- Top of page
A ‘dispersal sink’ (Andrewartha & Birch 1954; Pulliam 1988; Stacey, Johnson & Taper 1997) may be defined as any habitat in which, in the absence of immigration, the resident population is expected to decline to extinction (r < 0), because local births are insufficient to compensate for local deaths (births < deaths). Dispersal sinks are cited in the literature on metapopulation dynamics as a case in which immigration is clearly important in ‘rescuing’ populations from extinction (Stacey et al. 1997). Dispersal sinks are assumed to occur in suboptimal habitat, whereas ‘source’ populations (births > deaths) from which immigrants derive, are assumed to occur in optimal habitat.
Studies claiming to have demonstrated the existence of a dispersal sink were criticized by Watkinson & Sutherland (1995) because: (1) inferences regarding the demography of the resident population were often drawn from measurements of birth or death rates alone, rather than simultaneous measurements of both parameters; and (2) immigration was often inadequately measured, if at all. Only two studies on annual plants were judged by Watkinson & Sutherland (1995) to have adequately addressed these issues. Diffendorfer (1998) added a third study on another annual plant to the list, but otherwise drew the same conclusions as Watkinson & Sutherland (1995). Kadmon & Tielborger (1999) advocated an experimental approach to the evaluation of dispersal sinks because experiments they conducted on desert annuals failed to support the existence of dispersal sinks in conditions, where prior natural history studies had.
We propose that a third criterion should be added to those proposed by Watkinson & Sutherland (1995), namely an evaluation of the effects of measurement on immigration, births and deaths. The rigorous measurement of immigration in open populations in the field generally requires much more intensive handling and disturbance of the resident population than would otherwise be needed in a study of demography (Stenseth & Lidicker 1992; Clinchy 1999). If repeated capture and handling adversely effects either the birth or death rate of residents, then what might be a stable (births = deaths), or even a source (births > deaths) population in the absence of repeated capture and handling, may become a declining population (r < 0; Cypher 1997), that will then need to be ‘rescued’ from extinction by immigration. In addition, the artificial ‘removal’ of residents resulting from deaths due to capture and handling may actually induce an influx of immigrants (Stenseth & Lidicker 1992), leading to the conclusion that the site is a dispersal sink. Despite the logical necessity of testing for biases in demographic parameters attributable to handling effects, this is very rarely done in this or any other context (Wobeser 1994; Williams & Thorne 1996; Haydon et al. 1999).
Efford (1998) reported the existence of a dispersal sink among common brushtail possums (Trichosurus vulpecula Kerr) occupying optimal habitat in the Orongorongo Valley (OV) of New Zealand. Birth and death rates were both measured at the OV site, satisfying the first of the two criteria proposed by Watkinson & Sutherland (1995). However, the OV study did not adequately measure immigration. Since fewer than two-thirds of the resident young at the OV site were tagged while still with their mothers, ‘almost all unmarked yearlings … could be explained as surviving native young that had escaped tagging’ (Efford 1998, p. 510). Moreover, territorial expansion (i.e. ‘apparent recruitment of peripheral adults’; Efford 1998, p. 515) could not be distinguished from true immigration (defined by the new home range being disjunct from the abandoned one; Lidicker & Stenseth 1992, p. 23).
We conducted a spatially and temporally replicated removal experiment on common brushtail possums in uniformly suitable old-growth eucalypt forest in south-eastern Australia, designed to address the question: does immigration ‘rescue’ populations from extinction? We took every possible step to ensure that our measurement of immigration was made without error. By using multiple methods of capture and identification we attempted to completely enumerate all residents in the removal areas and in a surrounding ‘border zone’, so we could unambiguously identify unmarked individuals as immigrants, and distinguish range expansion from true immigration (Stenseth & Lidicker 1992; Clinchy 1999). In conjunction with measuring immigration, we also measured both birth and death rates in the resident population. Consequently, our study meets both criteria proposed by Watkinson & Sutherland (1995).
