In order to explore the results of the simulation in the field, I used a combination of new data on Black-legged Kittiwakes Rissa tridactyla and Common Guillemots Uria aalge from two colonies on the Orkney Islands, northern Scotland, and results from previously published experimental work at St. Abb's Head National Nature Reserve (NNR), Scottish Borders (Beale & Monaghan 2004b). From Beale and Monaghan (2004b) it is possible to estimate the shape of the dose–response curve over measured visitor levels for both seabird species at St. Abb's Head (sigmoid and exponential for Guillemots and Kittiwakes, respectively). Before using the published disturbance relationships to determine visitor management strategies, it is important to assess their generality. Consequently, I first assessed how generally applicable the published dose–response curves are by using them to predict nesting success in two different colonies in Orkney: Mull Head (east Mainland) and Marwick Head (west Mainland). Nesting success of Kittiwakes and Guillemots was measured in 2003 according to Joint Nature Conservation Committee monitoring guidelines (Walsh et al. 1995). Using site visits and photographs of the monitoring plots, I estimated the parameters identified in Beale and Monaghan (2004b) as important in determining nesting success in each species. In June 2004 I measured human visitor patterns in the same way as used to generate the original models. Assessment of visitor numbers was not possible during 2003, but the local recorder was confident that the distribution had varied little between years (D. Paice, pers. com.).
For both colonies, I calculated the average value of each parameter and used these means to estimate the mean nesting success for the monitoring plot. I further ranked the nests within each monitoring plot according to the disturbance pressure (people minutes per hour divided by distance to the nest), and produced separate estimates of nesting success for the top (high disturbance) and bottom (low disturbance) thirds of the ranked list. Where the disturbance pressure was the same for several nests and a division required, I selected nests from the tied rank at random. As there may be considerable variation between years and sites in seabird nesting success (Murphy & Schauer 1994), absolute values of the predictions offer a less stringent test of the relationship than the relative changes in predicted and actual values. Consequently, I assessed relationship accuracy by comparing both absolute precision and the magnitude and direction of changes within each colony.
Guillemot nesting success predicted by the published dose–response curve was remarkably accurate at both Orkney sites, particularly for Marwick Head, though in fact there was relatively little variation in disturbance pressure between the most and least disturbed sites (Fig. 4). The small degree of variation in disturbance pressure resulted in total overlap of 95% confidence intervals within distance classes indicating that these divisions were not statistically different. Nevertheless, the models successfully predicted the observed direction of change in mean nesting success at Marwick Head, despite nesting success increasing with disturbance pressure. This contrary pattern is likely to be due to a negative correlation between disturbance pressure and the number of Guillemots neighbouring the nests (a factor known to reduce nesting success, perhaps due to increased probabilities of eggs being knocked from ledges: Beale & Monaghan 2004b), something not observed at Mull Head (Marwick r2 = −0.226, n = 115, P = 0.015, Mull r2 = 0.149, n = 109, P = 0.123). It seems therefore that the Guillemot dose–response curve is likely to be generally applicable. By contrast, predictions of Kittiwake nesting success based on the models in Beale and Monaghan (2004b) were not able to predict nesting success in the new colonies accurately, so the dose–response curve may be site-specific in this species.
Figure 4. Predicted (asterisks) and actual (diamonds) nesting success (± 95% confidence limits) for Guillemots nesting in two Orkney colonies in 2003. Sites are divided into three groups based on the ranked level of disturbance at each nest: high, average or low. Note that predictions accurately mirror direction of change at both sites.
Download figure to PowerPoint
Given these patterns, it is possible to assess optimal management for Guillemots at each site. Clearly, as predictions for Kittiwakes were unreliable away from St. Abb's Head, it is not possible to provide general management guidelines for this species. At St. Abb's Head, however, an exponential dose–response curve implies that management that aims to spread visitors around the reserve as much as possible will always be the preferred solution. The sigmoid dose–response curve for Guillemots indicates that optimal management is dependent on the position and maximum rate of increase in the dose–response curve relative to the current disturbance pressure at each site. From Beale and Monaghan (2004b) it is evident that the maximum rate of increase occurs at a disturbance pressure of c. 0.62 people minutes/h/m and the current mean at St. Abb's Head is approximately 0.5. Consequently, the current levels at this site fall within the exponential stage of the sigmoid relationship, where an even spread of visitors is again favoured. In Orkney, at Mull Head, mean disturbance pressure was approximately 0.3 people minutes/h/m, and again an even spread of visitors would be optimal. Only at Marwick Head, where mean disturbance pressure reached 2 people minutes/h/m, was increased aggregation optimal.
Before discussing in detail the implications of these results it is perhaps interesting to explore the biological mechanisms that can lead to the different dose–response curves discussed above. In the context of the empirical observations described here, it is important to note that although human disturbance does lead to reduced nesting success in Guillemots, most nest failures are not a simple direct consequence of disturbance (Beale & Monaghan 2004b). In this species and at the sites visited, visitors do not flush birds directly from nests (Beale & Monaghan 2004b) and the daily number of nest failures is not related to the daily number of visitors (Beale & Monaghan 2005). The mechanism that leads to failure must be indirect and may relate to physiological costs associated with human disturbance and is discussed in more detail below (Beale & Monaghan 2004b).
