EDITOR'S CHOICE: Saving the hihi under climate change: a case for assisted colonization

Authors


Summary

  1. Climate change is affecting the spatio-temporal distribution of environmental conditions, forcing species to shift their range in response. Species not capable of dispersing naturally may benefit from conservation translocations. A key aspect of translocation planning is release site selection: under the 2012 IUCN guidelines for reintroductions and other conservation translocations, selected sites are expected to match the biotic and abiotic needs of the candidate species now and in the future.
  2. Here we present a methodological framework to identify optimal translocation sites under climate change. Our method is the first to explicitly combine statistical and predictive population modelling to understand the relationship between climate, climate change and population dynamics, in order to perform robust habitat suitability analyses for conservation decision-making.
  3. We use the hihi Notiomystis cincta, a bird endemic to New Zealand, as a case study. We focus on the population of Tiritiri Matangi Island, which has been provided with ad libitum supplementary food since 1996. This offers the unique opportunity to study the direct impact of climate and future change in climatic conditions on a population free of confounding constraints.
  4. Climate is found to drive hihi population dynamics, even though they are not limited by the availability of food. Thus, despite the current management of the species, climate change remains a major threat to its long-term persistence. Moreover, under predicted climate change for the country, hihi suitable habitat will shift southward: the two current largest hihi populations will face unsuitable climatic conditions in the coming decades, and habitat that was not part of the species' historical range may become suitable.
  5. Synthesis and applications. Assisted colonization is increasingly being considered as an adaptation tool for species threatened by climate change. Justifying the use of this extreme conservation action, however, requires robust evidence that it is necessary and clear guidance on where to translocate individuals of threatened populations. We show how both requirements can be met using habitat suitability modelling if knowledge of the relationship between climate, climate change and the species' population dynamics is systematically used to guide the modelling process.

Introduction

Climate change is one of the most severe threats to biodiversity (Parmesan & Yohe 2003; IPCC 2007). Environmental conditions such as averages and extremes of temperature and precipitation, seasonality and primary production dynamics are changing at the global scale (Foden et al. 2008). This trend is expected to continue in the future (Parmesan 2006), and consequences on species have already been observed (Wilson et al. 2007; Hoegh-Guldberg & Bruno 2010). For some species, climate change might trigger the emergence of a significantly different set of environmental conditions over most of their range, creating a disparity between the new conditions experienced by the species and its current realized niche. To avoid extinction, these species will have to either adapt to new conditions or shift their ranges to follow suitable ones (Davis & Shaw 2001; Chauvenet et al. 2012a).

Not all species that will need to shift their ranges will be able to do so. First, they may not be mobile enough or too slow moving to track rapidly shifting suitable environmental conditions (Schloss, Nunez & Lawler 2012). Alternatively, they may be faced with a physical barrier to their dispersal such as having to cross an ocean, a mountain or a city. Those species can be helped using conservation translocations (Seddon 2010). These are defined as ‘the intentional movement and release of a living organism where the primary objective is a conservation benefit’ (IUCN 2012) and are commonly used for species conservation and ecosystem restoration purposes (Seddon 2010). There are two main types of translocations: reintroductions, when species are moved in parts of their historical range, and assisted colonizations (also known as benign introductions; IUCN 2012; Chauvenet et al. 2012a), when species are moved into an area where they were never recorded before. Assisted colonization is generally considered a riskier choice than reintroduction due to the documented evidence of extreme and irreparable negative consequences on host ecosystems from introduced species (IUCN 2012). Therefore, if managers are to choose this conservation path, their choice must be backed by robust evidence that it is best for the species, likely to be a successful endeavour, and that it will not have negative consequences on the host ecosystem.

Today, one major impediment to the use of any conservation translocation is their poor overall success rate. The often-cited reason for this lack of success is the ad hoc selection of translocation sites (Griffith et al. 1989; Osborne & Seddon 2012). The need for a framework that addresses site selection issues is thus paramount, especially as climate change is making identifying suitable long-term habitat for threatened species even more difficult. The recent IUCN guidelines for reintroductions and other conservation translocations (IUCN 2012) offer an explicit list of requirements when selecting translocation sites under climate change. Sites must meet the following criteria: answer all biotic and abiotic needs of the candidate species at all its life stages; be adequate for its seasonal needs; be large enough to meet the conservation targets; provide enough connectivity between suitable sites; and be far away enough from potential unsuitable sink habitats. Moreover, translocation sites must remain suitable in the future under predicted climate change (IUCN 2012). Therefore, the best practice for translocation site selection will involve a two-step process. First, one must understand what aspects, if any, of climate drive the candidate species' population dynamics and investigate how climate change may affect their long-term dynamics. Second, based on this information, one must identify the distribution of habitat in which species can persist on the long term under predicted change.

