Winter waterfowl counts

In winter, one coordinated mid-week survey of the waterfowl using each lake has been undertaken in the middle of the months October to March, since 1989/90. On each occasion, individual members of a 9–12 person survey team visited each lake in the study area, recording the numbers and species of waterfowl present. The analyses described in this paper were based on average counts for each lake over all surveys for the six winters 1989/90 to 1994/95. There were 36 possible count occasions for each lake, and the majority of lakes were surveyed on every occasion. Arithmetic mean counts were used in preference to medians to reflect differences between lakes in the overall usage by each species; seasonal patterns and differences between years were not analysed.

Lake characteristics

Extensive, if rather incomplete, information exists on the physical, environmental, recreational and biotic characteristics of lakes in the Cotswold Water Park (Millett 1992, 1993, Bell 1996). Variables used in the analyses presented in this paper are summarised in Table 1.

As an indicator of the trophic status of each lake, average Trophic Ranking Scores (TRS values) for emergent and open water plants in each lake were calculated according to Palmer (1989). This author listed TRS values for aquatic plants based on their range of tolerance of trophic status, ranging from 2.5, for species characteristic of dystrophic and oligotrophic waters, to 10, for species characteristic of eutrophic waters. Aquatic plant species occurring in the study area are listed by Bell (1996). The commonest open water species are Elodea nuttallii (TRS value 10), Myriophyllum spicatum (TRS value 9) and Potamogeton pectinatus (TRS value 10). The commonest emergent species are Typha latifolia (TRS value 8.5), Eleocharis palustris (TRS value 5.8), Mentha aquatica (TRS value 7.3) and Schoenoplectus lacustris (TRS value 7.7).

Wintering waterfowl community types

Analysis of community composition was based on the 12 most abundant and widespread species in the study area (see Table 2), to avoid domination of the results by the patterns of occurrence of relatively rare species. The data were transformed by log_e(x+0.l) to down-weight the importance of high values, and centred about both species- and lake-means (Digby & Kempton 1987) so that differences between lakes were measured in terms of species-composition rather than absolute abundance.

The transformed and centred data matrix was analysed in two ways. Firstly, the lakes were classified into groups with similar waterfowl community types using an hierarchical clustering procedure. The CLUSTER procedure of the SAS statistical package (SAS Institute Inc. 1988) was used to perform clustering of euclidean distances by the Lance & Williams (1966) flexible strategy, with b set to -0.25 (see Webster & Oliver 1990). Secondly, canonical variates analysis (CVA) of the data matrix was used to ordinate the lake communities along gradients of species-composition in a way which emphasised the differences between the cluster groups (Digby & Kempton 1987). The results of the CVA were used to aid interpretation of the cluster analysis. The analysis was performed using the CANDISC procedure of the SAS statistical package (SAS Institute Inc. 1988).

One-way analysis of variance (ANOVA) was used to compare lake characteristics and the abundance (absolute numbers and densities) and diversity (number of species) of wintering waterfowl between cluster groups, using transformed data where appropriate (see Table 1). The Shannon diversity index H′ (e.g. Begon, Harper & Townsend 1986) was used to measure wintering waterfowl diversity in each cluster group.

Relationship between waterfowl and lake characteristics

Lake area and perimeter length are crude measures of the amount of lake habitat available to wintering waterfowl. For each of the 12 most abundant and widespread species in the study area, the lake size parameter which appeared best to measure the amount of available habitat was selected by tests of linearity in regression analysis. The selected measure of lake size was used in all further analyses for each species, e.g. in calculating waterfowl densities.

Analysis of covariance (ANCOVA) was used to compare waterfowl numbers between assemblage types, with lake size (area or perimeter length, as appropriate) as a covariate; both waterfowl numbers and lake size were log-transformed (see above). The analysis proceeded in two stages. First, the interaction between assemblage type and lake size was examined, testing the hypothesis that the slope of relationship between log-transformed waterfowl numbers and lake size differed between assemblage types. If this interaction term was not significant (P > 0.05), the analysis proceeded to the second stage, which was to examine the independent effects of lake size and assemblage type in a reduced ANCOVA model without an interaction term. A significant effect of assemblage type indicated that the value of a given unit of habitat for a species differed according to assemblage type.

