In a management perspective, the purpose of seabird monitoring programmes is to document the state and trends of selected populations, thereby enabling a targeting of attempts to counteract undesirable tendencies as early as possible. However, changes in seabird populations result from the combined effects of adult mortality, emigration and recruitment rates (which in turn are affected by a number of factors, including both stochastic and long-term climatic changes), as well as factors directly imposed on their environment by human interference (e.g. Nisbet 1989, Nettleship 1991, Cairns 1992, Croxall & Rothery 1991, Wooller et al. 1992).

The traditions of Norwegian seabird monitoring go back to the early 1960s when the late Einar Brun (Brun 1965, 1966, 1969a,b) paved the way by censusing most of the major breeding cliffs for auks (Alcidae) along the mainland coast. Brun was the first to use systematic census methods for cliff-nesting seabirds in Norway. By repeating many of his counts in the 1970s, he was able to demonstrate marked changes in several species and colonies (Brun 1979), thus illustrating the most important purpose of population monitoring: namely to document notable trends in numbers.

A dramatic decrease in breeding numbers of common guillemots Uria aalge in the 1960s and 1970s (Brun 1979) caused much concern and was one of the main reasons why the Norwegian Directorate for Wildlife and Freshwater Fish (now the Directorate for Nature Management) organised a national seabird programme (The Seabird Project) in 1979–1984. The programme was aimed at ''gathering data and material to use as a basis for the estimation of population sizes, detection of population changes and the effects of negative factors thereon, plus suggesting action which would result in a reasonable management of seabirds as a national and international resource” (Røv et al. 1984).

The monitoring of breeding seabirds in Norway was expanded and reorganised in 1988 (Lorentsen 1990) to involve nation-wide monitoring of eight species and regional monitoring of another six species. Except for the Briinnich's guillemot Uria lomvia, which has only marginal populations in mainland Norway, and the common eider Somateria mollissima, which is only monitored in southeastern Norway, we report results for all these species (see Table 1 for species names). Monitoring of wintering seabirds was primarily motivated by the concern for Norway's internationally important winter populations of great northern diver Gavia immer, white-billed diver G. adamsii, red-necked grebe Podiceps grisegena, common eider, king eider Somateria spectabilis, Steller's eider Polysticta stelleri, long-tailed duck Clangula hyemalis and velvet scoter Melanitta fusca (Røv et al. 1984). We therefore restricted our analyses to those species, but omitted the white-billed diver due to its small and highly variable numbers (Nygård 1994). In practice, however, all marine birds observed on localities within the 10 fixed counting areas (see Table 2) were monitored. The monitoring results for breeding populations have been reported annually (the latest by Lorentsen 1995), and an analysis of the national trends for wintering seabirds was recently presented by Nygård (1994).

Any environmental monitoring programme should be subjected to critical evaluations from time to time, to ensure that it will, in due time, produce answers to the questions it aims to address, that these questions are still relevant, and that the answers are sought in a rational way with respect to available resources. Using results from an assessment of Norwegian seabird monitoring, this paper deals with statistical and strategic considerations of general relevance in this context. The distance between the two points farthest apart on the Norwegian mainland coast, the lighthouses at Lista and Hornøy, is almost 1,800 km (or equal to the distance between London and Gibraltar). To minimise the risk of bias from merging different populations of each species, we therefore restrict our analyses to focus on regional trends only. We aim at illustrating what trends can be unveiled from the Norwegian monitoring data, and to examine by more extensive statistical methods to what extent they could have been obtained with less field effort. The results of this exercise are discussed on the background of what we consider to be the principal purpose of the activity, which can be summarised as follows:

The monitoring of seabird populations should, in as short a time scale as possible, document any significant changes in numbers of any population recognised as being particularly important, either from a conservation point of view (e.g. when numbers are of national or international importance, or when the population is threatened) or when the population is known to play a key role in its natural environment (e.g. by its transport of nutrients from sea to land). This is the ideal objective. In prac tice, the establishment of any monitoring programme necessitates a number of shortcuts to make the most out of the available funds. In this process, representativeness and realism are key aspects. Consequently, the monitoring of seabirds should be focused on a set of species for which appropriate monitoring methods are available and which reflect the variety of coastal habitats, ecological adaptations, environmental threats, and special concerns. No doubt, any such selection will be controversial in a cost-benefit perspective. Although the principal aim of monitoring is to document the trends, and not to explain them, we argue that seabird ecology and life-history patterns make it necessary to consider a number of other aspects when deciding on the distribution and sampling frequency of study plots.