In this paper, we report: (1) evidence of handling effects resulting from our intensive efforts to measure immigration without error; and (2) the errors that may be introduced into population projections, as a consequence of such efforts. We use demographic estimates from our study, and the OV study, to demonstrate that whether or not a site is judged to be a dispersal sink (r < 0) may largely depend on whether or not deaths in association with handling are assumed to be ‘natural’. Because we distinguished between true immigration and range expansion in response to removals, we are able to evaluate the relative contribution of each of these processes to the apparent ‘rescue’ of our study populations from extinction. We discuss the process whereby apparent dispersal sinks may actually be the result of our efforts to accurately measure immigration, births and deaths.
- Top of page
Demonstrating the existence of a dispersal sink requires accurate measurements of immigration, births and deaths (Watkinson & Sutherland 1995). Accurately measuring immigration presents special problems (Stenseth & Lidicker 1992), which we overcame by extensive and intensive sampling and re-sampling (Clinchy 1999). In addition to evidence of an effect of handling on the survival of pouch-young (Table 3), several independent lines of evidence indicated that our intensive efforts to measure immigration without error also had an adverse effect on adult survival (Figs 1 and 2, Table 2). Whether or not our study site appears to be a dispersal sink (r < 0) depends on whether deaths associated with capture and handling are assumed to be ‘natural’ or not (Scenarios 1 vs. 3, Table 1, Fig. 3).
Females that were handled more frequently were less likely to survive (Fig. 1). This could be because handling stress directly affected survival or because animals that were already in poorer condition were more trappable (Sedinger et al. 1997). In the latter case, the survival rate ought to be the same in the presence or absence of handling. In comparatively long-lived animals, such as common brushtail possums, the observed age structure is a function of the average survival rate over many preceding years (Krebs 1999). Based on the observed age structure (Fig. 2), the long-term average survival rate was clearly higher prior to the study than it was during the study (= 81·3% when likely handling deaths are included; Scenario 1, Table 1), suggesting that the stress of handling was directly affecting survival. Our estimate of the ‘background’ survival rate (= 90·1% when probable handling deaths are excluded; Scenario 3, Table 1) corresponds much more closely with the observed age structure (Fig. 2).
Just as at the OV site, symptoms of wobbliness and boniness and rumpiness were observed at the PL site, in association with: (1) trap-deaths; (2) changes in ambient temperature (season); (3) each other (Table 4); (4) weight loss; and (5) heavy intestinal nematode and ectoparasitic mite infections. These commonalities clearly suggest a common cause at both sites. This cause could be handling stress or some other underlying disease state. The fact that females with ‘terminal’ symptoms (W, B + R, any other combination) were handled more frequently is not conclusive since, as noted above, animals already in poorer condition may have been more trappable. However, four lines of evidence indicate that handling stress was in fact the cause: (1) this suite of symptoms (W, B, R) closely resembles those (ataxia, muscular degeneration and fur loss) typically observed in association with capture myopathy and handling stress in other species (Fiennes 1982; Mann & Helmick 1996; Williams & Thorne 1996); (2) these same symptoms have all been reported in response to stress in captive common brushtail possums (Humphreys et al. 1984; Presidente 1984); (3) necropsies revealed evidence of histopathology and biochemical anomalies consistent with a diagnosis of handling stress; and (4) most importantly, our diagnoses of the proximate causes did not reveal any evidence that the deaths of animals with these symptoms were due to ‘old-age’ (senescence), starvation or any other underlying disease state.
We conclude that there is clear statistical (Figs 1 and 2) and symptomatological (Table 2) evidence of handling effects. To gauge the scope of the errors introduced into our population projections as a result of these handling effects, it was necessary to identify which individuals most probably died as a direct result of capture and handling (Table 1). Our estimate of the scope of these errors is doubly conservative, since inclusion in this category was restricted to: (1) only those individuals that showed those likely symptoms of handling stress that were known to be associated with poorer survival (W, B + R, any other combination); from among (2) only those individuals whose deaths were otherwise undiagnosable (Table 2).
If our conclusions about the adverse effects of handling are incorrect, then the PL site would appear to be a dispersal sink (Scenario 1, Table 1, Fig. 3). This is clearly inconsistent with the modelled prediction that the PL site represents optimal habitat for possums (NPWS 1994), since dispersal sinks are generally assumed to occur in suboptimal habitat. Moreover, the measured rate of true immigration is not sufficient to maintain the population size under this scenario (Scenario 1 with immigration, Fig. 3), in which case the observed densities are inexplicable.
