Direct and indirect effects of area, energy and habitat heterogeneity on breeding bird communities


  • Micael Jonsson,

    Corresponding author
    1. Department of Ecology and Environmental Science, Umeå University, 90187 Umeå, Sweden
    2. Department of Forest Ecology and Management, Swedish University of Agricultural Sciences, 90183 Umeå, Sweden
    Search for more papers by this author
  • Göran Englund,

    1. Department of Ecology and Environmental Science, Umeå University, 90187 Umeå, Sweden
    Search for more papers by this author
  • David A. Wardle

    1. Department of Forest Ecology and Management, Swedish University of Agricultural Sciences, 90183 Umeå, Sweden
    Search for more papers by this author

Micael Jonsson, Department of Ecology and Environmental Science, Umeå University, 90187 Umeå, Sweden.


Aim  To compare the ability of island biogeography theory, niche theory and species–energy theory to explain patterns of species richness and density for breeding bird communities across islands with contrasting characteristics.

Location  Thirty forested islands in two freshwater lakes in the boreal forest zone of northern Sweden (65°55′ N to 66°09′ N; 17°43′ E to 17°55′ E).

Methods  We performed bird censuses on 30 lake islands that have each previously been well characterized in terms of size, isolation, habitat heterogeneity (plant diversity and forest age), net primary productivity (NPP), and invertebrate prey abundance. To test the relative abilities of island biogeography theory, niche theory and species–energy theory to describe bird community patterns, we used both traditional statistical approaches (linear and multiple regressions) and structural equation modelling (SEM; in which both direct and indirect influences can be quantified).

Results  Using regression-based approaches, area and bird abundance were the two most important predictors of bird species richness. However, when the data were analysed by SEM, area was not found to exert a direct effect on bird species richness. Instead, terrestrial prey abundance was the strongest predictor of bird abundance, and bird abundance in combination with NPP was the best predictor of bird species richness. Area was only of indirect importance through its positive effect on terrestrial prey abundance, but habitat heterogeneity and spatial subsidies (emerging aquatic insects) also showed important indirect influences. Thus, our results provided the strongest support for species–energy theory.

Main conclusions  Our results suggest that, by using statistical approaches that allow for analyses of both direct and indirect influences, a seemingly direct influence of area on species richness can be explained by greater energy availability on larger islands. As such, animal community patterns that seem to be in line with island biogeography theory may be primarily driven by energy availability. Our results also point to the need to consider several aspects of habitat quality (e.g. heterogeneity, NPP, prey availability and spatial subsidies) for successful management of breeding bird diversity at local spatial scales and in fragmented or insular habitats.


Island biogeography theory (IBT) (MacArthur & Wilson, 1963, 1967), which predicts that area and isolation are drivers of species richness, has been one of the most influential theories for explaining species richness patterns in island systems and other insular communities. According to IBT, larger insular areas (i.e. islands or habitat patches) have higher species richness because they gain more species through immigration and hold larger populations that are less prone to stochastic extinction. This theory makes no predictions regarding density, but density may decrease with increasing species richness (resulting from increasing area or decreasing isolation) due to higher numbers of competitors, predators, or large species, or to a less stable environment (i.e. ‘density compensation hypothesis’, MacArthur et al., 1972). Conversely, density may also increase with increasing species richness due to increased utilization of niche space (Lehman & Tilman, 2000). Over the past half century many studies have found support for IBT (Connor & McCoy, 1979; Rosenzweig, 1995), but several have also reported richness patterns that are inconsistent with IBT (e.g. Connor et al., 2000; Lomolino & Weiser, 2001; Barrett et al., 2003; Wardle et al., 2003a; Jonsson et al., 2009). Consequently, the general applicability of IBT to island systems and other systems with island-like characteristics has frequently been questioned (Brown & Lomolino, 2000; Lomolino, 2000; Brotons et al., 2003).

Several alternative, but rather complementary, theories to explain patterns of species richness have been proposed. The species–energy theory, an extension of IBT, predicts that species richness should reflect the total amount of energy available (Wright, 1983). Many studies have found support for species–energy theory but others have not, and as several processes may be involved in determining species richness, it is still unclear as to what the underlying mechanisms behind the species–energy relationship may be (Hurlbert, 2004; Evans et al., 2005, 2006). In species–energy studies on bird communities, net primary productivity (NPP) or other measures of (plant) productivity have often been used as a surrogate for energy availability. However, while plant productivity represents the amount of energy input to the ecosystem, it is not always directly relevant to the amount of energy available for the study organism, particularly in the case of secondary or higher consumers. Hence, it is not well known how more precise measures of energy availability (i.e. availability of food for consumers) are related to animal community patterns (but see Hurlbert, 2006), or how important energy availability is relative to the influence of area, isolation and other factors that may drive species richness. Nevertheless, some studies have provided evidence that the influence of area on patterns of species richness can be modified by energy availability (e.g. Wylie & Currie, 1993; Storch et al., 2005; Hurlbert, 2006; Kalmar & Currie, 2006).

