (1987/88–2000/01). The climatic data for each breeding season were then scored against these averages as (>) ‘wet’ or (<) ‘dry’. This produced four measures of rainfall. Figure 4(a & b) illustrates the temporal distribution of the rainfall (raindays) during the dependency period and cyclone season. The influence of each of these rainfall measures on Φ was examined by constraining the Φ rates to be a function of climate within the most parsimonious (suite of) age-structured model(s).

Figure 4.

Measures of rainfall (collected daily 1 September 1987 to 31 August 2001) used to examine the influence of climate on Mauritius kestrel survival: (a) proportion of rain days during the period post fledging that juvenile Mauritius kestrels remain dependent on their parents; (b) proportion of rain days during the cyclone season.

climate and density

To examine the additive effect of density and climate on Φ we identified the most parsimonious (suite of) model(s) constrained by density and added the most parsimonious climatic variable.

model selection

Corrected Akaike's information criterion (AICc) in MARK (Cooch & White 2001) was used to select the most parsimonious model from a set of candidate models. This procedure was preferred over the use of a likelihood ratio test (LRT) to test between two nested models as some of the analysis was conducted using non-nested models. For one model to be selected above another (Anderson & Burnham 1999) recommend that the difference between Akaike's information criterion (AIC) (ΔAICc in MARK) should >2. If ΔAICc is between 1 and 2 then these models should be considered nearly tied and any subsequent inference should be based on the subset of models. The logit link function, in program MARK, was used throughout the modelling procedure.

An assessment of the fit of selected models to the data was made using the standard goodness-of-fit (GOF) bootstrap test available in program MARK. Some 100 replicates were run.

Results

survival: time and age

Of the eight candidate models (Table 2) selected to examine the influence of time and age on Φ, AICc in program MARK identified model 1 as the most parsimonious. It was 14·2 times better supported by the data than the next most parsimonious model. Model 1 showed that (i) Φ exhibited by juvenile Mauritius kestrels differed from adults, (ii) juvenile Φ was time-dependent while adult survival was constant and (iii) P were different for the two age classes but constant over time.

Table 2.  Models tested show the influence of time and age on both survival (Φ) and reporting rate (P). Models are denoted according to the model-specific variation exhibited; t denotes full-time specific variation; c denotes constancy among years; two age classes are used (juvenile/adult), e.g. Model 1 reads survival is time dependent for juveniles/constant for adults with constant re-sighting probability for both juveniles and adults. Models are ordered according to Akaike's information criterion (AICc). Model parsimony increases with decreasing AICc. The log link function was used for all models
ModelAICcΔAICcAICc weightsNo. parametersDeviance
1. {Φ(t/c) · P(c/c)}1487·946 0·0000·93416358·522
2. {Φ(t/c) · P(t/c)}1493·252 5·3060·06627340·570
3. {Φ(t/c) · P(t/t)}1503·63615·6900·00039324·816
4. {Φ(t/c) · P(t)}1511·03023·0810·00028356·198
5. {Φ(c/c) · P(t)}1513·55325·6070·00016384·129
6. {Φ(t/t) · P(t/t)}1517·65729·7110·00050314·144
7. {Φ(t/t) · P(t)}1524·41636·4700·00039345·596
8. {Φ(t) · P(t)}1577·69389·7470·00027425·011

Juvenile Φ estimates generated by model 1, Table 2, are illustrated in Fig. 5. The estimates appear to show a comparatively high (60–73%) stable period (1987/88–1993/94). This is followed by a sharp rise in Φ and then a progressive decline (1994/95–1997/98) after which the Φ rate stabilizes.

Figure 5.

Juvenile Mauritius kestrel survival probabilities: 1987/88–2000/01. Survival estimates are generated from Model 1, Table 2. Vertical bars are 95% confidence intervals. Adult survival probability remained constant at 0·782 (95% LCI 0·742, UCI 0·817).

survival: density

The fit of this model to the data was not improved by constraining the most parsimonious time and age model linearly (model 1, Table 2) by the two selected measures of density in turn (models 12 and 13, Table 3) or by constraining the most parsimonious time and age model with an Allee effect (models 15 and 16, Table 3). However, constraining the model in a non-linear fashion with density (Nt) operating at a single threshold level identified model 9, Table 3, as the most parsimonious. Model 9 was 21·17 times better supported by the data than the next most parsimonious model. Figure 6 illustrates the functional form of this threshold relationship between juvenile Φ and density.