Methods

Results

Monitoring of breeding seabirds

Among the breeding seabirds, the overall population trends differed highly from species to species as well as between different regions (Table 1). This was also evident for the species and areas selected for the more detailed analyses (Fig. 1). Fulmars Fulmar glacialis and gannets Sula bassana increased considerably in numbers during the monitoring period, but the fulmar is only counted in the southernmost colonies. The cormorant Phalacrocorax carbo carbo increased in most parts of the country, while shag P. aristotelis numbers increased in southern and central Norway only and may have decreased further north. There was an extreme year to year variation in the number of breeding pairs in the large shag colony at Lille Kamøy (ranging from 0 to 2,400), and the colony was empty during 1986–87 and 1992–94. Common gulls decreased in Telemark and maybe also in Nordland, although the latter trend was not significant. Populations of lesser black-backed gull Larus fuscus intermeclius in South Norway seemed fairly stable, whereas the populations of the northern subspecies L. f. fuscus have decreased so severely that there is a risk they may disappear. In contrast, both the herring gull L. argentatus and the great black-backed gull L. marinus tended to increase in all colonies monitored. The kittiwake decreased in most colonies along the coast. Large annual fluctuations make common terns Sterna hirundo difficult to monitor, but the population in Telemark has decreased significantly. Common guillemots decreased significantly along the entire coast. In northern Norway, dramatic reductions in numbers took place during 1985–87, when the stock of Barents Sea capelin collapsed. During those years, breeding numbers dropped by 80–90% and more than half the population was lost (e.g. Anker-Nilssen & Barrett 1991). The puffin was stable in the southwest and increased slightly in eastern Finnmark (Hornøy), whereas colonies in Lofoten and Vesterålen, and particularly at Røst which hold the largest populations, have decreased markedly (Anker-Nilssen & Røstad 1993, Anker-Nilssen & Øyan 1995).

Table 1.

Population trends for 12 breeding seabird species at individual localities or within different regions along the Norwegian coast. The significance of each trend was tested using Monte Carlo simulations. Where the number of study plots is zero, the whole locality or area was counted.

Figure 1.

Variation in numbers (expressed as percentages of the maximum year) within the monitoring plots for kittiwakes breeding at Hornøy (N_ma = 2,123 apparently occupied nest sites), common guillemots breeding at Runde (N_max = 2,791 individuals) and puffins breeding at Hernyken (N_max = 2,993 apparently occupied burrows) and within the monitoring localities for red-necked grebes wintering at Smøla (N_max = 274 individuals) and common eiders wintering in Trondheimsfjorden (N_max = 8,531 individuals, see Tables 1 and 2). Only plots counted in all years are included. For common guillemot, the mean of the 7–10 annual counts is used. The results of more detailed statistical analyses are presented in Table 3.

The significance of counting frequency

For the five species studied, the GLM procedure on ranked data (Table 3) revealed the same population trends as found by Monte Carlo simulations (see Tables 1 and 2), but the trends were substantiated to a much higher probability. It was also evident that when using this variance in bird numbers between plots, the trends could be detected even after halving the number of plots or localities counted or reducing the count frequency to every second year. As could be expected, there were also significant differences in bird numbers between plots, but after controlling for this variation the year effect was still evident in most cases. For common eider, however, no year effect was found, but the counts in Trondheimsfjor- den only indicated an annual increase of about 4% (see Table 2).

Table 2.

Population trends for a selection of seven seabird species wintering within the regular counting areas along the Norwegian coast. The significance of each trend was tested using Monte Carlo simulations.

Discussion

Our analyses are based on data for a wide range of species and geographical areas and show that for breeding seabirds significant trends in population size could be detected in many cases. For monitoring based on only one annual count of total bird numbers, there are few other appropriate methods which can be used to reveal such trends than the use of Monte Carlo simulations (but see Underhill & Piŷs-Jones (1994) for bootstrapping procedures and Thomas (1996) for a review of methods). This method has, however, obvious limitations and statistically significant trends can only be detected after a number of years (see Methods).

The existence of multiple plots for the estimation of within-year variation in bird numbers allows the use of more powerful statistical analyses that are capable of detecting trends at an earlier stage and often to a much higher probability. This conclusion was supported by the cases we examined using both methods, and where all trends could be detected by the GLM procedure even after halving the number of plots or localities and reducing the count frequency to every second year. This strongly suggests that some populations can be monitored adequately by less effort than at present, although this involves the risk of missing the chance to pinpoint when any disastrous reductions in numbers take place.

Figure 2.