Here we propose a novel methodological framework to comprehensively and systematically identify current and future suitable translocation sites for species threatened by climate change. It involves the combined use of statistical and predictive population modelling, as well as species distribution modelling, to robustly identify the distribution of habitats suitable for potential translocation. This framework is the first to explicitly and comprehensively address the main pitfall of habitat suitability modelling: the usual lack of a priori understanding of the species' ecology or how climate influences its population dynamics (Guisan & Thuiller 2005; Araújo & Guisan 2006), which can lead to uninformative variable selection, and makes the evaluation of results difficult. Although habitat suitability models have been used before to identify the distribution of current and future suitable habitat for species (Carroll et al. 2009; Morueta-Holme, Fløjgaard & Svenning 2010), this approach remains rare (Keith et al. 2008), and there is no evidence in the published literature of the systematic combined use of the proposed tools to select suitable translocation sites and plan translocations under climate change.

We used the hihi (stitchbird) Notiomystis cincta as a case study. The hihi is a vulnerable New Zealand endemic bird that currently persists in four intensely managed translocated populations and one remnant island population. The species represents the perfect case study for our framework as it offers the rare opportunity of studying the direct impact of climate change on population dynamics, while other potentially influential factors, such as food limitation unrelated to climate, are accounted for through management. Moreover, there is a keen interest in its management and conservation with an ongoing recovery programme led by the Department of Conservation of New Zealand, which aims to increase hihi numbers and guarantee their long-term persistence. The recovery programme currently has a two-pronged approach with, on the one hand, intense management of current translocated populations and, on the other hand, establishment of new hihi populations through translocations. All the populations are located in areas from which hihi cannot disperse naturally, potentially rendering the species highly susceptible to future changes in climatic conditions.

Materials and methods

Hihi are small, sexually dimorphic, forest-dwelling passerine birds. Once found throughout the North Island, hihi declined to one single offshore population on Little Barrier Island (LBI; Hauturu) following European colonization and the consequent habitat loss and introduction of mammalian predators and diseases (Taylor, Castro & Griffiths 2005). Since the early 1980s, there have been several attempts at reintroducing hihi on and around the North Island, New Zealand. Today, hihi persist in five wild populations (four translocated and LBI; Chauvenet et al. 2012b; Fig. 1). Due to evidence that hihi were food limited (Armstrong & Ewen 2001; Armstrong, Castro & Griffiths 2007; Chauvenet et al. 2012b), the translocated populations are provided with ad libitum supplemental food (i.e. sugar water) through feeders at fixed locations. In order to quantify the link between climatic conditions, hihi vital rates and population dynamics, we focused our work on the population inhabiting Tiritiri Matangi Island (Tiri). All individuals from this population have been closely monitored since the establishment of the population in 1995 (Ewen, Armstrong & Lambert 1999; Ewen, Thorogood & Armstrong 2011). A detailed description of the monitoring procedure on Tiri can be found in Ewen, Thorogood & Armstrong (2011) and is thus not repeated here.

Figure 1.

Populations of hihi on the North Island of New Zealand; stars denote extant populations, and filled circles denote extinct populations.

Step (i) How does climate influence hihi vital rates?

The impact of climate on hihi survival was investigated using Cormack–Jolly–Seber (Cormack 1964; Jolly 1965; Seber 1965) model in MARK (White & Burnham 1999). Age-specific survival during the breeding (i.e. September–February; φ) and non-breeding (i.e. March–August; δ) seasons was modelled separately. To model hihi reproductive success, we defined female overall breeding success (γ) as the total number of fledglings raised by a female during an entire breeding season, Tf, divided by the total number of eggs she laid during that season, Te. We used generalized linear mixed models with a binomial error structure and a logit link function implemented in R (version 2·15·1; R Development Core Team 2012) to model γ. As measures were repeated across females and years, female identity and breeding year were set as random effects. The complete list of variables used to analyse hihi survival and female overall breeding success is provided in Table 1; a detailed description of the variables and the models run is given in Appendix S1 in Supporting Information. Variables used were both extrinsic (i.e. climate related) and intrinsic characteristics of individuals (e.g. age) and the population (i.e. density). For all these analyses, the best model of the candidate set was selected using the corrected Akaike Information Criterion for small data sets (AICc; Burnham & Anderson 2002).