Regression analysis, with selection of variables using Mallow's C_p criterion (Burnham & Anderson 1992), was used to explore the relationship of waterfowl densities with lake characteristics. The REG procedure of the SAS statistical package (SAS Institute Inc. 1988) was used to select models and estimate regression parameters. Waterfowl densities were transformed by log_e(x+0.01), and 36 lake characteristic variables were considered for inclusion in models (see Table 1); there were complete data for only 45 lakes. Owing to this small sample size, it was not possible to consider models within groups of lakes defined by assemblage types. Instead, assemblage type was included as a fixed term in the models, parameterised as a group factor by a set of dummy variables. For comparison, models without assemblage type and with assemblage type only were also estimated.

Given the small sample size and large number of variables, the likelihood of spurious results is large, arising particularly from cross-correlations between the variables. Detailed interpretation of models for individual species would not be justified. Instead, generalisations were sought, by examining the direction and magnitude of effects across species: t-values of the partial regression coefficient for each selected variable (Sokal & Rohlf 1981) were averaged across models, considering positive and negative effects separately. These averages were calculated for groups of species defined by dabbling, diving, grazing and fish-eating waterfowl guilds.

In order to identify models which were more easiy interpreted for individual species, a reduced set of variables was defined which summarised what were considered to be the most important habitat factors and which avoided the major cross-correlations (see Table 1). This approach was adopted after it was found that preliminary analyses, in which environmental variables were summarised and reduced in number by finding principal components (Webster & Oliver 1990), improved neither the amount of variation explained by models nor their ease of interpretation.

Wintering waterfowl communities

Five main types of wintering waterfowl assemblage were distinguished by cluster analysis of data for the 12 most abundant species (Fig. 1, Table 2).

Figure 1.

The hierarchy of relationship between wintering waterfowl assemblage types in the Cotswold Water Park identified by cluster analysis.

Lakes in group 1 were important for dabbling waterfowl, with high densities, absolute counts and proportions of mallard Anas platyrhynchos, teal A. crecca and gadwall. Other species were unimportant in this assemblage compared with their occurrence in other groups - this applied particularly to pochard and coot - although there were relatively high densities of mute swan Cygnus olor, great crested grebe and tufted duck.

There were relatively low numbers of most species except goldeneye Bucephala clangula and Canada goose Branta canadensis in group 2 lakes. Mute swan and cormorant Phalacrocorax carbo were also prominent members of this assemblage.

Large numbers of many species occurred in lakes of group 3, but the assemblage was chiefly characterised by the diving waterfowl, particularly coot and pochard. Wigeon Anas penelope and Canada goose, both grazing species, were also important in this assemblage.

Group 4 lakes were important for all diving waterfowl, particularly pochard, although at lower densities than in assemblage 3. Dabbling and grazing waterfowl were unimportant in this assemblage.

Lakes in group 5 supported relatively low numbers of most species. Mute swan, tufted duck and great crested grebe were the most characteristic species of this assemblage, present in relatively high densities.

Table 2.

Summary of wintering waterfowl counts in groups of lakes defined by assemblage types (1–5): (A) average counts; (B) average proportions (%) and (C) average densities (birds/ha^-1) of the 12 most abundant species; (D) average counts, density and diversity of total wintering waterfowl.

The major distinction in these assemblage types was between those dominated by dabbling species and those dominated by diving species - assemblages 1 and 4 represent the respective extremes along this axis. The dichotomy is well represented by CVA of the waterfowl data (Fig. 2). Assemblages 1–4 are well separated on the first two axes (see Fig. 2a). The first axis (CAN1) separates assemblages 4 (negative) and 1 (positive), and the canonical structure of the species contributions (see Fig. 2b) shows this to be a distinction of diving species in the negative direction and dabbling ducks and wigeon in the positive direction. The second dimension of variability, represented by CAN2, separates assemblage 3 (negative) from assemblage 2 (positive); pochard, tufted duck and coot make negative contributions, whilst other diving waterfowl and teal make positive contributions. Assemblage 5 is central to both CAN1 and CAN2, showing that it is not distinct on either of these dimensions of variability.