Relationship between the number of sampling areas monitored and the sum-of-squares (SS) in a two-way ANOVA, partitioned into three components: year SS, plot SS and remainder SS, all expressed as percentages of total SS. Each plot represents a mean value derived from 500–5,000 ANOVAs performed for a corresponding number of randomised data sets (see Methods for details of the statistical procedure). Data sets and symbols as in Figure 1.

Our examination of the relationship between sums-of-squares and plot number also substantiates the conclusion that the effort can be reduced in some cases. All sources of variation tended to level off before the number of plots exceeded 10–15. For puffin and common eider this was far less than the actual number of plots being counted. Accordingly, a reliable measurement of the total variance in numbers of these species could have been obtained by reducing the number of plots or localities. It is, however, difficult to give any general recommendations on the number of sampling areas needed to capture an adequate proportion of the variance. The existing plots and localities are not mutually exclusive and, for logistical and financial reasons, there is no great effort saved by selecting such sampling areas at random.

Table 3.

Population trends and differences between plots for five selected seabird species and areas. The data sets are the same as those presented in Figure 2; the trends were also analysed after reducing the numbers of years and localities. The significance of each trend was first tested by Monte Carlo simulations using the total number of birds within the plots (i.e. only one data point per year). PAC indicate the percental annual change recorded for the selection. Then a GLM procedure (Littell et al. 1991) was applied to ranked data for individual plots, where the F-values and P-levels indicated are based on type III sum of squares. When choosing every second year or plot, the first year or plot was always included.

The variance in bird numbers depends on many factors, including the size of study plots, and annual, seasonal and daily fluctuations in bird numbers. Rothery et al. (1988) concluded that for common guillemots a monitoring scheme consisting of five plots of approximately 200–300 birds counted on 10 days in June was a reasonable effort (see also Wanless et al. 1982, who analysed Orkney data and concluded there was little extra to gain from conducting more than 10 counts). Even at Runde, where the number of guillemots within the plots averaged only 84.8 (N_plots = 25, N_years = 11, SE = 5.5), 5–7 plots could be considered sufficient to produce a reliable estimate of the year-to-year variation. Nevertheless, our study suggests that for many species monitoring only five plots is probably too small an effort. When the remainder effect is large, as it was for the two wintering populations, a much higher effort is needed to determine what proportion of the variance is attributable to variation between years. Vice versa, little effort is needed to produce very reliable estimates when the remainder effect is small, as it was for the three breeding populations examined. We have not tested the effect of plot size and the possible interdependency between counts in plots situated close together. Such factors are complex to deal with, as they will depend on both the physical structure of the area and the biological composition of the sample population (e.g. Anker-Nilssen & Røstad 1993).

A key question to nature management is how much a population should change before it deserves special attention, either through conservation measures or by beginning research aimed at explaining the underlying causes of the change. For seabirds, there is no simple answer to this question. Not only is there a great variety of life history strategies and ecological adaptations among the species monitored, but in many cases our understanding of the birds origins, and variations in their demography and distribution, is too poor to differentiate between actual population changes and bias caused by other sources of variation. The latter is particularly applicable to the winter populations. The different monitoring methods also represent a varying bias, for example by being unequally sensitive to daily changes in weather conditions and, in winter, to annual changes in ice conditions.

Population trend is most robust as an indicator of environmental changes that affect the survival of birds that have started to breed. Most seabirds have a delayed onset of breeding, and changes in survival of the younger life stages, i.e. nestlings, juveniles and immatures, will not be reflected in the breeding numbers until one to several years later, by which time the effects may well be obscured by other factors. With respect to food supply, seabird foraging and reproductive success are parameters more sensitive to changes than adult survival, although periods of extreme food shortage may be accompanied by high mortality of adults (e.g. Cairns 1987). In relation to oil pollution and fishing gear, however, acute reduction of immature and adult survival is the most likely consequence.

Thus, both the progress and the importance of a numerical change are highly dependent on a number of different factors. These factors also make it difficult to select a common mathematical model to describe the course of the various changes that are documented or predicted. However, with respect to major unidirectional changes occurring over a few (2–4) consecutive years, the shape of the trend deserves less concern than the total change in numbers, unless it indicates that numbers may bounce back the following year (see Greenwood et al. 1995). Similarly, slow but long-lasting changes represent no crucial statistical problem, as in most cases GLM or Monte Carlo simulation procedures will then be considered appropriate analysis tools for identifying the overall trends.