Another theory that has potential for explaining species richness patterns on islands is niche theory (e.g. Chesson & Kuang, 2008). Based on this theory it is predicted that more species can coexist on more heterogeneous islands, as such islands offer more opportunity for niche differentiation (MacArthur, 1969; Tilman, 2004; Chesson & Kuang, 2008). Although some studies relating measures of habitat heterogeneity to species richness have found support for this prediction (Southwood et al., 1979; Siemann et al., 1998; Ricklefs & Lovette, 1999; Jeffries et al., 2006), others show that habitat heterogeneity by itself may be a relatively poor predictor of richness (Ricklefs & Lovette, 1999; Triantis et al., 2003, 2008; Panitsa et al., 2006; Russell et al., 2006).

In sum, the three theories make predictions about effects of habitat area, isolation, energy availability and habitat heterogeneity. In this study, we sought to determine their relative effectiveness in predicting bird community patterns across 30 islands in a well-characterized lake-island system for which area-dependent differences in fire regime among islands have created gradients in plant diversity, forest age and prey densities that are negatively related to both NPP (per unit area) and island size (Wardle et al., 1997, 2008; Crutsinger et al., 2008; Jonsson & Wardle, 2009; Jonsson et al., 2009). In this system, we collected data on the abundance and species richness of insectivorous birds on all 30 islands, and related these to measures of area (island size), isolation, habitat heterogeneity (forest age and plant diversity), and energy availability (NPP and invertebrate prey). We used these data to test which of the three theories best predicted breeding bird community patterns on these islands, so as to yield insights about the mechanisms responsible for determining patterns of biological diversity across island ecosystems. A general problem when evaluating the relative importance of these factors is that they affect community structure through a number of different pathways. As an example, island area may affect richness through its effects on colonization rates, via its effects on energy availability and abundance, or via its effects on habitat diversity. A consequence of these problems is that common statistical approaches, such as multiple linear regression, may be uninformative. As such, we compared analyses based on multiple regression models with those based on structural equation models because the latter quantify the strength of both direct and indirect relationships.

As all three theories entail the influence of area on species richness, a significant effect of area does not in itself distinguish between the theories. First, to test for IBT, we investigated if isolation explained any patterns in bird species richness in addition to that explained by area. Further, as we have no reason to believe that the number of predators, competitors or large species, or environmental stability vary with species richness in our study system (Jonsson & Wardle, 2009; Jonsson et al., 2009), we predicted that bird density should be positively correlated with area as a consequence of increased utilization of niche space on large and species-rich islands (Lehman & Tilman, 2000). Second, to test for species–energy theory, we evaluated whether bird abundance increases with increasing total energy availability, as measured by either NPP or prey abundance, and whether this increase in bird abundance in turn leads to increased species richness. Further, we predicted that bird density should be positively related to per unit area energy availability. Third, to test for niche theory, we predicted that a positive relationship between species richness and habitat heterogeneity would be present, as measured by both plant diversity and forest age. No clear-cut prediction could be derived for bird density because increased heterogeneity could lead to either more high-quality habitats or more low-quality habitats.

Materials and methods

Study system and independent variables

Bird censuses were performed on each of 30 islands in two freshwater lakes (Lakes Hornavan and Uddjaure) in the boreal forest zone of northern Sweden (65°55′ N to 66°09′ N; 17°43′ E to 17°55′ E). These islands differ in size (from 0.02 to 15.0 ha) and fire disturbance history (Wardle et al., 1997). Of these islands, 10 are ‘large’ (>1.0 ha), 10 are ‘medium’ (0.1 to 1.0 ha) and 10 are ‘small’ (<0.1 ha) (Wardle et al., 2003b). Large islands are more frequently struck by lightning and have therefore burned more frequently than have the small islands (Wardle et al., 1997, 2003b). This variation in disturbance history across the island-size gradient results in the islands being at different successional stages – large islands are at an early successional stage, while small islands are at a much later stage. The gradient in successional stages among these islands is well characterized, with the small islands having lower productivity, lower decomposition rates, deeper humus, less available soil nutrients, lower microbial biomass, and higher plant species richness compared to the large islands (Wardle et al., 1997, 2003b, 2008).