Table 1. Covariates used in the survival and/or reproductive success analyses
NameDescriptionSurvivalReproductive success
BNB
  1. B, breeding season; NB, non-breeding season.

Intrinsic factors
AgeIndicating the age of individuals at each time step and was tested as both continuous and categorical (i.e. 3 age classes: juvenile, first year and older, and 4 age classes: juvenile, first year, 2–6 years and older)YesYesYes
Clutch numberNumber of clutches laid by females during the breeding seasonNoNoYes
DensityThe total number of individuals at the beginning of each breeding and non-breeding seasonYesNoYes
SexDistinguishing between males and femalesYesNoNo
Extrinsic factors
RainfallThe total precipitation across each breeding and non-breeding seasonsYesYesYes
TemperatureThe average monthly temperature across each breeding and non-breeding seasonsYesYesYes

Step (ii) How will climate change impact hihi population dynamics?

Model description

A stochastic matrix model was built in R to project the hihi population into the future and investigate the impact of different aspects of climate change on its dynamics (Fig. 2). In the model, hihi survival during the breeding (φ) and non-breeding seasons (δ) and female overall breeding success (γ) were defined according to the findings of step (i). The population was divided into nine classes: fledgling (from time of fledging until adulthood reached the next September) and 8-year-long adult age classes, for example, 1–2 years of age, 2–3 years of age, etc. No bird was allowed to age past 9 years as this is the oldest age ever recorded in the wild. We used ‘year’ as a time step and started the model at the beginning of the breeding season at time t0, setting the initial abundance to 100 first-year individuals. The abundance N at each time step (i.e. beginning of the breeding season at time + 1) was then modelled as a function of the number of individuals in each class at time t, the total number of fledglings produced during the breeding season (F) and age- and season-dependent survival φ(i) and δ(i) (i denoting the age class of individuals: fledgling stage = 1, between 1 and 2 years of age = 12 and between 2 and 9 years of age = 29) between time t and + 1 (eqns 1 and 2; Fig. 2):

display math(eqn 1)

where

display math(eqn 2)

and Nt(i) was the number of individuals in age class i at time t; Nt(j)/2 was the number of females in age class j at time t (assuming equal sex ratio; j denoting the age class of females: between 1 and 2 years of age = 12, between 2 and 6 years of age = 26, and between 6 and 9 years of age = 69); and E was the number of eggs laid per female. At each time step, E was sampled from a normal distribution of mean = 4·2 and SD = 1 (corresponding to the average number of eggs laid when females lay only one clutch during the breeding season) and rounded to the nearest whole number.

Figure 2.

Graphical representation of the stochastic matrix population model used to project the hihi population under different aspects of climate change. ‘Nf’ and ‘Nm’ represent the number of females and males in the population in age classes i and j.

Climate change simulations

To investigate the impact of climate change on hihi population dynamics, we ran a set of 15 simulations. Each one was defined by the values of the climate parameters that were found to significantly affect the survival and reproductive success of hihi in step (i). One of these 15 simulations represented the ‘baseline’ conditions, such that at each time step, values for the weather variables were randomly sampled on the normal distribution of the average observed weather conditions (and their standard deviation) since the population was reintroduced. The other 14 simulations were run to account for different aspects of climate change as suggested by global and regional projections from the IPCC (2007, 2012), for example, increased average temperature or decreased average rainfall. Table 2 provides a detailed list of the simulations considered.

Table 2. List of simulations run with and without temporal autocorrelation
SimulationMean Temp + 2 °CMean Temp + 4 °CTemp SD × 2Total rain − X%Rain SD × 2
  1. ‘Baseline’ corresponds to current climatic conditions.