Table 3.  Models tested show the influence of measures of population density on both survival (Φ) and reporting rate (P) within Model 1 (Table 2). Models are denoted according to the model-specific variation exhibited; t denotes full-time specific variation; c denotes constancy among years; Nt and Nt+1 denote the two measures of kestrel pair density; TH denotes the inclusion of a threshold and is followed by the breeding season after which density increases. Allee denotes the inclusion of the Allee effect in a model: Models are ordered according to Akaike's information criterion (AICc). Model parsimony increases with decreasing AICc. The logit link function was used for all models
ModelAICcΔAICcAICc weightsNo. parametersDeviance
9. {Φ(TH′94(Nt)) · P(c/c)}1477·311  0·0000·889 5 370·505
10. {Φ(TH′93(Nt)) · P(c/c)}1483·416  6·1050·042 5 376·610
11. {Φ(TH′92(Nt)) · P(c/c)}1484·103  6·7920·030 5 377·296
12. {Φ(Nt) · P(c/c)}1484·198  6·8870·028 5 377·392
13. {Φ(Nt+1) · P(c/c)}1487·206  9·8950·006 5 380·400
14. {Φ(t/c) · P(c/c)}1487·946 10·6350·00416 358·522
15. {Φ(Allee(Nt)) · P(c/c)}2219·046741·7350·000 41114·265
16. {Φ(Allee(Nt+1)) · P(c/c)}2259·040781·7290·000 31156·279
Figure 6.

Juvenile Mauritius kestrel survival probabilities: 1987/88–2000/01. Survival estimates are generated from the threshold density model (Model 9, Table 3). Vertical bars represent 95% confidence intervals. Adult survival remained constant at 0·782 (LCI 0·742, UCI 0·817).

survival: climate

Constraining the most parsimonious time and age model (model 1, Table 2) by the four selected measures of rainfall in turn (Table 4) significantly improved the fit of the model to the data for only one of the four constraints. The significant climatic variable was ‘wet’ or ‘dry’ cyclone seasons (cf. model 17, Table 4).

Table 4.  Models tested show the influence of climate on survival (Φ) within Model 1 (Table 2). Models are denoted according to the model-specific variation exhibited; t denotes full-time specific variation; c denotes constancy between years; DP denotes the dependency period; CY denotes the cyclone season: % denotes raindays as a fraction of the time period while * denotes the period scored as ‘wet’ or ‘dry’. Models are ordered according to Akaike's information criterion (AICc). Model parsimony increases with decreasing AICc. The logit link function was used for all models
ModelAICcΔAICcAICc weightsNo. parametersDeviance
17. {Φ(CY*) · P(c/c)}1484·8190·0000·650 5378·013
18. {Φ(CY%) · P(c/c)}1487·5732·7540·164 5380·767
19. {Φ(t/c) · P(c/c)}1487·9463·1270·13616358·522
20. {Φ(DP%) · P(c/c)}1490·3825·5630·040 5383·576
21. {Φ(DP*) · P(c/c)}1493·3448·5250·009 5386·538

survival: climate and density

The additive effect of density and climate proved significant with the most parsimonious model (model 22, Table 5) being constrained by both climate and density. The GOF bootstrap test, in program MARK, for model 22, Table 5, was not significant (P = 0·68 [32–33/100 in our example]), indicating that the model was an adequate description of the re-sighting histories.

Table 5.  Models tested show the influence of density and climate on survival (Φ). Models are denoted according to the model-specific variation exhibited; t denotes full-time specific variation; c denotes constancy between years; Nt denotes kestrel pair density; TH denotes the inclusion of a threshold and is followed by the breeding season after which density increases; CY* denotes the cyclone season scored as ‘wet’ or ‘dry’. Models are ordered according to Akaike's information criterion (AICc). Model parsimony increases with decreasing AICc. The logit link function was used for all models
ModelAICcΔAICcAICc weightsNo. parametersDeviance
22. {Φ(TH′94(Nt) + CY*) · P(c/c)}1474·566 0·0000·793 6365·729
23. {Φ(TH′94(Nt)) · P(c/c)}1477·311 2·7450·201 5370·505
24. {Φ(CY*) · P(c/c)}1484·81910·2530·005 5378·013
25. {Φ(t/c) · P(c/c)}1487·94613·3800·00116358·522

The Φ estimates (Fig. 7) clearly illustrate the influence of density and climate on juvenile Φ. Juvenile Φ is reduced as population density exceeds a critical value but fluctuates among cohorts at a level dictated by the climatic conditions experienced by each cohort.