For this study, we used vascular plant species richness and abundance data for each island (previously recorded in June 2006; Wardle et al., 2008) to calculate plant diversity (Shannon–Wiener index) and used this as a measure of habitat heterogeneity. Across the islands, vascular plant species richness measured in 10-m radius plots ranges from 4 to 15 species. Three tree species (Betula pubescens, Picea abies and Pinus sylvestris) and three shrub species (Empetrum hermaphroditum, Vaccinium myrtilus and Vaccinium vitis-idaea) are the most common vascular plant species on the islands. The trees dominate in terms of biomass, and are therefore the most important components of the vegetative structural complexity (i.e. habitat heterogeneity). On large islands, Pinus sylvestris makes up on average 79.3% of the total vascular plant biomass, while Betula pubescens represents 11.0%, and Picea abies 6.0%, leading to low spatial heterogeneity in the vegetation structure. The species composition is more even both on small islands, with Pinus sylvestris at 12.6%, Betula pubescens at 19.5%, and Picea abies at 58.3%, and on medium islands, with Pinus sylvestris at 42.5%, Betula pubescens at 27.0% and Picea abies at 25.5% (Wardle et al., 1997). As these tree species differ significantly in their physical structure, and as their species composition change markedly and predictably with island size, there is also an important difference in vegetation structure complexity, and therefore habitat heterogeneity, across the island-size gradient.

The use of vegetation diversity as a measure of habitat heterogeneity is consistent with earlier investigations in which plant community characteristics (e.g. foliage height diversity, plant species richness) have been used as surrogates for habitat heterogeneity of consumer organisms (e.g. MacArthur & MacArthur, 1961; Siemann et al., 1998; Ricklefs & Lovette, 1999; Haddad et al., 2001). However, as several forest characteristics may be involved in creating habitat heterogeneity when forests grow older, we also used forest age (i.e. time since the last major fire on the island using data from Wardle et al., 2008) as an alternative measure of habitat heterogeneity, consistent with Jeffries et al. (2006). In our study system, both plant diversity and forest age were found to be negatively correlated with island size (r2 = 0.258, P = 0.004 and r2 = 0.534, < 0.001, respectively). These negative relationships provide opportunities for distinguishing between the effects of area and proxies of habitat heterogeneity that in most other island systems are positively correlated with each other.

The densities of spiders and beetles also vary across the island-size gradient, with small islands showing significantly higher densities than large islands, most likely due to the negative relationship between island size and plant diversity (Crutsinger et al., 2008; Jonsson et al., 2009). In earlier work, where terrestrial invertebrates were caught using pitfall trapping and sweep netting of vegetation, Jonsson et al. (2009) found a negative relationship between the combined spider and beetle densities and island size (r2 = 0.463, < 0.001). Similarly, the density of emerged aquatic insects (mainly dipterans and trichopterans) found on the islands show a negative relationship with island size (r2 = 0.401, < 0.001), and are positively correlated with terrestrial invertebrates on the islands (Jonsson & Wardle, 2009). The birds assessed in the present study are obligatory or partial insectivores, and spiders, beetles and emerged aquatic insects (i.e. arthropods) arguably represent the most important prey source for these birds. The other main terrestrial invertebrate group, i.e. wood ants, are well defended and are competitors rather than food for the studied bird species (Haemig, 1992). Other terrestrial invertebrates, such as earthworms and caterpillars, are of very low density in our study system (Jonsson & Wardle, 2009) and therefore of negligible importance as a food source. For this study, we used measures of combined spider and beetle density from Jonsson et al. (2009) and emerged aquatic insects density (Jonsson & Wardle, 2009; M. Jonsson & D.A. Wardle, unpublished data) as two measures of food (i.e. energy) availability to the birds. Terrestrial invertebrates were collected in early July of 2006 and 2007, and aquatic insects were collected in early August of 2008 and 2009. Further, to determine the energy input to each island, we used measurements of net primary productivity (NPP; kg m−2 year−1) performed on each island by Wardle et al. (2003b). This measure is based on growth data using tree ring widths and allometric relationships for the tree species, and measures of shoot biomass productivity for the understorey species (Wardle et al., 2003b). This measure of NPP is positively related to island size (r2 = 0.384, < 0.001).

The islands also differ in distance to mainland (i.e. isolation) from 0.3 to 3.3 km. However, as large islands may have similar effects as the mainland on nearby small islands, we derived another measure of island isolation that took all nearby landmasses into account, and not just the mainland (Jonsson et al., 2009). This index of isolation, calculated separately for each island, is based on the total area around the entire island between the shoreline and 500 m from the shoreline, and the proportion of land in that area. It is determined as 1 − (a/(b – a)) where a is the total landmass in the surrounding area and b is the total surrounding area (including both land and water). This type of isolation is known to influence bird community patterns, especially when birds are reluctant to travel relatively large distances through open habitats (e.g. Bélisle & Desrochers, 2002). Across the 30 islands, this measure of isolation was independent of island size (r2 = 0.012, P = 0.566).

