- Top of page
Although the interplay between productivity and trophic cascades has been the subject of recent reviews (Brett & Goldman 1996; Schmitz, Hambäck & Beckerman 2000; Halaj & Wise 2001; Shurin et al. 2002), determination of the amount of resources on which a community is based has been elusive. The amount of resources may differ among systems because of productivity differences, but also because of areal changes (Schoener 1989) and external inputs (Polis, Anderson & Holt 1997). Field studies that do not simultaneously consider these three sources of variation are potentially ignoring a main determinant of energy available. Few are the systems that allow us to compare the effect of productivity without there being a large variation in other environmental or community attributes (Vander Zanden et al. 1999; Post et al. 2000). This could explain the contradictory results or the lack of significant patterns in the review literature.
The functional relationship between species richness in the community and the number of links per species is the subject of debate (e.g. Martinez 1992; Winemiller et al. 2001; Schmid-Araya et al. 2002). Species richness can have a positive or humped relationship with productivity, depending on the spatial scale of observation (Chase & Leibold 2002). If species richness changes with productivity and if links per species are related to richness, an association is expected between productivity and the number of links per species. However, attempts to reveal this association have not detected a significant pattern (Townsend et al. 1998).
Since 1987, a long-term ecological research project has been conducted at Aucó, central Chile. At this site, precipitation is positively associated with perennial and ephemeral herb cover, seed bank density, as well as with insect, bird and mammal abundance (see reviews by Jaksic 2001; Meserve et al. 2003). Thus, the study area is characterized by large interannual variations in primary productivity and in the total number of interacting species.
The objective of our study is to analyse the variation in prey link density among top predator species, in response to variation in primary productivity. Although we do not have data on the whole food web, top predator species represent a relevant component of communities (Oksanen & Oksanen 2000) and theories about food web connectance do not discriminate between trophic groups (Williams & Martinez 2000; Schmid-Araya et al. 2002). In addition, there is a scarcity of data on food webs at different times, or in comparable environments but with different productivity (Dunne et al. 2002a). Thus, the analysis of trophic relations of a community subset is a promising approach to understanding the connection between energy supply and food web structure (Holt & Polis 1997; Winemiller et al. 2001). At variance with previous comparisons between environments, the system studied here does not show large variation in community composition or loss of energy sources, thus reflecting the fine-tuning that may occur in natural systems over time. We analyse separately the association between productivity and (i) mean number of trophic connections among all predators and (ii) prey richness for each predator species.
- Top of page
The study site is located at Las Chinchillas National Reserve near Aucó (31°30′ S, 71°06′ W; elevations range from 400 m to 1700 m), 300 km north of Santiago, Chile. The geographical and biological characteristics of the Reserve have been detailed elsewhere (Jaksic et al. 1993). From 1987 to the present, the diets of the strigiforms Speotyto cunicularia (burrowing owl), Bubo magellanicus (Magellan horned owl), Glaucidium nanum (Austral pygmy owl), Tyto alba (barn owl), of the falconiform Falco sparverius (American kestrel), and of the canid Pseudalopex culpaeus (Culpeo fox) have been monitored. These six species are the most abundant and persistent top predators at the site throughout the study. The data analysed here consist of 15 years of records (1987–2001), comprising 18 707 regurgitated raptor pellets and 6557 fox faeces, totalling 87 698 individual prey. The number of samples and individual prey identified for each predator and year are presented in Table 1. Details of sample collection are explained elsewhere (Jaksic et al. 1993).