  2. X = 7·5% during the non-breeding season and X = 10% during the breeding season.

  3. Mean Temp, average temperature; Total Rain, total rainfall; SD, standard deviation.

Baseline     
T1X    
T2X X  
T3 X   
T4 XX  
R1   X 
R2   XX
TR1X  X 
TR2X XX 
TR3 X X 
TR4 XXX 
TR5X  XX
TR6X XXX
TR7 X XX
TR8 XXXX

We first performed those 15 simulations without temporal autocorrelation in climatic conditions. Then we investigated the impact of three different patterns of autocorrelation on rainfall (droughts) and temperature (hot spells): more frequent and longer-lasting droughts, more frequent and longer-lasting hot spells, and positive covariation between droughts and hot spells. Temporal autocorrelation was modelled using discrete first-order Markov chain models, that is, the state of the system at time t was dependent on the state of the system at time − 1 (Truscott & Gilligan 2003). We defined different ‘climate states’, which described the environmental conditions that could be experienced by the population at any time, for example ‘drought’ or ‘non-drought’; we then set transition probabilities from one state to another. Because future temporal autocorrelation in temperature and rainfall in New Zealand has not been explicitly quantified, values for the transition probabilities were chosen to represent a significant departure from a system where there is no autocorrelation, while making sure that the population did not only experience droughts or hot spells (Appendix S2, Supporting Information).

Each simulation was repeated 1000 times. The total number of time steps of each simulation run was 1000, but we discarded the first 200 steps to measure the population dynamics at an equilibrium state. To quantify the impact of climate change on hihi population dynamics and compare the results between simulations, we collected the mean population size and its standard deviation and the probability of the population size dropping below two quasi-extinction thresholds set at 10 and 20 individuals.

Step (iii) How will climate change impact the spatial distribution of the hihi's suitable habitat?

Habitat suitability modelling was used to map the current, and predict the future, distribution of suitable habitat for the hihi. Potential predictor variables' selection was guided by the results from steps (i) and (ii), that is, climate variables found to influence hihi population dynamics were considered as potentially informative for their distribution. Similarly, this knowledge was used to validate the output of the models in terms of variables that came out as best predictors for the species distribution. The presence record was defined as every location in which hihi currently persists or was recorded to have bred and survived for 2 years or more since the recovery programme started (Fig. 1). We used maxent to model habitat suitability; it is a machine learning algorithm that uses presence-only data (maxent 3·3·3; Phillips, Anderson & Schapire 2006).

To use in maxent, we downloaded GIS layers for predictor variables describing ‘current conditions’ (i.e. 1950–2000), and ‘future conditions’ predictions for 2050 and 2100, from the worldclim data base (version 1·4; www.worldclim.org; Hijmans et al. 2005). Those layers were of climate variables found to be significant in steps (i) and (ii), as well as of the 19 Bioclim variables (version 1·4; www.worldclim.org; Hijmans et al. 2005; Table S1, Supporting Information). For future conditions, three emission scenarios (B2, A1B and A2) and two climate models were considered (GCMs: HADCM3 from the Hadley Centre for Climate Modelling and CSIRO from the Commonwealth Scientific and Industrial Research Organisation). Each scenario represents a different potential future for the Earth; A1B is often referred to as a ‘middle-of-the-road’ scenario where human population growth gradually declines after 2050 and the world relies both on fossil and non-fossil fuels equally (predicted temperature increase range 1·7–4·4 °C); B2 corresponds to a world with a slow and continuous human population growth, but more ecologically friendly (predicted increase 1·4–3·8 °C), and A2 refers to a world with a continuously fast increasing human population, regionally driven economic development and decline in the environment (predicted increase 2–5·4 °C; IPCC 2007). Two different global climate models (or GCMs) were used as it is known that different models may yield different results and those two models are amongst the most commonly used in the literature (Synes & Osborne 2011; Luedeling & Neufeldt 2012). Moreover, there was no regional climate model for New Zealand readily available.

We chose to work at a 2·5 arc-minute resolution, which is the second smallest grid size available from worldclim. Because suitable hihi habitat is not only dependent on climatic conditions, but also on the right habitat type and natural food supply, the extent of the suitable habitat predicted by MaxEnt was clipped to a layer of the current distribution of native forest from the New Zealand Land Cover Database (http://www.mfe.govt.nz/issues/land/land-cover-dbase/index.html).