Figure 7.

Juvenile Mauritius kestrel survival probabilities: 1987/88–2000/01. Survival estimates generated by the additive model of density and climate (Model 22, Table 5) are shown, separated according to climatic conditions (i.e. wet or dry). Vertical bars represent 95% confidence intervals. Adult survival remained constant at 0·782 (LCI 0·742, UCI 0·817).

The juvenile Φ estimates generated from the most parsimonious time and age model were significantly correlated (r = 0·868, P < 0·00) with those generated from the density and climate model (with climate operating independently of density). Thus 75·4% of variation in Φ exhibited among cohorts can be explained by the influence of increasing population density and the temporal distribution of rainfall during cyclone seasons.

Discussion

analysis

Model assumptions

Monitoring of the Mauritius kestrels in the Bambous Mountains focused on the breeding population. Sightings and identification of non-breeding individuals were infrequent. As Mauritius kestrels are generally nest site faithful this monitoring technique conveys a bias in recapture rate at (t + 1) to kestrels breeding at time (t). The fact that the breeding population has remained the monitored element of this subpopulation leads us to assume that any bias would then be consistent throughout the study. Like many other studies of non-colonial bird populations monitoring of the non-breeding portion of the population is notoriously difficult (Newton 1979, 1998) and assessments of population dynamics are based around the breeding birds. As a result of this the annual estimates of survival are likely to be underestimated.

Model robustness

For the purpose of this analysis, whereby program MARK was used to explore biological hypotheses, not to generate the most meaningful survival estimates, we used an age-structured model distinguishing only between juveniles and adults. This was consistent with previous analyses (M.A.C. Nicoll, C.G. Jones & K. Norris, unpublished data). We did run additional analyses to explore the sensitivity of age structure (not shown in this paper). This resulted in more parsimonious age-structured models containing additional age classes, but only juvenile survival demonstrated time dependence. Therefore for the purpose of this study we retained the simple two-age class structure.

Throughout the analysis our interest has focused on the survival estimates and has been conducted in a stepwise process using the parsimony approach. To ensure that our conclusions reached about survival are robust and have not been inadvertently influenced by excluding an undetected change in re-sighting probability (i.e. constant to time dependent) we repeated the analysis with time-dependent re-sighting probability for both juveniles and adults. The order of model parsimony obtained remained the same but with significantly larger AICc values as expected.

density dependence

The recognition of different survival probabilities among adult and juvenile Mauritius kestrels agrees with many results found in other long-term population studies (Gaillard et al. 1998; Frederiksen & Bregnballe 2000; Lieske et al. 2000; Larter & Nagy 2001). The survival probabilities exhibited by juvenile Mauritius kestrels over the course of the population's development (Fig. 5) indicate that there is (i) a gross decline in survival and (ii) marked variation in survival among cohorts. We examined this gross decline in survival and found that it was strongly associated with density, exhibiting a non-linear threshold functional form. Juvenile survival was negatively correlated with population density only above a certain critical threshold. In our study population we suspect this threshold was attained once the core area, containing the optimal habitats, was saturated. As density increased beyond this critical level population expansion occurred in peripheral areas, composed of potentially lower-quality habitats. The discrete nature of the study area and the nest site non-limited nature of the population support this. This threshold non-linear density-dependent process acting on survival appears to be characteristic of populations experiencing large fluctuations in density (Bjornstad et al. 1999; Lima et al. 2002). This could explain its comparative rarity in the literature as the majority of studies involve established populations that experience only small fluctuations in density (Gaillard et al. 1998). The importance of considering non-linear density-dependent processes operating on demographic parameters should, however, not be underestimated.