In the more general case, it is important to note that an exponential or sigmoid dose–response curve is probably the most appropriate shape of the dose–response curve, although over measured values of visitor pressure linear and exponential may also be possible. If disturbance impact is measured in terms of nest failure, for example, once a nest and any replacement clutches have failed, the impact cannot continue to grow, so a maximum must exist. It is, however, quite possible that the visitor levels necessary to cause total failure of all susceptible nests, or even a slowing in the rate of failures, are practically unobtainable and a linear or exponential dose–response curve is appropriate. This also indicates the most likely biological mechanism that could result in a logarithmic dose–response curve: responses are a linear function of disturbance until all susceptible birds have failed and a plateau is reached. This may be the case if birds have a fixed desertion probability that varies as a linear function of disturbance pressure. More complex are the exponential-type dose–response curves. In this case, birds are relatively unresponsive to small levels of disturbance, but over a certain threshold impacts start to increase dramatically. Such patterns could be the result of purely behavioural processes: if birds assess risk on the basis of a threshold rather than a simple linear increase, an exponential dose–response curve is likely. They could also be caused mechanistically, however, if the proximate cause of nest failure is related to a physiological process such as increased heart rate (Nimon et al. 1995, Wilson & Culik 1995, Beale & Monaghan 2004b). If birds abandon nesting attempts when reserves are depleted below a certain threshold (Coulson & Johnson 1993, Cadiou & Monnat 1996), and the rate of depletion is affected by disturbance pressure (Nimon et al. 1995), birds are unlikely to be affected by disturbance until disturbance pressure exceeds that required to tip the physiological threshold to abandonment. Clearly in the case of a sigmoid relationship, an exponential process and a logarithmic process are combined to give the final result, and mechanisms at each stage are likely to be similar to those described above.
It is important to examine what impact differences between the real world situation and the model system may have on the generality of the results. First, it is obvious that real seabird colonies are not distributed on an evenly spaced grid and are often located in areas where access around the entire periphery of the reserve is not possible. The results of this simulation, however, suggest that what is important is not so much the absolute distribution of either people or birds, but the relative variation in visitor pressure that nests receive under scenarios of increased or decreased spatial aggregation of visitors. Whatever the distribution of nests and visitors, increasing aggregation will increase the range of visitor pressures experienced by nests within the colony, and it was this variation in range that results suggested should lead to a change in management option from always spreading visitors around to increasing aggregation under certain circumstances. The potential for such a threshold to exist in the field therefore is clear. By contrast, the numerical conditions (i.e. the necessary but not sufficient conditions where steepness > 2 and step < 22) that lead to this threshold are strongly dependent on the assumptions made in the simulation. A different form of sigmoid curve and a different range of visitor numbers and pressure would result in significantly different numerical conditions, but no change in the overall rules of thumb. It is clear, for example, that only when the dose–response curve is approximately logarithmic or sigmoid can visitor aggregation be favoured. In practice it seems unlikely that the benefits of aggregation when the dose–response curve is logarithmic are biologically meaningful, so considering only the case of a sigmoid relationship, it can be seen that only when the maximum rate of increase is below the mean visitor levels can aggregation be preferred, but meeting this condition is not sufficient to identify the preferred management as the maximum rate of increase also needs to be known.
In general, results from the simulations suggest that the optimal visitor management depends on the sensitivity of the species, the shape of the dose–response curve and on the levels of visitor disturbance that are likely to be experienced at any site. It is not possible therefore to determine a general rule to explain whether visitors to a nature reserve should be aggregated into one area or spread as thinly as possible across the area as a whole. However, empirical observations of the shape of the dose–response curve, coupled with observations of visitor distributions within a nature reserve, can allow a site- and context-specific solution to be determined. If, however, visitor numbers change substantially from the levels present when this is determined, it is important to reassess this management. Moreover, if visitor pressure is fairly low and the birds being visited are not suspected of being particularly susceptible to disturbance, it is likely that in most cases spreading visitors as thinly as possible is likely to be optimal. In practice, management aimed at aggregating visitors into one small area is, perhaps, more easily achievable than spreading visitors more thinly (Sutherland 2000), and some thought about how this may be achieved is necessary but is impossible without further research into how people behave on nature reserves (e.g. Underhill-Day & Liley 2007). It may be, for example, that visitors aim to spend a fixed period of time in a reserve, in which case building a good network of paths that encourage visitors to keep wondering what is around the corner rather than lingering in one site for the whole time would achieve this end. Until further research on practical ways to manage visitor access is completed, however, such suggestions are mere speculation.
In conclusion, the empirical observations and simulation results suggest that unless a species of seabird (and probably other colonial animals visited during the breeding season) is very strongly sensitive to disturbance or in rare cases where visitor pressure is very high, maintaining as even a spread of visitors as possible is likely to be the most generally preferable visitor management strategy. It is interesting that this contrasts rather markedly with many current visitor management strategies where efforts are made to contain visitors in one small area (Higham 1998, Nisbet 2000). It is also important to note that the reductions in impact that may result from optimal management could be relatively small and potentially not biologically meaningful: a disturbance-induced decrease in overall nesting success of around 10% from the expected nesting success in the absence of visitors was estimated at St. Abb's Head (Beale & Monaghan 2005), so without reducing overall visitor numbers any benefits gained from spreading visitors further around the reserve are likely to be minimal. Therefore, only when disturbance really does have an important impact on demographic processes is any attempt to manage the spatial distribution of visitors likely to have significant conservation benefits.