Bird censuses

A census of obligatory and partly insectivorous bird species was performed during a period of 1 week in the middle of both June 2006 and June 2007. At this time of the year, most birds have arrived in northern Sweden and males are claiming their territories, but couples are only in the initial phases of nesting. The census was carried out by walking the length of an island in a zigzag pattern (<50 m between each individual transect) in order also to cover the width of the island, and recording all birds that were encountered. As the distance between transects was <50 m, it is likely that all foraging, singing, alerting and calling birds were detected. Occasional stops were made only to ensure that an individual was not counted more than once. The time spent counting and the total transect length varied with island size, but never exceeded 60 min; this was, however, sufficient to record all the potentially breeding birds on each island. The counts were used to assess bird density and bird species richness. The bird data, as with the invertebrate data, were then averaged over 2 years to obtain one average value for each island.

Data analysis

Linear regressions were used to explore univariate relationships between measures of the bird community and the following independent variables: island size, island isolation, terrestrial prey availability, emerged aquatic prey availability, NPP, forest age and plant diversity. For predicting bird density, we used measures of energy availability calculated on a per unit land-area basis, while for predicting bird species richness we used measures of total energy availability calculated on a whole island basis. As a bird density measure we used the number of individuals per unit area because this was the quantity that predictions were derived for. We recognize that the investigation of relationships between variables that share a common term (e.g., ‘area’ in both island size and bird density) has been the subject of debate, with some authors criticizing this approach for being ‘spurious’ (e.g. Brett, 2004), and with others providing arguments based on mathematics and logic to support such analyses (e.g. Prairie & Bird, 1989; Peters, 1991). However, consistent with the recommendations of Brett (2004), we use this analysis merely to test our results against an appropriate null model, i.e. that the number of individuals per unit area is independent of island area.

Multiple regression analyses were then performed to test hypotheses derived from each of island biogeography theory (IBT), species–energy theory and niche theory. All of the theories predict effects of area, the only exception being effects on density under niche theory. We therefore chose to test if other mechanisms proposed by each theory explained any variation in addition to that of area alone. Therefore, island isolation was included in the model testing IBT, total NPP and total terrestrial and aquatic invertebrate prey abundance in the models testing species–energy theory, and plant diversity and forest age in the models testing niche theory. To test for a mechanism behind energy effects on species richness, we also included total bird abundance as a predictor in models testing species–energy theory. Further, to test our predictions of drivers of bird density in relation to IBT, we included bird species richness as a predictor (in line with the ‘density compensation hypothesis’, MacArthur et al., 1972). We tested all combinations of hypothesized predictors in the multiple regression analyses, with the restriction that area was included in all models. Akaike information criterion (AIC) values were used for model selection.

Further, the whole data set was structured into a path-analysis network, including all plausible pathways (Fig. 1), and analysed in a structural equation model (SEM; Malaeb et al., 2000; Grace et al., 2007). In these analyses, variables were removed in a stepwise fashion, guided by AIC values, to obtain the most parsimonious model (the one with the lowest AIC value). As above, per unit area energy availability was used for the bird density model, while total energy availability was used for the bird species richness model. Several tests were used to assess model fit, i.e. the chi-square test, comparative fit index (CFI), root square mean error of approximation (RMSEA), and standardized mean square error of approximation (SRMR). For the SEM, Mplus software version 5.2 (Muthén & Muthén, 1998–2007) was used. For the single and multiple regression analyses, PASW Statistics 18.0 (© SPSS Inc., 2009, Chicago, IL, USA) was used. Data were log transformed if necessary to meet the requirements of parametric data analyses.

Figure 1.

 Illustration of all plausible interaction paths in the study system of 30 Swedish islands for (a) the analysis predicting bird species richness using total (per-island) energy measurements [i.e. terrestrial and aquatic prey, and net primary productivity (NPP)] (see Fig. 3a), and (b) the analysis predicting bird density using per unit area energy measurements (see Fig. 3b).


In total, 21 bird species were recorded in the study system, and species richness ranged from 0 to 10 species across the islands. None of the 21 species dominated in terms of abundance; four species each represented between 10.2 and 17.4% of the total abundance, five species each represented 4.5 to 8.1% of the total abundance, and then there was a tail of 12 species that represented 2.9% or less of the total abundance (see Appendix S1 in the Supporting Information).