The lakes in each group differed significantly (ANOVA, P < 0.001) in total numbers, densities and diversity of waterfowl (see Table 2). Group 3, important for diving and grazing birds, had highest numbers of species present, total waterfowl numbers and densities, whereas group 2, important for goldeneye, had the lowest values. The dabbling assemblage of group 1 had the highest Shannon diversity. Group 4 had almost as low a Shannon index as group 2, indicating the dominance of the diving bird assemblage by just a few species.

Figure 2.

The results of canonical variates analysis of wintering waterfowl count data: (A) scores on the first two axes (CAN 1 and CAN 2), according to species and (B) according to assemplage types.

The lakes were also distinct in terms of lake characteristics (Table 3). Substratum conditions appeared to be important: lakes in groups 1 and 2 generally had a higher occurrence of sand and gravel in the substratum than those in groups 3–5. Water chemistry was also important in defining the divide between the dabbling and diving bird assemblages: both DOM (Dissolved Organic Matter) and dissolved phosphate levels were highest in group 1, the dabbling assemblage, and lower in groups 3–4, more important for diving birds. The assemblage with the highest waterfowl abundance, densities and species richness, group 3, was defined by the lowest dissolved phosphate levels. Lakes in group 4, supporting a wintering assemblage of mainly diving waterfowl, were typically the largest and deepest in the study area; lakes in group 2, supporting the poorest wintering assemblage, were the smallest and most shallow. This poorest group was also at the lower end of the scale in terms of aquatic invertebrate and plant diversity and plant abundance; these measures were highest in group 1, lakes with a diverse dabbling waterfowl assemblage. Average TRS values, calculated using both open water and marginal plant species, were lowest in group 3 lakes and highest in groups 1 and 5. There was no significant overall difference between the groups in lake age, but lakes with the diverse diving assemblage (group 3) tended to be younger (13.0 ± 2.3 years, mean ± S.E.) than other lakes (25.9 ± 1.8 years) (F_l.80 = 4.43, P = 0.038).

Table 3.

Significant differences between wintering waterfowl assemblage types (1–5) in site characteristics, tested by one-way ANOVA.

Wintering waterfowl and habitat

Analysis of covariance shows that there was a significant effect of assemblage type on the relationship of abundance with lake size in all 12 species considered (Table 4). The linearity of the relationship (regression slope) was similar between assemblage types (nonsignificant interaction term) in eight species, but the assemblage main effect was highly significant in each case, indicating differences between the lake groups in terms of the value of a given unit of habitat or open water. In the remaining four species, there were differences between the assemblage types in the slope of regression (significant interaction term), suggesting even more profound differences between the lake groups. The ANCOVA models explained much of the variance in waterfowl numbers, so that it can be seen that lake size and the lake characteristics related to the differences in assemblage types (see Table 3) account for much of the value of these lakes for wintering waterfowl.

Regression models for waterfowl densities are summarised in Table 5. Assemblage type on its own was significant for all species except goldeneye. In six species, the models with assemblage type included more environmental variables than those without, suggesting that when differences between lakes are accounted for, more environmental factors are identifiable. In great crested grebe, wigeon and tufted duck, better models (larger R²) were identified when assemblage type was included, but with a smaller number of environmental variables, indicating a great deal of consistency and predictability about their occurrence in different assemblages. Conversely, in gadwall and goldeneye, the model was poorer (smaller R²) for being constrained to include assemblage type, suggesting that the interpretation of the assemblage types in which these species occurred is not particularly meaningful in terms of real biological communities. Overall, the models with both assemblage type and environmental variables were very successful in accounting for the densities of individual species on lakes, explaining up to 95.7% of the variance.

Table 4.

Summaries of ANCOVA models explaining wintering waterfowl numbers in terms of assemblage types and lake size. Lake size expressed as area for diving and fish-eating species and perimeter length for dabbling and grazing species. Where the interaction term is not significant, the R² and the P-values for lake size and assemblage type are for a model excluding this term.