The fit of the maxent model to the data was assessed using the area under the ROC (AUC; Zweig & Campbell 1993; Fielding & Bell 1997). An AUC of 1 indicated that the predictor variables perfectly explained the species distribution, while a value of 0·5 indicated that model predictions were not better than random; thus, the higher the AUC value, the better. Moreover, we used the ‘maximum of the sum of the sensitivity and specificity’ approach (Liu et al. 2005), which maximized the agreement between observed and predicted distributions, to identify the threshold that separates suitable from unsuitable habitat for the species.

Results

Step (i) How does climate influence hihi vital rates?

The total number of capture occasions in the sighting record was 31 and the total number of individuals = 1773. There was very little overdispersion (ĉ = 1·2), and the ĉ adjustment option in MARK was used to correct for it. Model selection was thus based on QAICc. Regardless of the season, models where recapture probabilities were time dependent always performed better than those where they were constant across seasons. Recapture probability on Tiri was high: 0·77 on average between 1996 and 2010 (SD = 0·15).

There were a number of models that plausibly fitted the data for hihi survival during the breeding season (φ; Table S2, Supporting Information). The best model contained age as a categorical variable separating first-year individuals (age class = 12) from older individuals (age class = 29), rainfall as an additive effect and the interaction between density and temperature (there are no fledglings in the population during the breeding season). Individuals that were older had an overall higher probability of survival than first-year birds. Moreover, when density was low, temperature had little effect on the survival of hihi. However, when density was high, above-average temperature had a significant negative impact of the survival of hihi of both age classes.

There was overwhelming support for one model of hihi survival during the non-breeding season (δ; Table S3, Supporting Information). It contained an interaction between age and temperature, and rainfall as an additive effect. Age was a categorical variable distinguishing between fledglings (age class = 1), first-year individuals (= 12) and older individuals (= 29). Juvenile survival was significantly lower than adult survival. Rainfall decreased the survival of all age categories to an extent, but that of juveniles the most.

Female hihi laid on average 1·6 clutches per breeding season (SD = 0·6), but there were instances when they laid 3 or 4 clutches. The average number of eggs laid across a breeding season per female was 6·5 (SD = 2·6), and the average number of fledglings produced was 3·0 (SD = 1·9). A few models plausibly fitted the data for female overall breeding success (γ; Table S4, Supporting Information). The best model showed γ as a function of age, density, temperature and clutch number as additive effects. Age was a categorical variable separating first-year females (age class j = 12), females between 2 and 6 years of age (age class j = 26) and females older than 6 years (age class j = 69). Thus, we were able to detect senescence in hihi female reproduction with older individuals (6–9 years of age) having a significantly lower reproductive success than females in their prime (2–6 years of age). Moreover, first-year females had a lower γ than older females. Overall, an increase in the average temperature and an increase in density both decreased γ.

Step (ii) How will climate change impact hihi population dynamics?

The ‘baseline’ simulation provided a benchmark against which to compare results of simulations that included climate change: under current conditions, the average hihi population size on Tiri was predicted to be 198 (SD = 68), and the probability of the population reaching either quasi-extinction threshold was 0 (Fig. 3).

Figure 3.

Impact of the different aspects of climate change with no temporal autocorrelation on: (a) the average hihi population size and (b) the probability of the population being smaller than 20 and 10 individuals.

Overall, we found that simulated changes in average temperature and rainfall according to climate change predictions were not enough to increase the population's probability of extinction; they had to be coupled with increased variance and temporal autocorrelation in droughts and hot spells (Figs S1–S3, Supporting Information). However, an increase in temperature by 2 or 4˚C alone severely reduced the population's carrying capacity, while an increase in average amount of rainfall did not. Finally, temperature autocorrelation increased the chance of the population declining below 10 or fewer individuals the most, yielding a 100% chance of extinction for some simulations (Fig. 3).

Step (iii) How will climate change impact the spatial distribution of hihi's suitable habitat?

The distribution of suitable habitat for the hihi was best predicted by the BioClim variables, and the top predictors of hihi habitat suitability were almost all temperature-based variables: variables with the largest contribution in predicting hihi suitable habitat were as follows: mean temperature of driest quarter (50·3%), mean diurnal range (maximum temperature − minimum temperature; 24·3%), mean temperature of coldest month (9·9%), precipitation seasonality (5·8%) and maximum temperature of warmest month (5·1%). The rest of the predictor variables contributed a marginal percentage or no percentage to the habitat suitability models. This result was in agreement with the findings of the predictive population model in step (ii), which showed temperature as the major driver of hihi dynamics, thus supporting our confidence in the habitat suitability modelling results. The average AUC for maxent was 0·891 (SD = 0·027).

