Drivers of wildlife population dynamics are generally numerous and interacting. Some of these drivers may impact demographic processes that are difficult to estimate, such as immigration into the focal population. Populations may furthermore be small and subject to demographic stochasticity. All of these factors contribute to blur the causal relationship between past management action and current population trends. The urban Peregrine Falcon Falco peregrinus population in Cape Town, South Africa, increased from three pairs in 1997 to 18 pairs in 2010. Nestboxes were installed over this period to manage the interface between new urban pairs of Falcons and the human users of colonized buildings, and incidentally to improve breeding success. We used integrated population models (IPMs) formally to combine information from a capture–mark–recapture study, monitoring of reproductive success and counts of population size. As all local demographic processes were directly observed, the IPM approach also allowed us to estimate immigration by difference. The provision of nestboxes, as a possible stimulant of population growth, improved breeding success and accounted for an estimated 3–26% of the population increase. The most important driver of growth, however, was immigration. Despite low sample sizes, the IPM approach allowed us to obtain relatively precise estimates of the population-level impact of nestbox deployment. The goal of conservation interventions is often to increase population size, so the effectiveness of such interventions should ideally be assessed at the population level. IPMs are powerful tools in this context for combining demographic information that may be limited due to small population size or practical constraints on monitoring. Our study quantitatively documented both the immigration process that led to growth of a small population and the effect of a management action that helped the process.
Urbanization has mostly negative effects on biodiversity (McKinney 2002, Mcdonald et al. 2008, Chamberlain et al. 2009), although some species have adapted well and are profitting from the expansion of urban habitats (DeStefano & DeGraaf 2003, Rutz 2008). Human habitation offers a variety of intentional or incidental subsidies to wildlife, such as provision of supplementary foods (Fuller et al. 2008), provision of protective denning or nesting structures, and the removal of predators (Evans et al. 2009, Smith et al. 2010). However, human activities also pose risks to urban wildlife, such as the release of non-native organisms, chemical pollution and collision with vehicles, overhead lines or other fixed structures (Chace & Walsh 2006, Bradley & Altizer 2007). Whether a species can successfully exist in urban areas depends on the net effect these factors have on a population. Because different drivers can affect different demographic rates (Gaillard et al. 2000, Altwegg et al. 2005), the net effect of various human impacts on wildlife is usually not easy to anticipate.
Similar issues arise when evaluating the effectiveness of conservation interventions. Many interventions target particular demographic rates and even if the intervention improves the target rate, it may not have a sufficient effect at the population level. For example, head-starting sea turtles improves early-life survival but has only a small effect on turtle numbers overall (Crouse et al. 1987). Similarly, control of invasive bullfrogs initially focused on removing adults, but ultimately this approach proved to be a less effective control strategy than removing metamorphic frogs (Govindarajulu et al. 2005). Providing nestboxes is a common tool in avian conservation and generally improves breeding success (Newton 1994, Arlettaz et al. 2010, Libois et al. 2012) but it can also lead to lower fledgling survival, for example if nestboxes lead birds to breed in environments with higher risks to fledglings (Klein et al. 2007). Examining population-level consequences of such interventions is therefore important (Catry et al. 2009).
A change in population size is determined by additions through recruitment and immigration, and losses through death and emigration. Avian population dynamics are conveniently represented using matrix population models as a tool to examine population-level consequences of changes in demographic rates (Caswell 2001). Matrix models are parameterized using age-specific survival and reproduction rates, which may be estimated from capture–mark–recapture experiments and nest monitoring. However, species of conservation concern often have small populations, or it may be too difficult to obtain adequate sample sizes without harmful disturbance of sensitive taxa. Where population census data are also available, integrated population models offer an efficient method for combining these different data types to yield more accurate and comprehensive estimates of demography (Schaub et al. 2007, Schaub & Abadi 2011).