While the demographic results from the PL site could be dismissed as the consequence of ‘a couple of bad years’, results from the OV site are based on more than 15 years of data (Efford 1998). Under the assumption that handling is innocuous (Scenario 1), the resident population at the OV site is projected to decline at virtually the same rate as the resident population at the PL site (Table 1, Fig. 3). As with the PL site, the projected decline at the OV site is clearly inconsistent with the fact that it too appears to be in optimal habitat (Efford 1998).
The remarkable similarities in the projections for both the PL and OV sites (Table 1, Fig. 3) may at first seem surprising. Common brushtail possums were introduced into New Zealand from Australia (Cowan 1990), and the two countries are ecologically very different. The similarities in projections are less surprising, however, considering the overwhelming importance of adult survival to the projected population growth rate (Table 5). As a consequence, similarities in the factors affecting adult survival are largely sufficient to explain the similarities in the projections. Since predator densities (Fitzgerald & Karl 1979; Catling & Burt 1995; Efford 1998; Clinchy 1999) and parasite burdens (Heath et al. 1998; Obendorf et al. 1998; Stankiewicz et al. 1998; Clinchy 1999) are comparable at both sites, and deaths due to predators and parasites were not numerous enough to have created a dispersal sink at the PL site (Table 2), it seems likely that an additional factor must also have been at work at the OV site. We suggest that, just as at the PL site, that other factor is handling stress.
Logically, animals found moribund in traps represent observable deaths in association with capture and handling that may or may not be the direct result of capture and handling. As is typical of animals that die during capture (Fiennes 1982; Williams & Thorne 1996), no gross pathology was evident in animals found moribund in traps at either the PL or OV sites. Where there is no other evidence it will not be possible to say definitively whether such deaths are ‘natural’ or not. Since this may affect whether a site is judged to be a dispersal sink, we propose that all studies reporting the existence of an apparent dispersal sink should also report the frequency of all observable deaths that occur in association with capture and handling.
The PL and OV sites would both appear to be dispersal sinks if animals found moribund in traps were the only ones deemed to have died as a direct result of capture and handling (Scenario 2, Table 1, Fig. 3). Of course, deaths due to handling may not always occur in the presence of the researcher. Animals may be released, apparently unharmed, that may not die until days or weeks later (Fiennes 1982; Williams & Thorne 1996). The perplexing fact that females at our site were ‘dropping dead’ for no apparent reason (Table 2) would not have been evident if these animals had not been radio-collared. In addition, the scarcity of both predators and scavengers meant that we were often able to retrieve intact carcasses for diagnosis, something that is rarely feasible (Wobeser 1994). Since results regarding the OV site were based solely on trapping (Efford 1998), it is not possible to evaluate how many animals at that site may have been ‘dropping dead’ following their release. If we assume that such deaths occurred with the same frequency at both sites, the resident population at the OV site is projected to be stable (r ≅ 0·0; Scenario 3, Table 1, Fig. 3).
Given the strikingly similar symptoms of handling stress and the parallels in projected outcomes at both the PL and OV sites, it seems reasonable to conclude that the resident population at the OV site was unable to replace itself (births < deaths) largely because of the trapping study. Since numbers remained stable at the OV site (Efford 1998, his Fig. 3), new animals must have been continually entering from outside the study area. The results from our experimental removals indicate that the reason why these new recruits were drawn into the area may also have been largely because of the trapping study.
In the 2 years following our ‘pulsed’ experimental removal of the 19 core residents, 15 females that were originally resident in the surrounding ‘border zones’ expanded their ranges into the core and re-occupied many of the dens of the core residents (Clinchy 1999). Range expansion in response to the disappearance of a neighbour is common in territorial species (Carpenter 1987). Efford, Warburton & Spencer (2000) called this the ‘vacuum effect’. Given that deaths due to capture and handling are analogous to removals, the OV study may be viewed as a low-intensity, ‘press’ (continuous) removal experiment. Reproduction (Table 3) and survival (Fig. 1) will be higher among animals living just outside of the study area precisely because they are not repeatedly captured and handled. The creation of vacancies on the study grid as a result of deaths due to capture and handling presumably draws in these neighbours from just off the grid (i.e. the vacuum effect), and numbers therefore remain stable. Efford (1998, p. 515) acknowledged that such ‘apparent recruitment of peripheral adults’ at the OV site, could not be ruled out.