Demographic models are being used with increasing frequency in conservation biology and commercial management to predict how a population may behave in the future (Norris & Stillman 2002). The accuracy of the projections from these population viability analyses is largely dependent on the quality of the ecological data for the target species underpinning the model (Brook et al. 2000; Coulson et al. 2001). Therefore awareness of the functional forms exhibited by density dependence in key demographic parameters and use of the most relevant form within a demographic model are fundamental for model-based management projections. As survival is a key demographic parameter frequently instrumental in population regulation, inclusion of an inappropriate density-dependent functional form (e.g. linear over a non-linear threshold) within a demographic model could bias the predicted population response. Potentially this could result in misleading projections and the implementation of unsuitable management recommendations. This issue of model structure has been examined within the context of population management for commercial purposes by Pascual et al. (1997) and Runge & Johnson (2002). In both cases they found that the functional form of a significant key demographic parameter influenced the recommended management action and subsequent long-term stability of the system.

climate

In addition to the density-dependent process reducing juvenile survival the influence of stochastic events was also detected. Among consecutive cohorts variation in survival of nearly 20% was attributable to the temporal distribution of rainfall experienced by first year birds postfledging and then as newly independent juveniles during the cyclone season. Survival was reduced when the number of rain days experienced by the cohort exceeded the seasonal average for the study period. Thus wetter cyclone seasons reduced the survival probability relative to drier ones. This stochastic climatic effect probably influenced juvenile survival by reducing (i) the time available to hunt, (ii) prey abundance and (iii) hunting efficiency. Weather (particularly rainfall) has been observed to affect hunting effort and efficiency in Mauritius kestrels directly (M.A.C. Nicoll, C.G. Jones & K. Norris, personal observation) in a similar fashion to other raptors (Cave 1968) and indirectly through affecting the behaviour of its primary prey arboreal geckos of the Phelsuma genus that hide during rain (M.A.C. Nicoll, C.G. Jones & K. Norris, personal observation). The influence of weather on prey abundance and hunting efficiency in raptors has been documented in a range of species from the Eurasian kestrel Falco tinnunculus (Village 1991) to the golden eagle Aquila chrysaetos (Steenhof, Kochert & McDonald 1997). For newly independent Mauritius kestrels developing their hunting skills, increased periods of prolonged rainfall could therefore result in starvation. Newton (1986) and Village (1991) observed peak periods of mortality in newly independent juvenile sparrowhawks and common kestrels associated with a decrease in prey abundance.

climate and density

Combined density dependence and stochastic processes account for 75·4% of the temporal variation in juvenile survival exhibited during the population's development. For resident terrestrial vertebrate populations it would be generally expected that the effects of climate on demographic parameters would become more evident as a population approaches the carrying capacity for the environment (Portier et al. 1998; Frederiksen & Bregnballe 2000). In our study the most parsimonious model, constrained by density and climate, implies that rainfall experienced postfledging is operating as an episodic event impacting on survival of juvenile cohorts in a constant fashion independent of population density. However, if we divide the annual survival estimates into two temporal periods, threshold and post-threshold, we can see that the data within each period are limited, e.g. only two cohorts experience ‘dry’ cyclone seasons and four experience ‘wet’, post-threshold (cf. Fig. 7). We believe that this paucity of data precludes any further meaningful examination of a density-dependent/climate interaction with the current time series. However, additional data collected as part of the ongoing monitoring programme might enable us to assess this at a later date. From our analysis we can only conclude that climate impacts on juvenile survival through the process outlined earlier while the relationship between climate and density requires additional research. With rainfall being so influential on survival in such a formative period in the life of a Mauritius kestrel, any long-term change in annual rainfall distribution could have significant effects on kestrel population dynamics. The current global warming and associated climate change (Pain & Donald 2002) could potentially affect the population stability of a tropical species such as the Mauritius kestrel.

Acknowledgements

The Mauritius kestrel recovery programme has been sponsored by the National Parks and Conservation Service, Government of Mauritius, the Peregrine Fund, the Mauritian Wildlife Foundation and the Durrell Wildlife Conservation Trust. We are grateful to all those in the field, captive breeding centre, laboratory and office that have made this work possible. We would also like to thank Tim Coulson, Rhys Green and two anonymous referees for their constructive comments on earlier manuscript drafts.

Ancillary