Island biogeography theory

Univariate analyses showed that bird species richness was positively correlated with island size (Fig. 2a), while density was negatively related to island size (Fig. 2g). No significant relationship was found for island isolation (species richness: r2 = 0.001, P = 0.999; density: r2 = 0.158, P = 0.406). The multiple regression analysis showed that island isolation failed to explain any variation in bird species richness beyond what was explained by island size alone (Table 1). Thus, as area is an important component of all three theories, we did not find any specific support for IBT. Our prediction that bird density would be greater on larger and more species-rich islands was only partially supported, because the best model included a negative influence of area and a positive influence of bird species richness (Table 2).

Figure 2.

 Relationships of island size (a, g), plant diversity (b, h), forest age (c, i), net primary productivity (NPP) (d, j), terrestrial prey abundance (e, k), and aquatic prey abundance (f, l) with bird species richness and bird density on 30 Swedish islands. The bird and invertebrate data are averaged over 2 years (2006 and 2007 for birds and terrestrial invertebrates, and 2008 and 2009 for aquatic insects). Total (per-island) energy measurements (i.e. terrestrial and aquatic prey abundance, and NPP; all log-transformed) are used for bird species richness, while per unit area energy measurements are used for bird density. r2 and P values are presented for each regression (n = 30 for each regression). Linear relationships are shown where < 0.10.

Table 1.   Results of multiple regression analyses testing the ability of island biogeography theory (IBT), species–energy theory (SET), and niche theory (NT) to explain patterns in bird species richness, using bird and invertebrate data averaged over 2 years (2006 and 2007 for birds and terrestrial invertebrates, and 2008 and 2009 for aquatic insects) for 30 Swedish islands. Island size (area) is used as a null model that all other models are tested against. Island isolation is included to test IBT; total net primary productivity (NPPtot), terrestrial prey abundance (terrpreytot), aquatic prey abundance (aquapreytot) and bird abundance (birdab) are included to test SET; and plant diversity and forest age are included to test NT (n = 30 islands). The bird, NPP and prey values are on a per-island basis. The sign (‘+’) indicates the direction of significant (< 0.05) relationships between dependent and independent variables. The models are listed in order of decreasing Akaike information criterion (AIC) values, where bold values indicate that the model being tested is better then the null model.
RichnessIndependent variablesAICR2P
 IBT, SET, NTarea (+)4.940.820<0.001
 IBTarea (+); isolation5.690.830<0.001
 SETarea; aquapreytot; NPPtot10.080.820<0.001
area; NPPtot7.420.820<0.001
area (+); aquapreytot7.400.820<0.001
area; terrpreytot (+); aquapreytot; NPPtot6.960.853<0.001
area; terrpreytot (+); NPPtot4.120.852<0.001
area (+); terrpreytot (+); aquapreytot4.070.853<0.001
area (+); terrpreytot (+)1.450.852<0.001
area; birdab (+); aquapreytot; terrpreytot; NPPtot7.980.919<0.001
area; birdab (+); aquapreytot; NPPtot10.410.917<0.001
area (+); birdab (+); aquapreytot; terrpreytot10.550.918<0.001
area; birdab (+); terrpreytot; NPPtot11.050.919<0.001
area (+); birdab (+); aquapreytot12.580.915<0.001
area; birdab (+); NPPtot13.280.917<0.001
area (+); birdab (+); terrpreytot13.380.918<0.001
area (+); birdab (+)15.250.915<0.001
 NTarea (+); forest age; plant diversity9.740.822<0.001
area (+); plant diversity7.280.821<0.001
area (+); forest age7.120.822<0.001
Table 2.   Results of multiple regression analyses testing how variables related to island biogeography theory (IBT) and species–energy theory (SET) are able to explain patterns in bird density, using bird and invertebrate data averaged over 2 years (2006 and 2007 for birds and terrestrial invertebrates, and 2008 and 2009 for aquatic insects) for 30 Swedish islands. Island size (area) is used as a null model that all other models are tested against. Island isolation and bird species richness (richness) are included to test IBT; and net primary productivity (NPP), terrestrial prey density (terrprey) and aquatic prey density (aquaprey) are included to test SET (n = 30 islands). The bird, NPP, and prey values are on a per unit area basis. No predictions were made for niche theory. Signs (‘+’ or ‘−’) indicate the direction of significant (P < 0.05) relationships between dependent and independent variables. The models are listed in order of decreasing Akaike information criterion (AIC) values, where bold values indicate that the model being tested is better then the null model.
DensityIndependent variablesAICR2P
 IBT, SETarea (−)−59.440.403<0.001
 IBTarea (−); isolation−57.360.4110.001
area (−); isolation; richness (+)−58.590.517<0.001
area (−); richness (+)61.270.517<0.001
 SETarea (−); aquaprey; NPP−54.620.4100.003
area; terrprey (+); aquaprey; NPP−56.610.4990.001
area (−); aquaprey−57.050.4050.001
area (−); NPP−57.200.4080.001
area; terrprey; NPP−58.600.4830.001
area (−); terrprey (+); aquaprey−59.110.492<0.001
area; terrprey−60.980.478<0.001