The contributions of the variables to the models are summarised according to ecological group of waterfowl in Figure 3. These results do not bear detailed interpretation, partly because correlations between the variables make their effects difficult to construe. However, the following points emerge: a) dabbling species generally favoured the most organically-rich lakes - high in DOM, detritus and dissolved nitrates (see Fig. 3a); b) diving species benefited from an abundance of aquatic invertebrates over gravelly substrata (see Fig. 3b); c) diving waterfowl appeared more susceptible to the effects of water-based recreation than any other group of species; d) grazing species were favoured by the use of lakes for both coarse and game fishing (see Fig. 3c); e) more than any other group, grazing waterfowl occurred on lakes with little marginal and other vegetation; f) fish-eating species favoured the deeper lakes (see Fig. 3d); g) fish-eating species were positively associated with some water-based recreational activities, but were adversely affected by the use or management of lakes for game fishing.

Models for waterfowl densities identified using a reduced set of environmental variables but constrained to include assemblage type are summarised in Table 6. Smaller amounts of variance were explained than by full models (see Table 5), for all species except gadwall, but some important effects were highlighted. The most important of these was the strong impact of water-based recreation on densities of the two Aythya species (pochard and tufted duck). Some effects of potential food supply were also apparent: tufted duck favoured lakes with greater diversity and abundance of aquatic invertebrates, whilst pochard also favoured lakes with abundant aquatic plants. There were also some spurious correations, however, such as the strong positive association between Canada goose densities and aquatic invertebrate diversity, which seems unlikely to be a causal relationship. The trophic and base-status of lakes appears to be relevant for some species. Densities of pochard and coot declined with increasing eutrophication, as measured by the average TRS-values for open water plants, whilst cormorant and gadwall appeared to be favoured by increasing dissolved phosphate levels. Great crested grebe and, to a lesser extent, teal densities were negatively correlated with alkalinity, whilst cormorant showed a strong negative association with pH. The model for cormorant highlights some other factors which may be important in defining habitat for this species: lake islands may be important disturbance-free resting areas, whilst the effects of water depth may be related to the fish populations of lakes and the ease of foraging.

Table 5.

Variation in wintering waterfowl densities explained by lake characteristics and/or assemblage types. Numbers of variables are the number of environmental variables included in the regression models. R² is the coefficient of determination, adjusted for degrees of freedom. P is the significance of the model using assemblage types only.

Figure 3.

Contributions of lake characteristic variables to regression models for individual waterfowl species. Scores are t-values for regression coefficients, averaged across species, separately for positive and negative effects.

Table 6.

Summaries of regression models for wintering waterfowl densities on lakes. All models include terms for assemblage types. The signs indicate the direction and significance of regression coefficients: (+)/(-), 0.15>P>0.05; +/-, P<0.05; ++/--, P<0.01; +++/---, P<0.001.

Waterfowl communities

Biological communities are composed of organisms which interact or are mutually dependent (Lincoln & Boxshall 1987). The extent to which wintering waterfowl recorded together on a lake constitute a community in this sense is arguable. Different species may have overlapping requirements for habitat and food, hence may be expected to compete, but interspecific interactions are rarely observable (Owen & Black 1990) and resource partitioning through behavioural or morphological adaptations may reduce direct competition (e.g. Lack 1945). However, spatial constraints on occupancy may apply even when there is no direct trophic dependence on a habitat, as in the case of waterfowl roosting on a lake, for which it may be simply a disturbance-free stretch of water. We are thus justified in considering patterns of variation in the assemblages of waterfowl using wetlands and in attempting to determine the ecological basis for such variation.

A potential pitfall in the interpretation of species' roles in identified assemblages, and hence in the interpretation of assemblages as biological communities, is that short-term movements may change species distributions. In the present analysis there is a danger that wintering waterfowl assemblages identified from day-time surveys will not adequately represent trophic dependence on habitat, since many waterfowl species, particularly diving ducks, are known to feed at night. Observations of pochard and tufted duck in the Cotswold Water Park suggest that day-time feeding is adequate to account for at least the day-time fraction of the daily energy budget in these species (R. May & M.C. Bell, unpubl. data), but it is still probable that there are night-time movements between lakes.