Kirby & Bell (in press) have recently proposed a quantitative method to identify alert conditions in population trends (i.e. what changes may deserve particular concern). Their method, which is based on the Underhill index and bootstrapping procedures (Underhill & PrŷsJones 1994), seems to be very sound and may well become an international standard for such decision-making. However, the principle will at least require considerable time to establish, and many will need an easier way to obtain reasonable advice. Considering the natural variation in seabird numbers and attempting to simplify a complex task (being well aware of the risk of oversimplification), we therefore suggest, as a rule of thumb, that conservation authorities should pay special attention to any change greater than ±25%, provided it is statistically significant at the 10% level. Such an event may well be the result of important changes in survival or recruitment rates, whether they are linked to altered conditions for the local population (e.g. affecting adult survival or reproductive performance) or (less frequently) to changes in the exchange rates between different breeding populations (especially in cormorants, terns, and some gull species). Intentionally, our criterion is made irrespective of the species and time spans involved. Whenever the 25% limit has been exceeded, the most probable explanatory hypothesis should be discussed and form a basis for consideration of relevant actions or additional research.

It is important to bear in mind that statistical probability of detecting a trend is not the only consideration to be made when deciding on the number of plots and their sampling frequency. The need to use the data in a wider context deserves particular attention. Monitoring only population numbers may well tell us something about the present status of the populations, but as most seabirds are characterised by low reproductive rates, delayed maturity and high adult survival, the time scale for population changes are long (e.g. Nisbet 1989). In addition, small or short-term episodes in the marine environment may first affect the number of breeding birds several years later, although survival rates or breeding performance may be affected more suddenly. Consequently, when a population change of special concern has been documented, it is often too late to unveil its causes and, thereby, to produce reliable predictions and identify proper actions. To overcome some of these difficulties, it is clearly necessary to 1) design sampling schemes that permit running comparisons of population trends with any parallel data series on other population parameters and environmental factors, and 2) to extend the seabird monitoring concept to include parameters that are likely to be among the most important for the regulation of the populations. In this context, increasing plot numbers and sampling frequency may well be essential in order to increase the sensitivity of statistical comparisons. One example is the puffin scheme at Hernyken, where a large number of small and evenly distributed sampling areas (only 10 m² each) enables the documentation of density-dependent population changes between successive years, which, in turn, can be correlated with recruitment rates (Anker-Nilssen & Røstad 1993, Anker-Nilssen & Øyan 1995).

Theoretical seabird population studies have demonstrated the importance of adult survival rates (e.g. Croxall & Rothery 1991, Erikstad et al. 1994), and even small changes in adult survival may have profound effects on population development. For instance, if the annual adult survival rate drops from 95 to 90%, a population will need to double its recruitment rate in order to maintain its numbers. Thus, adult survival should clearly be included in population monitoring programmes. Furthermore, breeding success, and preferably also food choice, should be monitored to be able to address the causes of possible changes in survival and recruitment rates (Nisbet 1989, Wooller et al. 1992), and to detect rapidly any major impact on these parameters.

The dynamics of seabird populations may only be understood when the principal population parameter data are coupled with long-term population trends. However, to fully understand also the dynamics of the ecosystem that seabirds inhabit and to be able to predict future changes in the populations, it is also necessary to include information concerning their most important food supplies in the analyses. The bulk of seabirds breeding in northwestern Europe are fish-eating species, and several species have been affected by variations in commercial fish stocks (e.g. Furness & Ainley 1984, Montevecchi 1993). Regrettably, up to now, multidisciplinary cooperation between seabird scientists and fisheries researchers has been almost totally lacking.

Monitoring programmes are often implemented with limited funds (e.g. Nisbet 1989), emphasising the selection of a limited number of key species and sites where it is possible to combine the monitoring of population numbers with the monitoring of adult survival rates and parameters affecting recruitment (e.g. breeding success and food supply). For breeding seabirds such key sites should, preferably, be selected in different oceanographical regions, and species representing the main trophic positions of seabirds in the marine ecosystem should be selected. When choosing between species and localities, special attention should be paid to the long data series that exists for some colonies with respect to population numbers, adult survival, food supply and breeding success.

In order to indicate the validity of results of the more extensive monitoring, key sites should be supplemented by a net of sites where the main purpose is to reveal an adequate proportion of the total variance in bird numbers at the lowest possible effort. At the low-effort sites, 10–15 monitoring plots will usually be enough, and in some cases it may also be sufficient to count every second year only. Caution is needed, however, as there may be substantial variation between years due to various sources of variation that cannot easily be controlled for (e.g. differences in weather conditions, field personnel, age distribution of birds present, timing of breeding, and the proportion of adults that skip breeding), but which clearly will reduce the possibility of detecting any factual trends in bird numbers from year to year.