Species–energy theory

In univariate regressions, bird species richness was significantly positively related to total NPP (Fig. 2d), total terrestrial prey abundance (Fig. 2e), and total aquatic prey abundance (Fig. 2f). Also, bird species richness and bird abundance were positively correlated (r2 = 0.808, < 0.001). Bird density was significantly negatively correlated with per unit area NPP (Fig. 2j), but positively correlated with both terrestrial and aquatic prey density (Fig. 2k,l, respectively). The best multiple regression model for predicting species richness was one that included only island size and bird abundance (Table 1). For bird density, the best model included island size and terrestrial prey density (Table 2). Thus, as terrestrial prey density showed a positive influence on bird density, and as bird abundance had a positive influence on bird species richness, there was some support for species–energy theory.

Niche theory

Bird species richness showed negative relationships with both plant diversity (Fig. 2b) and forest age (Fig. 2c). The multiple regression analyses testing the effects of plant diversity and forest age (both used as measures of habitat complexity) on bird species richness showed that none of the investigated models produced lower AIC values than the model with island size alone (Table 1). Thus, we found no support for niche theory. We did not make any predictions on bird density patterns in relation to niche theory, but bird density was unrelated to plant diversity (Fig. 2h) and positively related to forest age (Fig. 2i).

Structural equation modelling

The most parsimonious SEM predicting bird species richness (i.e. the one with the lowest AIC value) included total NPP, bird abundance, island area, forest age and terrestrial prey abundance (Fig. 3a). Only total NPP and bird abundance had direct effects on bird species richness, whereas forest age had a weak, but significant, negative influence on bird abundance. Also, forest age and area had positive influences on terrestrial prey abundance that, in turn, showed a positive influence on bird abundance. Isolation and plant diversity were not present in the most parsimonious model, nor was the influence of bird species richness on bird abundance. This SEM showed a strong fit according to all test statistics used (χ2 = 2.719, P = 0.994; CFI = 1.000; RMSEA < 0.001; SRMR = 0.008).

Figure 3.

 Results from structural equation models for (a) bird species richness and (b) bird density on 30 Swedish islands. The bird and invertebrate data are averaged over 2 years (2006 and 2007 for birds and terrestrial invertebrates, and 2008 and 2009 for aquatic insects). Total (per-island) energy availability (log-transformed) is used in the bird species richness model, while per unit area measurements of energy availability [i.e. terrestrial and aquatic prey abundance, and net primary productivity (NPP)] are used in the bird density model. Bold lines indicate significance at P = 0.05 (thick bold lines indicate < 0.01). Dashed lines illustrate non-significant paths that were kept in order to obtain the most parsimonious model. The standardized coefficients are given for each arrow.

The most parsimonious structural equation model (SEM) for bird density included island area, forest age and per unit area energy availability (terrestrial and aquatic prey density and NPP), whereas plant diversity, bird species richness and isolation were excluded (Fig. 3b). Terrestrial prey density showed a direct positive influence on bird density, while island area showed a direct negative influence, and forest age and aquatic prey density showed indirect influences via their positive effects on terrestrial prey density. The influence of bird species richness on bird density that was found in the multiple regression analyses did not explain enough variance to be included in the most parsimonious SEM. Therefore, these results are in line with the prediction by species–energy theory. This SEM showed a strong fit for all test statistics used (χ2 = 2.979, P = 0.812; CFI = 1.000; RMSEA < 0.001; SRMR = 0.033). For both SEMs, a significant amount of variance was explained for each dependent variable, although a greater amount of variance was explained for bird species richness than for bird density (Fig. 3a,b).