The analysis of wintering waterfowl assemblages in the Cotswold Water Park indicated that community structure depended on the ecological roles of the different species - largely a distinction between diving waterfowl assemblages, dependent on open water habitats, and dabbling waterfowl assemblages, more dependent on littoral habitats. A similar conclusion was drawn by Suter (1994) with respect to waterfowl wintering on Swiss lakes north of the Alps: species with similar food and feeding methods had similar distributions across the lakes. However, Suter (1994) found that distribution was more similar between species with similar food requirements but different feeding methods, such as coot and mute swan, than between species belonging to the same guild. In the present study, the distinction between diving and dabbling communities was very clear, although there were intermediate types.

The assemblages were also very distinct in terms of lake characteristics. Two main conclusions may be drawn from comparison of lakes between groups defined by assemblage types. Firstly, the findings support the notion that the scale of biotic diversity is reflected at all biotic levels within a lake. The poorest lakes for wintering birds were also the poorest for aquatic plants and benthic invertebrates. The converse appeared also to be true, although a diverse assemblage of dabbling waterfowl seemed to be a better indicator of generally high biotic diversity than a diverse assemblage of diving waterfowl. It is unclear, however, whether the link was causal, i.e. resulting from a dependence of birds on plants and invertebrates as a food resource, or systemic, i.e. reflecting a general effect of habitat diversity on all levels of biotic diversity. The modelling of lake characteristics important for individual species sheds more light on the existence of trophic links, at least for diving species.

The second conclusion concerns the link between the trophic status of lakes and the wintering waterfowl assemblage. Bell (1996) found a strong relationship between age and trophic status of lakes in the Cotswold Water Park, as measured by average Trophic Ranking Scores for open water plants, which implies a natural process of eutrophication occurring over a scale of a few decades. The eutrophication process appears to be driven by an interaction between the evolution of aquatic plant communities and the nutrient and base-status of the lake water. In terms of water chemistry, the trend is marked by an increased availability of dissolved phosphates. Lakes supporting the assemblage types in which diving waterfowl featured largely appeared to be at an earlier stage in the eutrophication process than those supporting assemblages dominated by dabbling species. These differences suggest that there is a relationship of wintering waterfowl community composition with a natural process of lake evolution towards increasingly eutrophic status, there being a parallel evolution from diving to dabbling assemblages. To a certain extent, this hypothesis is borne out by differences in age between lakes supporting different assemblage types: lakes supporting the diverse diving assemblage tended to be younger than all other lakes.

Several other workers have noted a relationship between the trophic status of lakes and wintering or breeding waterfowl (e.g. Rutschke 1987, Hoyer & Canfield 1994, Staicer, Freedman, Srivastava, Dowd, Kilgar, Hayden, Payne & Pollock 1994). Suter (1994) found that wintering diving duck in Switzerland tended to be associated with the less nutrient-enriched lakes, principally because their main food source, the zebra mussel Dreissena polymorpha, prefers less eutrophic waters; conversely, densities of fish-eating waterfowl were greatest on hypertrophic lakes, because of the abundance of cyprinid fish in these waters.

No effect of the recreational use of lakes on community structure was detected. That recreation can have profound effects on wintering waterfowl densities is apparent from this and other studies, both in the Cotswold Water Park (Fox et al. 1994) and more generally (Tuite et al. 1984). It appears that, at least in this study area, whilst water-based recreation can affect the numbers of birds using lakes (see below), it has little bearing on the types and relative proportions of different species, except inasmuch as recreation is correlated with habitat factors.

Wintering waterfowl and habitat

The diversity of waterfowl species and ecological types recorded on lakes in the Cotswold Water Park suggests that there is a corresponding diversity of waterfowl habitats. Inevitably, there must be overlaps in the ecological requirements of the different species such that the abundance of individual species should not be considered in isolation. The suitability of a lake for one species will be affected by the presence of other species, either because the species compete for the same resources or simply because there are spatial limits on the occupancy of different habitats. It can be shown that the reduction in the capacity of a unit of habitat to support a particular species caused by the presence of a second species may be modelled statistically in terms of interactions between the amount of habitat and the abundance of the second species.