Here, we examine the demographic effect of providing nestboxes in a growing urban population of Peregrine Falcons Falco peregrinus, using an integrated population model. In many parts of the Northern Hemisphere this cosmopolitan species is recovering from a condition of near-extinction, attributed largely to the recent historical effects of chemical pollution (Cade et al. 1988). The Peregrine's ability to use urban environments appears to be an important contributor to its renewed success in these areas (Kauffman et al. 2004). We studied the Peregrine population of the Greater Cape Town area, South Africa (which has no known history of anthropogenic depression), during a sustained period of continuous growth. This trend was particularly marked in the urban sub-population, which increased from three pairs in 1997 to 18 pairs in 2010. In addition to the effect of nestboxes, we wanted to know to what degree this rapid increase of urban Peregrines was driven either by immigration or by local recruitment. A high immigration rate would indicate strong connectivity of this population with surrounding non-urban populations.
Materials and Methods
Study species and site
We studied Peregrines in the Greater Cape Town area from 1989 to 2010, and instituted a colour-ringing scheme from 1997 onwards. The birds were ringed either as nestlings or after capturing them as newly established breeders. All individuals received a unique combination of coloured aluminium rings and a numbered SAFRING (South African Bird Ringing Unit, University of Cape Town) ring. Individuals were later re-identified by reading their colour-ring combinations through a 20–60× spotting scope. Peregrines in this population breed between September and December (Jenkins 2000) and all known territories were visited at least once during this period to establish whether the resident breeding pair was still present or one of the old birds had been replaced by a new breeder. Such a replacement usually only happened when the old bird had died, in which case we identified the new, marked resident, or attempted to capture and mark it if it was not already ringed. All pairs considered in this study of the strictly urban segment of the overall population nested on buildings, and we were able accurately to record the number of fledglings each pair produced. Before and during the breeding season, we thoroughly searched the study area for newly established territories, and in some cases were notified by the general public about the establishment of new urban pairs. We are confident that we found most such new pairs within the first year after they became established.
New urban Peregrine pairs took up residence and attempted to breed on various types of buildings, sometimes coming into conflict with the people living in or managing these structures. Nestboxes were generally used as a reactive management tool to mitigate such conflict situations, and were installed in territories where the birds had particularly poor prospects of breeding successfully. The nestboxes were set up in relatively quiet areas of the selected building so the birds could breed undisturbed and away from sites where they might foul exposed structures or otherwise interfere with human activities. The nestboxes used were open-fronted, measured about 0.7 × 0.6 × 0.6 m, and were constructed from marine plywood, lined with a 5-cm-deep layer of river sand and fine gravel, with multiple drainage holes in the floor.
We made use of three separate datasets, collected between 1997 (when the use of colour rings first permitted collection of resighting data) and 2010. First, the capture–mark–resighting data consisted of 612 encounters (captures or resightings) of 149 individuals. Of these individuals, 136 were ringed as nestlings. Some of these birds were resighted as non-breeders (the vast majority as immatures). After taking up residence in a breeding territory, birds were generally not observed to leave these and we consequently treated all of them as breeders with a correspondingly higher resighting probability. Secondly, the number of fledglings each pair produced was known for every year of the study. The third dataset consisted of a census of all breeding pairs per year, which varied between 3 and 18 pairs.
Parameter estimation using integrated population modelling
We used the recently developed integrated population modelling approach to estimate demographic parameters and the population trajectory. The model allowed us to combine demographic data from different sources in a single model and to generate more precise estimates of a range of key parameters, some of which would not otherwise be estimable from any of the individual data sources alone (e.g. Besbeas et al. 2002, Abadi et al. 2010a, Schaub et al. 2012). See Schaub and Abadi (2011) for a review of integrated population modelling.
We followed the steps described in Schaub and Abadi (2011) to set up the integrated population model for Peregrines. First, we described the life cycle of Peregrines using a population matrix model that assumed a post-breeding census (Caswell 2001) with five age classes as:
where Nj,t is the number of individuals of age class j (age class 0: juveniles (0–1 year old); 1–3: sub-adults; 4: breeders) in year t and φj,t is the local survival probability of an individual of age class j between year t and t+1. Permanent emigration is confounded with mortality in such data. Note that the survival probability is the same for all individuals 2 years or older. ft is the number of fledglings per breeding pair in year t, which is independent of age. αj,t is the probability that an individual starts breeding at age j+1. ωt is the immigration rate in year t. Immigrants thus enter as adults in the population. Our census data only include breeding adults, and we were only able to observe new individuals when they were established as breeders in the study area. So this assumption closely matches our data.