The positive relationship between species richness and habitat area is one of the strongest generalizations in community ecology (MacArthur & Wilson, 1963; Rosenzweig, 1995). In the present study, we sought to address whether this relationship reflects direct effects of area on richness or indirect effects mediated by other factors. Using structural equation models (SEM) (Fig. 3a), we were able to show that the influence of area was best explained as indirect effects mediated by energy availability, measured either as availability of terrestrial prey or as net primary productivity. We also observed significant influences of habitat heterogeneity (indicated by forest age) through two separate pathways (Fig. 3a). However, these two indirect effects were comparatively weak and operated in different directions, suggesting that the net effect of habitat heterogeneity on species richness is likely to be negligible. Our results therefore strongly support the view that available energy is an important, and probably a primary, determinant of community patterns in our study system (cf. Wright, 1983; Storch et al., 2005; Hurlbert, 2006; Kalmar & Currie, 2006). Our results also demonstrate that SEM can provide important insights beyond what can be achieved with traditional regression approaches. For example, in contrast to the SEM, the best multiple regression model included bird abundance but no direct measures of energy availability as predictors, thus providing, at best, weak evidence for the role of energy availability. Moreover, the regression analysis was unable to tease apart the different pathways along which area and forest age affect species richness.

The SEM revealed that two different measures of total energy availability (terrestrial prey abundance and total NPP) influenced bird species richness independently of one another (Fig. 3a). While the effect of terrestrial prey abundance on species richness was mediated via bird abundance, NPP also exerted a direct effect on richness. It is unclear as to the precise mechanism through which NPP might exert such an effect on bird species richness. However, some of the bird species in our study system (i.e. short-distance migrants) may be partially affected by NPP independently of invertebrates because they are also able to feed on plant material (e.g. buds and seeds) at some times of the year, especially when invertebrate abundance is low (i.e. Mönkkönen et al., 2006). Increased NPP could be potentially linked to an increased supply and heterogeneity of this material, in turn promoting bird species richness. Another possibility is that while total invertebrate prey abundance is unaffected by NPP (Fig. 3a), the species composition and quality of this prey may be responsive to NPP (Southwood et al., 1979; Haddad et al., 2001), in turn influencing bird species richness.

Several mechanisms that involve random sampling and/or local extinction processes have been proposed as explanations for positive relationships between energy availability and species richness (see Honkanen et al., 2010, for a comprehensive review). Local extinction processes should not be important, as the breeding birds in our study system are either long- or short-distance migrants. It is therefore possible that between-island patterns are shaped by habitat selection of arriving birds. However, we expect that the bird numbers that arrive to breed on an island are largely independent of the output of fledglings from that island the previous year (although some homing may exist). While this suggests that only random sampling processes are important, it is still likely that, in a longer-term perspective, variation in mortality rate and reproductive success between different types of islands shapes the evolution of habitat selection behaviours, so that local density matches habitat quality (Fretwell & Lucas, 1970). Hence, even if the random sampling mechanism can yield a positive relationship between species richness and energy availability (i.e. the ‘energy-based sampling hypothesis’; Evans et al., 2005; Carnicer et al., 2008a; Honkanen et al., 2010), it seems probable that some degree of active choice is also involved, given the strong flight ability of the birds. Thus, one possible explanation for the observed patterns is a variation of the niche position hypothesis (Abrams, 1995), in which it is predicted that elevated levels of energy increase the number of rare resources that are productive enough to attract breeding pairs of specialist species. In line with this reasoning, recent studies have shown that components of the variation in species richness can be explained by niche-based processes, such as habitat selection and dispersal behaviour, and that the sampling process therefore is not completely random but depends on species-specific adaptive traits (Carnicer et al., 2007, 2008a). If both dispersal sampling and niche-based processes are important, it is expected that communities should exhibit a high degree of nestedness (Carnicer et al., 2008a). The fact that the bird communities in our study system were significantly nested (Appendix S2) supports this prediction.

In the multiple regression analyses bird density was driven by area, terrestrial prey density and bird species richness, but in contrast to our prediction there was a negative relationship between bird density and area. One possible explanation is based on recent theory about density–area relationships in systems for which densities are controlled by migration (Hambäck & Englund, 2005). If immigration rate is proportional to the width (or height) of the island (which is expected if it is detected by migrating organisms from some distance; Englund & Hambäck, 2007), and if emigration rate is independent of area, then the expected slope of the relationship between log(area) and log(density) is −0.5, which is close to our observed value of −0.36. An alternative hypothesis, which is more in agreement with the idea of active habitat choice, is that the area–density relationship reflects resource levels. For example, if aquatic subsidies drive terrestrial prey density, and this in turn drives bird habitat selection, then a slope close to −0.5 would be expected for bird density. Indeed we previously found that the slope for terrestrial prey density is −0.34 (Jonsson et al., 2009). The same direct influences as those found in the multiple regression analyses were found in the SEM on bird density, with the exception that bird species richness showed no significant influence. In addition, the SEM showed indirect influences of aquatic prey density and forest age on bird density, via effects on terrestrial prey density (Fig. 3b).