We caution against dispersal sink studies becoming self-fulfilling prophecies: (1) deaths due to capture and handling create a declining population (r < 0) fit to be ‘rescued’ from extinction by immigration; (2) the resulting ‘removal’ of residents creates vacancies that draw in apparent immigrants in the form of neighbours intent on securing all or part of the empty ranges (i.e. the vacuum effect); and (3) thus is the population ‘rescued’ from extinction by ‘immigration’. This conundrum is not easily resolved. Distinguishing between true immigration and the apparent immigration entailed by the vacuum effect (Lidicker & Stenseth 1992, p. 23) will probably require even more intensive capture and handling.
How often do handling effects generate apparent dispersal sinks? At present this is unanswerable, because so few studies have tested for biases in demographic parameters attributable to handling effects (Wobeser 1994; Williams & Thorne 1996; Haydon et al. 1999). Haydon et al. (1999) recently reported that significant biases in population projections for snowshoe hares were the result of handling effects. Biases in population projections resulted from comparatively small effects on adult survival in both possums (≅ 8·8%; Scenarios 1 vs. 3, Table 1) and hares (≅ 5·0%; Haydon et al. 1999). Given the general sensitivity of population projections to effects on adult survival, we suggest that any evidence of adverse effects of handling on adult survival, in particular, should raise alarm bells about the accuracy of estimated population growth rates and resulting conclusions about dispersal sinks (i.e. r < 0).
We examined 17 of the 22 animal studies cited in Diffendorfer’s (1998) recent review of empirical studies of source-sink dynamics (the remaining five were either inapplicable or inaccessible) for any acknowledgement that handling effects may have biased the results. Only four studies did acknowledge this (Southern 1970; Nettleship 1972; Grant 1975; Stearns & Sage 1980). Notably, those that did were significantly (t15 = 2·78, P = 0·014) older (mean publication date = 1974) than those that did not (mean publication date = 1986). Those that did also tended to be longer (mean length = 36 pages) than those that did not (mean length = 16 pages), but the difference was not significant (t15 = 1·64, P = 0·122). There are at least two explanations why more recent studies do not acknowledge handling effects: (1) improvements in techniques and procedures have eliminated handling effects; or (2) researchers are now more reluctant to report or discuss potential handling effects.
We have discussed how biases attributable to handling effects may lead to erroneous conclusions about ecological processes. If handling effects are not reported there will also be no way to anticipate them and thereby attempt to reduce potential harm (Sedinger et al. 1997). We chose to focus our analyses on handling per se since this was the most invasive procedure animals were subjected to. Of course trapping, independent of handling, may also have adverse effects. Even apparently healthy animals were significantly more likely (binomial P < 0·001) to have lost weight (32) than not (8) over a 3-night trapping session. Providing more bait and processing animals at night, rather than waiting until the next day, are both ways to reduce potential weight loss. Moreover, while it is necessary to sample several (3) times over a short period in order to accurately estimate population size (Krebs 1999), sampling every second or third night, with ‘nights off’ in between, ought to be just as effective. Viggers & Lindenmayer (1995) recommended the use of a sedative to reduce the level of stress induced when handling possums.
Our work addressed an important question in conservation biology regarding the role of immigration in ‘rescuing’ populations from extinction (Clinchy 1997; Sih, Jonsson & Luikart 2000). To answer this question properly will generally require extensive handling and disturbance of the study population(s) (Stenseth & Lidicker 1992; Watkinson & Sutherland 1995; Clinchy 1999). If the resulting effects on the health and wellbeing of the subjects are not considered this may lead to erroneous conclusions, and the wrong conservation actions being taken. Consequently, conservation and animal welfare concerns are both better served by the explicit reporting and discussion of the potentially adverse effects of measurement.