Table 1. Samples collected, number of invertebrate and vertebrate prey identified, and expected prey richness for six predators during 15 years of study in Aucó, Chile. N pellets, number of pellets; N faeces, number of faeces; N vert. prey, number of vertebrate prey; N invert. prey, number of invertebrate prey; N plant prey, number of plant food items; n.d., no data obtained that year; S expected, expected richness if 20 vertebrate and 20 invertebrate individual prey were identified (numbers in boldface indicate that fewer than 20 individual prey were identified)
|Species||Precipitation (mm)||513|| 57|| 104|| 63|| 201|| 307|| 163|| 71|| 94|| 96|| 436|| 16|| 148|| 252|| 205|
| S. cunicularia||N pellets|| 338|| 126|| 1415|| 406|| 63|| 746|| 663|| 584|| 222|| 16|| 192|| 467|| 409|| 210|| 573|
|N invert. prey||2459|| 642|| 3474|| 3535|| 492||3852||2242||5087||1913|| 94|| 975||1828|| 5555||1793||3011|
|N vert. prey|| 271|| 96|| 154|| 174|| 29|| 748|| 642|| 298|| 117|| 13|| 183|| 492|| 251|| 110|| 327|
|S expected|| 18·2|| 15·4|| 14·1|| 11·2|| 10·8|| 15·4|| 14·1|| 13·6|| 12·9|| 11·3|| 14·6|| 13·6|| 13·1|| 13·2|| 12·7|
|B. magellanicus||N pellets|| 378|| 693|| 140|| 81|| 319|| 577|| 833|| 158|| 123|| 195|| 38|| 56|| 28|| 180|| 209|
|N invert. prey|| 32|| 17|| 39|| 1|| 1|| 34|| 50|| 57|| 281|| 666|| 11|| 15|| 51|| 172|| 75|
|N vert. prey||1034|| 1328|| 162|| 140|| 384|| 781||1033|| 166|| 130|| 207|| 55|| 76|| 30|| 275|| 260|
|S expected|| 12·5|| 12·1|| 14·4|| 7·25|| 6·8|| 12·5|| 12·6|| 10·7|| 11·9|| 12·1|| 9·5|| 11·1|| 14·9|| 14·8|| 11·5|
|G. nanum||N pellets|| 176|| 247|| 149|| 96|| 1|| 53|| 123|| 136|| 1||n.d.||n.d.||n.d.||n.d.||n.d.|| 98|
|N invert. prey|| 316|| 241|| 262|| 131|| 10|| 129|| 69|| 134|| 0||n.d.||n.d.||n.d.||n.d.||n.d.|| 23|
|N vert. prey|| 224|| 262|| 173|| 99|| 1|| 59|| 124|| 138|| 1||n.d.||n.d.||n.d.||n.d.||n.d.|| 101|
|S expected|| 19·6|| 18·0|| 14·8|| 15·8|| 4|| 14·6|| 12·6|| 16·4|| 1||n.d.||n.d.||N.d.||n.d.||n.d.|| 15·7|
|T. alba||N pellets|| 155|| 252|| 21|| 56|| 101|| 509||1615|| 501|| 9|| 149||1127||1442|| 44|| 271|| 572|
|N invert. prey|| 0|| 1|| 0|| 0|| 0|| 5|| 16|| 1|| 0|| 2|| 1|| 8|| 2|| 0|| 0|
|N vert. prey|| 291|| 436|| 32|| 88|| 129|| 690||2086|| 568|| 10|| 176||1483||1862|| 57|| 517|| 681|
|S expected|| 7·1|| 4·6|| 5·6|| 6·1|| 6·8|| 6·0|| 6·3|| 6·7|| 5|| 7·4|| 4·9|| 4·9|| 8·4|| 5·8|| 6·1|
|F. sparverius||N pellets|| 53|| 9|| 2|| 45||n.d.||n.d.|| 89|| 50|| 3|| 5|| 10|| 28||n.d.|| 26|| 45|
|N invert. prey|| 189|| 30|| 0|| 152||n.d.||n.d.|| 221|| 73|| 12|| 19|| 15|| 98||n.d.|| 56|| 81|
|N vert. prey|| 60|| 10|| 3|| 48||n.d.||n.d.|| 107|| 49|| 4|| 5|| 10|| 30||n.d.|| 25|| 56|
|S expected|| 21·0|| 8|| 2|| 17·0||n.d.||n.d.|| 17·5|| 15·2|| 5|| 8|| 9|| 14·8||n.d.|| 16·0|| 14·8|
|P. culpaeus||N faeces|| 260|| 1182|| 2767|| 637|| 931||n.d.||n.d.||n.d.||n.d.|| 120|| 92|| 166|| 257|| 62|| 83|
|N invert. prey|| 926|| 4305|| 3398|| 3525|| 8663||n.d.||n.d.||n.d.||n.d.|| 439|| 227|| 90|| 63|| 154|| 28|
|N plant prey||7428||60584||66416||35458||37285||n.d.||n.d.||n.d.||n.d.||2382|| 158||1831||32741||2339|| 197|
|N vert. prey|| 369|| 1205|| 780|| 578|| 781||n.d.||n.d.||n.d.||n.d.|| 126|| 115|| 213|| 101|| 86|| 114|
|S expected|| 18·6|| 17·9|| 17·6|| 15·6|| 18·0||N.d.||n.d.||n.d.||n.d.|| 11·4|| 12·7|| 11·3|| 14·9|| 15·2|| 11·4|