As both habitat heterogeneity and productivity are positively correlated with area in most studies, it is frequently difficult to quantify the relative importance of each factor when studying influences of habitat heterogeneity and/or energy availability on community structure (Ricklefs & Lovette, 1999; Evans et al., 2005; but see Hurlbert, 2004, 2006; Storch et al., 2005). However, unlike previous studies, we were able to effectively separate effects of energy availability (in terms of prey availability) and habitat heterogeneity (i.e. plant diversity or forest age) from effects of island size, because prey density and habitat heterogeneity were not positively correlated with area. Further, although island size and NPP (as a measure of energy input per unit area to the ecosystem; Sala & Austin, 2000) were positively correlated, total NPP and not island size was most effective in explaining variance in bird species richness according to our analyses. In fact, our results showing that NPP and bird abundance together were of direct importance for bird species richness (Fig. 3a), while area was not, clearly support the view that increasing abundance in response to higher energy availability is a mechanism behind higher species richness in high-energy environments, consistent with the ‘energy-based sampling hypothesis’. Our results also suggest that if energy-availability data along an island-area gradient are lacking, it is possible to incorrectly attribute subsequent changes in animal community structure to the influence of area per se.

Although most studies exploring the species–energy relationship have used various measures of productivity to assess energy availability, these may not necessarily serve as useful proxies of the amount of energy available for the species under consideration (Hurlbert, 2006), especially in the case of higher-level consumers such as insectivorous birds. In circumstances for which coarse measures of productivity (e.g. NPP) are tightly linked with food availability (as certainly occurs in some cases; see Kaspari et al., 2000), they could indeed serve as useful predictors of the amount of energy that can be converted into consumer biomass. While we found that total NPP could predict some across-island variability in bird species richness, we also found that invertebrate prey abundance could explain some variability that NPP could not (Fig. 3a). This is likely to be due to the negative relationship between per unit area NPP and prey density in our study system, which clearly suggests that NPP is not the most important driver of the terrestrial invertebrate prey community in our system (Crutsinger et al., 2008; Jonsson et al., 2009). Instead, the results show that the terrestrial prey organisms in our system, many of which are themselves predators, are highly dependent on specific aquatic-derived prey sources (i.e. in the form of emerging insects; Jonsson & Wardle, 2009) that can be converted into biomass (cf. Hurlbert, 2006). Therefore, although our results show an importance of NPP, they also suggest that disregarding other, more direct measures of energy availability may lead to weaker mechanistic explanations of patterns in species richness. As such, these results are in line with recent research showing that direct measurements of available resources along resource channels, rather than the use of proxies, can be crucial for understanding determinants of species richness and food-web structure (Hurlbert, 2006; Carnicer et al., 2008b, 2009; Rooney et al., 2008).

In summary, our results show that energy availability is an important determinant of breeding bird species richness, most likely because high energy availability supports more individuals that, in turn, promote high species richness, consistent with the ‘energy-based sampling hypothesis’. Moreover, our results suggest that positive species–area relationships, which historically have been reported from a large number (and wide variety) of studies, may be explained by energy availability rather than by independent effects of area, and that it may be necessary also to consider measures of energy availability that are more direct than those provided by NPP proxies when attempting to understand factors that drive community-level attributes across islands (i.e. Hurlbert, 2006). In addition to the observed importance of energy availability, we found that habitat heterogeneity (i.e. forest age) indirectly influenced the bird communities through its effect on prey availability (i.e. Jeffries et al., 2006), and that these prey were also influenced by spatial (aquatic) subsidies (e.g. Jonsson & Wardle, 2009). Hence, our results are also relevant for management and conservation practices of systems that are subjected to habitat modifications (e.g. Simberloff, 1988; McCarthy et al., 2005) because they suggest that NPP, food resource availability, habitat heterogeneity and the input of energy from adjacent systems should simultaneously be considered in order to optimize management of local biodiversity, particularly in fragmented and insular habitats.


We thank Jofre Carnicer and an anonymous referee for helpful comments. Financial support was provided by The Swedish Research Council for Environment, Agricultural Sciences, and Spatial Planning (FORMAS).


Micael Jonsson investigates the importance of species diversity and composition for the functioning of terrestrial and freshwater systems, the importance of spatial subsidies for animal community structure and dynamics, and drivers of animal community patterns.

Göran Englund studies the role of heterogeneity and spatial scale in community and population ecology. His recent work concerns climatic effects on the distribution of freshwater fish, area–density relationships, and effects of spatial heterogeneity on predator–prey dynamics.

David A. Wardle works on the functioning of terrestrial ecosystems with emphasis on the feedbacks between the above-ground and below-ground subsystems. Much of his current work involves field projects in New Zealand rain forests and northern Sweden boreal forests.

Editor: Lisa Manne