Secondly, we constructed the likelihoods for capture–mark–resighting, census and productivity data. The likelihood for the capture–mark–resighting data, denoted by , was constructed using a multistate mark–recapture model in a state–space formulation (Schwarz et al. 1993, Gimenez et al. 2007), with six true states (alive as a juvenile, alive as a 1-year-old, alive as a 2-year-old non-breeder, alive as a 3-year-old non-breeder, alive as a breeder, or dead), and four observed states (seen as a juvenile, seen as a non-breeder, seen as a breeder, not seen). Here p1 and p2 are the recapture probabilities for non-breeders and breeders, respectively. For the census data, the likelihood denoted by was built using a state-space model (e.g. Schaub et al. 2007, Abadi et al. 2010b), which consisted of two processes. The state process described the dynamics of the true (unknown) population sizes over time as presented in Equation (1) above and, in addition, we used appropriate probability distributions (i.e. Poisson, binomial) to account for demographic stochasticty. The observation process linked the observed population counts to the true population sizes, assuming a Poisson process for the number of breeding pairs counted in the population. Even though the model does not attempt to correct for the possibility of missing breeding pairs, it does allow for variability introduced by the observation process. The likelihood for productivity data denoted by LPR(f) was constructed using a Poisson model (e.g. Schaub et al. 2007, Abadi et al. 2010b). Finally, we built the combined likelihood, assuming independence among the different datasets, as the product of the likelihoods for the individual datasets. That is,
In our case, the three datasets were based on the same individuals but simulations suggest that violation of the independence assumption does not bias greatly parameter estimates or their precision in models with live re-encounter data (Abadi et al. 2010a).
The census data provide information about all parameters and are key for the integrated population modelling (Supporting Information Appendix S1). Furthermore, the three datasets share information about common parameters that improve the precision of parameter estimates (Schaub & Abadi 2011).
We expected that nestboxes would increase the number of fledglings a pair could produce compared with pairs using open nests on buildings. We also expected potential carry-over effects of nestboxes on juvenile survival. These two parameters were therefore allowed to differ between birds born in a nestbox vs. those born in an open nest. We furthermore allowed for variation in juvenile survival and number of fledglings among the years of the study. We also let the immigration rate vary over the years. As immigration is probably driven by processes outside the study area, it did not make biological sense to us to treat the immigration rate as constant, which would have forced the number of immigrants to be directly proportional to local population size. With the limited data, we kept the other parameters constant over time. We implemented the model within a Bayesian framework specifying non-informative priors for each model parameter. We specified an N(0,100) prior for each regression coefficient in the model, a U[0,1] prior for the survival (immature and adult), breeding and recapture probabilities, and a U[0,10] prior for the standard deviation parameters. We also specified an N(5,100) prior truncated to positive values for the initial age-specific population sizes (see Supporting Information Appendix S2 for details). The analysis was conducted in jags (Plummer 2003) via the r package R2jags (Su & Yajima 2012). We ran three parallel chains of 50 000 iterations with a burn-in of 30 000 and kept every 10th observation to assess convergence of the Markov chain Monte Carlo (MCMC) to the targeted posterior distributions. All the r-hat values were below 1.02, suggesting convergence. We then ran a further 20 000 iterations to compute the posterior summary statistics. Details of the prior distributions used and the jags code are available in Appendix S2.