We tried to account for the presence of other species by considering lake suitability for a species within groups of lakes defined by assemblage types. The proportions of the different species present should be relatively similar between lakes within assemblage types, so that the effects of interspecific interactions may be assumed to be effectively constant. A further justification of this approach is that the ecological requirements of a species may differ between assemblage types; a species may be present as a roosting member of one assemblage, for example, whilst a feeding member of another, thus presenting two different types of relationship with habitat factors.

It was clear from these analyses that a great deal of variation in numbers of individual species was explicable in terms of lake size and assemblage type. Wintering waterfowl assemblage type appeared to summarise a great deal about the nature of lakes, such that within groups of lakes supporting similar assemblages, lake size is a good correlate of the amount of suitable habitat for assemblage members. The success of models including assemblage type implies that the pattern of lake use by waterfowl stems from more than just the sum of individual habitat preferences; individual species may also have different ecoogical roles in different assemblages.

The importance of lake size was not unexpected; Suter (1994), for example, found relationships of wintering waterfowl numbers on Swiss lakes with either lake area or perimeter length, depending on waterfowl guild. These results were parallelled in the present study: in preliminary analyses we found that abundance was most strongly related with lake area in those species dependent on open water habitats (diving and fish-eating waterfowl) and with lake perimeter length in those species dependent on littoral habitats (dabbling and grazing waterfowl). However, even in some of the diving and fish-eating species, non-linearity in the relationship of numbers with area suggested that not all units of lake habitat are of equal value.

Over and above the effects of lake size and the lake characteristics summarised by assemblage types, it was clear that habitat factors defining the availability of food for wintering waterfowl were of great importance in determining numbers and that the human use or management of lakes was an important modifying factor. Water-based recreation appeared to be an important factor reducing the suitability of lakes, particularly for diving waterfowl. In some species, however, human management of lakes appears to be beneficial: management of lake shores for angling benefits the grazing species, particularly where the removal of marginal vegetation facilitates movements between banks and the open water.

It is unclear whether waterfowl using the Cotswold Water Park are in any sense resource limited. Effects of food supply and other habitat factors on waterfowl numbers do not necessarily imply that numbers are at a sustainable maximum on any given lake, merely that birds are distributed according to the availability of resources. Nevertheless, the results of the modeling presented in this paper are of value in identifying some major determinants of waterfowl numbers, which may be of use in directing management of lakes for waterfowl.

Implications for management of lakes for wintering waterfowl

A natural process of eutrophication accompanied by changes in the nature of wintering waterfowl assemblages was identified which has considerable implications for the maintenance of conservation interest in the Cotswold Water Park. In the absence of intervention to halt or divert this process, a lake which is of great conservation interest for the numbers of diving duck in winter is likely to evolve into one of lesser interest in the future, 20 to 30 years hence. Measures which are applied to protect this conservation interest may very well become misplaced. In the case of wintering waterfowl, it is unlikely that an important lake will become totally uninteresting in the future: any sizeable stretch of water in the region will support some waterfowl, if only for roosting, simply because there is a paucity of other open water in the region. It may well be the case, however, that conditions will no longer favour the concentrations of diving duck for which the Cotswold Water Park is renowned.

In the short term, shifts in conservation interest are unlikely to be a problem, since enough new gravel pits are excavated to ensure the availability of habitat at all stages of evolution. If the planned extension of gravel extraction in the upper Thames valley takes place (Gloucestershire County Council, Upper Thames Plan), there should be a supply of new habitat well into the future. To ensure maximum conservation benefit from this habitat creation requires a flexible conservation strategy, however, and a recognition by both conservationists and planners that conservation and the human-use of lakes in the Cotswold Water Park can coexist if planned with due regard to the likely changes and shifts in conservation interest. In the longer term, there needs to be research into methods of maintaining the conservation value within individual lakes. Habitat management may have a crucial role to play in ensuring that the Cotswold Water Park continues to be of outstanding importance for national and international nature conservation.