Prey were analysed at the finest possible taxonomic resolution. Mammals were generally identified to species level, and those identified to genus level were considered as a different taxon. Many invertebrates were identified only to the ordinal level but those identified to genus or species level were considered as separate taxa. Because taxonomic identification is not independent of the species involved, we considered that those groups identified to the genus level were likely to comprise different species from those already identified to the species level. Perhaps some prey species were already represented in the genus and species categories, but we considered that the net effect was an increase in the accuracy of the prey richness estimation. Further, there is no reason to expect an association between productivity and taxonomic resolution. This means that no biases are expected in the analysis of the association between prey richness and productivity (see below), from the level of taxonomic resolution attained.
In the study of response function among variables a key feature of an index is that it should be monotonically related with the original variable: an increase in index value should reflect an increase in productivity in spite of variation in magnitude. This ensures that the reported patterns will be robust. Thus, monotonic increases or decreases, U-shaped responses, and humped patterns will retain the trend in spite of nonlinear relationships between productivity and the index used to measure it. Precipitation satisfies these requirements, at least in our study site. In north-central Chile rains do not occur during summer (December–February) and the inter-year variation in productivity is accounted for chiefly after autumn (March–June), with the start of the rainy season. Because our aim was to relate trophic structure with productivity, we considered that the year started in autumn.
Prey link density
The total number of prey identified affects the number of trophic links estimated per species. The higher the number of individual prey identified, the more species observed (Gotelli & Graves 1996; Naya, Arim & Vargas 2002). The effect of variation in sample size can be controlled with a rarefaction procedure (Gotelli & Graves 1996). It should be noted that rarefaction is more than an index of diversity; it is a powerful standardization method (Gotelli & Colwell 2001; Arim & Barbosa 2002). With this technique one can calculate the number of prey species expected to be observed if in all predators and sample times the same number of individuals were recorded (Gotelli & Graves 1996).
The expected prey species richness if 20 individual prey were recorded was estimated for each year and predator with the rarefaction procedure. Rarefaction is a valid approach when similar taxonomic categories are involved (Gotelli & Graves 1996). Considering the types of prey present in faeces and pellets, two rarefaction procedures were conducted, one for vertebrate and another for invertebrate prey. The expected richness from both rarefactions was added. For Culpeo fox (P. culpaeus) a third rarefaction was made in order to include plant food items. The Magellan horned owl (B. magellanicus) had only 11 seeds in a total of 7563 individual prey observed. This low frequency of occurrence, and knowledge of owl biology, indicate that these plant items are unlikely to be part of the diet and appear as a result of secondary ingestion (Arim & Naya 2003); thus they were not considered. With our rarefaction analysis, we obtained an estimation of prey link density, controlling biases from large variations in the number of individual prey recorded during each sample time.