Maxent found that 31·4% of the total native vegetation of North Island, or 17% of the total native vegetation of New Zealand, was classified as suitable for the hihi under current conditions (Fig. 4). When projecting the suitable range of hihi in the future under climate change, we combined the results of the two climate models (GCMs) as they yielded similar results for all investigated change scenarios (Table S5, Supporting Information). Hihi suitable range is predicted to move southward as a response to climate change (Fig. 5): the northern part of New Zealand, which was found highly suitable in current conditions, may become almost entirely unsuitable in the coming decades; the south-west and south-east of North Island as well as the northern part of South Island showed a larger extent of suitable hihi habitat in 2050 and in 2100 than under current conditions. As a result, of all the current hihi populations, one of the translocated ones (Tiri) and the remnant population on LBI may be faced with unsuitable conditions in the future (Fig. 5).

Figure 4.

Map of current suitable habitat for the hihi. Coloured pixels represent current native forest with black showing habitat that is suitable for the hihi.

Figure 5.

Map of future suitable habitat for the hihi in 2050 and 2100. Shown are the pixels selected as suitable by both HADCM3 and CSIRO, for three climate change scenarios: B2 (a, d), A1 (b, e), A2 (c, f), and for 2 timeframes: 2050 (a–c) and 2100 (d–f). Coloured pixels correspond to native forest with black pixels showing habitat that is suitable for the hihi.

Discussion

The methodological framework presented here was designed to optimize the selection of future translocation sites for species threatened by climate change. This framework, which relies on the combined use of statistical and quantitative modelling techniques to inform species distribution modelling, is the first one to explicitly address known issues of habitat suitability analyses that have been impeding their use for practical conservation decisions (i.e. robust choice of predictor variables and how to verify the reliability of results). For the hihi, it yielded an in-depth understanding of the relationship between climate and its population dynamics, which in turn enabled us to confidently make recommendations about (1) where the species may be translocated in the future and (2) the fact that assisted colonization of the species to the South Island of New Zealand should be considered as an option for the species' long-term persistence.

One important finding of this work is that, despite using a population that is not limited by food or breeding habitat availability as the focal population, we still found that climate change will have a negative impact on its population dynamics. The implication for the species as a whole is that, even with a substantial and continuous investment in management, climate change remains a significant threat to its long-term persistence. In addition, our models predict that the two northernmost populations, Tiri and LBI, will not remain within optimal climatic conditions under climate change. LBI is the last natural hihi population, is the largest one and has the highest genetic diversity of all extant populations (Brekke et al. 2011), while the population on Tiri is considered to be the most successful example of hihi reintroduction. The consequences of their extinction for the ongoing recovery programme would be devastating: both populations are regular sources of individuals for other reintroduced populations. Detecting, and correctly interpreting, changes in hihi numbers will thus be critical as climate changes environmental conditions. The current monitoring in place for hihi populations offers a chance to detect those rapidly; this knowledge can be coupled with population models within an adaptive management framework to make timely and robust decisions for current populations (Armstrong & Reynolds 2012; McCarthy, Armstrong & Runge 2012).

Since the beginning of the hihi recovery programme, all translocations have been reintroductions. Yet, a large proportion of the predicted future suitable habitat for hihi under climate change is in an area of New Zealand where the species was not found historically: the South Island. Thus, although a laudable strategy until now, choosing habitat based on where the hihi survived more than two centuries ago will not be good enough in the future. Instead, climate needs to be accounted for explicitly. Despite hihi populations not being limited by food availability due to the large investment into ad libitum supplemental feeding, evidence points to assisted colonization being the solution to guarantee the species' long-term survival. Adopting assisted colonization as the new best practice for hihi conservation will represent a significant shift away from the current management practice, but a necessary one.