Assessing the impact of nestboxes and immigration
We used the estimates obtained from the integrated population model to parameterize the projection matrix model described in Equation (1), and to evaluate the impact of nestboxes and immigration on the population trend. We took the observed number of fledglings and adults in 1997 and, not having directly observed the other age classes, assumed that the population was otherwise in demographic equilibrium. We projected the adult population forward until 2010 using the estimated demographic rates and actual proportion of territories with nestboxes. Then, we explored what might have happened to the population in a range of different scenarios. The first scenario explored what might have happened if there had been no nestboxes by repeating the projection with the assumption that all individuals reproduced and survived at the same rate as individuals without nestboxes. This scenario may underestimate the effect of nestboxes on the population because nestboxes were generally provided to pairs that had selected particularly poor nest-sites. As a second scenario, we therefore assumed that these pairs would not have been able to breed successfully at all, had they not had nestboxes. The true effect of nestboxes in this population lies somewhere between these two scenarios. The third scenario assumed that nestboxes were present but that there was no immigration to this population. We quantified the uncertainty in the effects of nestboxes on the final population size by randomly drawing demographic estimates from the posterior distributions and re-running the population projection 10 000 times. We used the 2.5th and 97.5th percentiles of the estimated differences in final population size between scenarios as a 95% confidence interval.
Estimates of demographic parameters
Nestboxes increased the number of fledglings produced per pair but had no statistically discernible effect on juvenile survival (Fig. 1), which varied around a posterior median of 0.297 (95% credible interval (CRI): 0.162, 0.483). Estimated median survival of birds in their second year was 0.716 (95%CRI: 0.486, 0.910) and the survival probability of older birds was 0.852 (95%CRI: 0.788, 0.906). The estimated median probability of starting to breed was 0.519 (95%CRI: 0.332, 0.709) for 2-year-old birds and 0.774 (95%CRI: 0.554, 0.927) for 3-year-old birds. The immigration rate varied little across years around a posterior median of 0.208 (95%CRI: 0.089, 0.424). This led to an estimated zero to two individuals entering the population each year (Fig. 2). The estimated median resighting probabilities for non-breeders and breeders were 0.480 (95%CRI: 0.326, 0.643) and 0.938 (95%CRI: 0.876, 0.977), respectively.
Impact of nestboxes and immigration on the population trend
The population model reproduced the observed population trajectory well (Fig. 3). Under the scenario without nestboxes, our model suggested a slightly lower trajectory with a final population size that was 0.5 (95%CI of the difference: −1.2 to 2.7) pairs lower than the model with the nestboxes if we assumed all pairs had breeding success as observed in territories without boxes, and 3.9 (1.3–11.5) pairs lower if we assumed that these pairs would not have been able to breed successfully at all. When we set immigration to zero, the projected population growth was slightly negative.
Attributing population change to particular conservation interventions is a difficult task because such interventions may affect a number of demographic components with different relative impacts on overall population growth (Green 1995). The most critical conservation interventions often involve small populations, with associated low sample sizes and imperfect detectability posing additional problems for robust statistical inference. In these situations, integrated population models are particularly useful tools for understanding the dynamics of a population (Schaub et al. 2007). We used an integrated population model (Besbeas et al. 2002, Schaub & Abadi 2011) to estimate the effects of providing nestboxes on a small but growing urban population of Peregrines in Cape Town and found that nestboxes indeed probably contributed to the growth of this population. However, immigration was even more important, showing that the connection to the non-urban population critically drove the observed increase in urban birds.
The challenge with estimating population-level effects of conservation interventions is that any of the four demographic components – reproduction, survival, immigration and emigration – could be affected, potentially in different ways. Using integrated population models and local demographic data, we were able to separate three of them, leaving only permanent emigration and mortality confounded. However, this is not a problem for understanding local dynamics because permanent emigration and mortality both lead to permanent losses for the local population.
Our analysis showed that immigration was an important driver of the observed population increase. The demographic data suggest that the population would have tended to decline in the absence of immigration. This result means that local recruitment was not enough to balance losses but that the losses are likely to some extent attributable to be permanent emigration rather than mortality. Our estimates of juvenile and second-year local survival were considerably lower than those observed in an urban Peregrine population in California (0.65 and 0.86, as compared with our estimates of 0.30 and 0.71), which had an adult survival almost identical to that of the Cape Town population (0.86 compared with our 0.85; Kauffman et al. 2003). This discrepancy may well reflect more permanent emigration of immature birds from our study area. The inferred population decline in the absence of immigration does not necessarily mean that this local population is a sink at the landscape level, because it probably contributed immigrants to other populations (Runge et al. 2006).
The provision of nestboxes is a common conservation intervention for birds where suitably protective nesting sites are thought to be limiting (Newton 1994). We found that nestboxes improved breeding success compared with open nest-sites on buildings. However, nestboxes were not allocated at random in our study, but were installed in territories where breeding attempts led to conflicts with the human occupants of the building on which the birds had decided to breed. These attempts were generally unsuccessful until a nestbox was installed, so the comparison of nesting success between pairs with nestboxes and pairs without is likely to underestimate the positive effect of nestboxes. Without nestboxes, the nest success of these pairs would have been somewhere between the success measured for un-manipulated pairs, and complete failure. Our model suggested that nestboxes contributed between 0.5 (former scenario) and 3.9 (latter scenario) pairs to the population that increased by 15 pairs in total during the study period, even though the uncertainties around these estimates were relatively high because of the small number of individuals upon which the estimates were based.
Nestboxes are generally found to improve reproductive success (Newton 1994, Kostrzewa & Kostrzewa 1997, Arlettaz et al. 2010, Libois et al. 2012) and have been linked to population growth in raptors (Catry et al. 2009, Brown & Collopy 2013). However, nestboxes may also have negative effects on target populations. Barn Owls Tyto alba fledging from nestboxes may have a lower juvenile survival than those fledging from natural nests (Klein et al. 2007), and the availability of artificial nest-sites has been suggested as a cause for the collapse of the tree-breeding Peregrine population in Germany (Kirmse 2001). We found no evidence to suggest that nestboxes had a negative effect on juvenile survival, or any other demographic parameter, in Cape Town's urban Peregrine population.
Peregrine populations are recovering in many areas around the world after decades of intensive conservation attention (Cade et al. 1988, Sielicki & Mizera 2008). Part of the reason for the escalating success of this species is that it has so successfully colonized urban environments (Kauffman et al. 2003). In built-up areas, Peregrines use high buildings as surrogates for cliffs, nesting and roosting on these protective central sites, and exploiting abundances of preferred prey species (particularly Rock Dove Columba livia and Common Starling Sturnus vulgaris), themselves thriving on a similar suite of human subsidies (Chamberlain et al. 2009). They may also benefit from the moderating effects of cities on the local climate (Haggard 1990), and the exposure to predation of night-flying birds by high levels of light pollution (Drewitt & Dixon 2008). On the other hand, urban Peregrines may be subject to a higher risk of mortality in collisions (with buildings, cars and overhead lines), by premature fledging and by direct persecution (Cade & Bird 1990), and may also be adversely affected by chemical pollution (e.g. Newsome et al. 2010). Overall, the combined effect of these various factors in many modern cities seems to be positive for Peregrines (and for a number of other raptor species, e.g. Rutz 2008), and urban areas are population sources at the landscape level for Peregrines in California and the midwestern US (Kauffman et al. 2004, Wakamiya & Roy 2009). Urban Peregrine populations have completely subsumed a previously rural population in western Germany (Wegner et al. 2008).
We found that the increase in Cape Town's urban Peregrines was in large part due to immigration and the provision of nestboxes. The substantial influence of immigration shows that this urban population is still tightly linked with surrounding non-urban birds (Brown & Collopy 2013) and that the welfare of the species in these more remote environments is still critical to sustaining population stability and growth in the city. While the driver of immigration in this instance is not yet known (although burgeoning numbers of prey species are implicated), it seems likely that the Peregrine is still in the process of colonizing urban Cape Town (Rutz 2008) and that the marked effect of immigration on the dynamics of this population may diminish as other factors impose constraints on continued growth.
Thanks to Rob Robinson for inviting us to present this work at the 2013 BOU conference From populations to policy impact: avian demography in a changing world. He and two anonymous reviewers provided helpful comments. F.A. was supported by fellowships from the Claude Leon Foundation and the University of Cape Town. A.R.J.'s fieldwork was partly funded by Steve Phelps and Peregrine Properties, BirdLife South Africa and Vodacom South Africa. R.A. was supported by the National Research Foundation of South Africa (Grant 85802). The NRF accepts no liability for opinions, findings and conclusions or recommendations expressed in this publication.