Assisted colonization, however, remains a controversial conservation action; it is first and foremost an introduction and thus carries the associated risks to host environments (Ricciardi & Simberloff 2009; Schlaepfer et al. 2009). Quantifying those risks to novel environments will be a must before making the decision to introduce hihi onto South Island (Chauvenet et al. 2012a). The 2012 guidelines for reintroduction and other conservation translocations (IUCN 2012) identify five major risks to host environments from species introduction: ‘ecological risk’, which is the potential for negative impacts from the introduced species to other species and/or ecosystem functions; ‘disease risk’, which is the risk of new pathogens entering the ecosystem and spreading to native species; ‘associated invasive risk’, which is the risk of the introduced species spreading beyond the intended range, becoming invasive and out-competing native species; ‘gene escape’ risk, which is the risk of the introduced species and a closely related species hybridizing; and ‘socio-economic risk’, which is the risk faced by people's livelihoods and well-being. Experience from past hihi translocations has taught us that most of those risks are minimal. For example, on mainland sites where the species has been reintroduced, evidence shows that hihi are unlikely to successfully disperse away due to their vulnerability to introduced mammalian predators; there is no reason to suspect that the species can affect human well-being or livelihood, as no evidence of this has emerged during the running of the thirty years of recovery programme; the species is so phylogenetically distinct that there is no chance of gene escape (Ewen, Flux & Ericson 2006); hihi are the subordinate of the nectar-feeding guild of New Zealand (Rasch & Craig 1988), hence very unlikely to out-compete other species. Nevertheless, a thorough risk assessment when the time comes to envisage assisted colonization for the species will allow weighting the costs and benefits of hihi-assisted colonization, and uncertainties associated with them, in order to make the best decision.

One key limitation of our approach remains its reliance on habitat suitability modelling techniques to identify suitable translocation habitat. While we provide a method that ensures that both the input and output of those models are as useful and robust as possible, the general use of habitat suitability modelling has been criticized for simplifying the relationship between environment and species distribution (Davis, Jenkinson & Lawton 1998; Pearson & Dawson 2003; Norberg et al. 2012). Indeed, the principle behind this method is that simple environmental characteristics will determine whether a species is able to survive in a given location (Elith & Leathwick 2009). In reality, species are part of communities, and interspecific relationships will play a major role in their persistence. This may be particularly important when investigating the impact of climate change on the distribution of suitable habitats as species will most likely encounter new communities (Davis, Jenkinson & Lawton 1998; Pearson & Dawson 2003; Norberg et al. 2012). Another inherent assumption made when using habitat suitability models is that the current species range is a good indicator of suitable habitat. However, this may not be true (Osborne & Seddon 2012). Similarly, current absence from habitat may not signify unsuitability of non-occupied range. As a result, the habitat in which hihi can thrive may actually be more varied and much larger than predicted. Nevertheless, because results from the habitat suitability modelling concurred with those from the predictive population modelling, that is, temperature-based variables were the best predictors of suitable habitat for the hihi, we were confident that the predictions made by maxent were robust.

Implications for management and conservation

Climate change is one of the biggest threats to biodiversity, and successful adaptation tools are intently sought after. One such tool, assisted colonization, is growing in popularity, as conservationists and managers alike can see that, to persist in the long term, species threatened by climate change may need actions that may be considered extreme. Habitat suitability modelling is a quick – and fairly easy – way to map the distribution of current and future suitable conditions for species. As such, it has the potential to be a powerful ally in identifying species with populations that may need to be moved elsewhere to persist, and where to move them. Yet, past use of habitat suitability models has often produced results which cannot be reliably ascertained, limiting their usefulness for successfully planning translocation. Our work addresses this key issue for species management and conservation by demonstrating that there is a systematic way for habitat suitability modelling to yield robust and reliable results. It is only with a quantitative understanding of the relationship between environmental conditions and the long-term dynamics and persistence of candidate species that decision-makers can parameterize and evaluate habitat suitability models and use this powerful tool to plan assisted colonization under a changing climate. We thus encourage managers to consider targeted monitoring of populations first, in order to maximize the robustness of any habitat suitability modelling.

Acknowledgements

The authors thank the Department of Conservation that coordinated hihi recovery group and Tiritiri Matangi Island field staff for their ongoing investment in hihi conservation. The authors are also grateful to two reviewers for their insightful comments on this manuscript. National hihi conservation benefits from funding from Wesfarmers Industrial and Safety NZ Ltd, and the research presented here resulted from a Department of Conservation contracts to J.G.E.; A.L.M.C. was supported by the AXA fellowship. J.G.E. was supported by an RCUK Fellowship.

Ancillary