The expected prey richness for each predator was the variable used to study variation in trophic connections. When the number of invertebrate or vertebrate prey was below 20, that year was not considered for statistical analysis. Extrapolation methods could be used to estimate expected richness from samples with fewer than 20 individuals, but extrapolation precision is too low in assemblages composed of fewer than 15–20 species (Melo et al. 2003). When the number of individual prey observed was below 20, the largest prey richness observed was five species or lower. To analyse the effect of excluding observations with fewer than 20 individuals on reported patterns, we made all the analyses including and excluding data points where species richness was estimated from fewer than 20 individuals.
Association between prey link density and productivity
The association between primary productivity and predator prey link density (hereafter, link density) was analysed using multiple regression. The maximum number of records available to study this association was 15 data points. Each record represents the number of trophic connections for a given year. We considered a maximum of seven explanatory variables. Six variables were related to productivity: precipitation during the current year, during the previous year, during the previous two years, and the quadratic value of these three variables. From these variables a subset was selected with a stepwise procedure (see above). The incorporation of the quadratic term allows detection of any type of monotonic, U-shaped or humped nonlinearity in the association. Precipitation data were centred at the mean in order to minimize potential correlation between quadratic and linear variables (Sokal & Rohlf 1995). The use of polynomial regression can be considered as a nonparametric approach to obtain information about a response function (Neter et al. 1996). One variable, the year of study, was included to control long-term tendency in prey species richness (see below). To test the significance of the association between response variable and single explanatory variables, t-tests for partial correlation coefficients were used. This test allows us to analyse the association between a response and an explanatory variable when the other variables included in the regression are held constant. Note that an adjustment should be made in the degrees of freedom for the number of variables held constant (Sokal & Rohlf 1995).
Data obtained over time may have serial correlation in residuals. Long-term trend in prey richness may be a biological phenomenon or may result from a gradual change in methodology. For instance, a consistent tendency between years may originate from variation in the level of taxonomic detection of some prey groups, or in small changes in areas of sample collection. It should be noted that changes in methodology over a long-term research project do not represent an intentional or measurable process. When long-term trends are outside the scope of study, they can be statistically held constant thus eliminating the associated biases (Neter et al. 1996; Shipley 2000). The incorporation of the year of study in a multiple regression model is a way to control for this bias (Sokal & Rohlf 1995; Freckleton 2002).
Because many independent variables were considered in relation to the number of available cases, a variable selection criterion was implemented. For polynomial functional relations, the use of stepwise multiple regression is recommended (Sokal & Rohlf 1995). We used a forward stepwise regression as a guide to select variables. This statistical method should not be used to select the model automatically (Sokal & Rohlf 1995; Shipley 2000). Therefore, from the selected variables we analysed the resulting model when some variables were removed or when no selected quadratic terms were included. In this way, the effect of the addition of the study year in the model was tested in all regressions to control for potential correlation of residuals (Neter et al. 1996). The objective was to find a biologically meaningful model with the least number of variables. A total of seven multiple regressions was applied: one with the mean number of expected trophic connections among all predators, and other six with the expected prey richness of each predator as the response variable. The seven multiple regressions were also applied considering previously excluded data with fewer than 20 individual prey. In each case, the results of both analyses are presented together.
- Top of page
Stepwise regression with mean number of links per predator as the response variable, selected the year of study and the quadratic value of precipitation as potentially independent variables (F2,12 = 4·28, P < 0·04). The addition of other variables did not alter the general pattern, so we proceeded with the selected model. However, year presented a marginally significant association (t12 = 2·18; P = 0·0502). The functional relationship between mean link density and precipitation was positive (Fig. 1), but not significant (t12 = 1·13; P = 0·28). Note that in Fig. 1 and the following, year was included in the model, the y-axis represented the residuals from the regression between year and prey richness. The work with residuals in these cases was preferred in order to improve the visualization of the pattern of interest. Nevertheless, statistical analyses were multiple regressions (see methodology) and not regressions of residuals, as was suggested by Freckleton (2002).
Figure 1. Association between mean link density among all predators and precipitation in Aucó, Chile. Precipitation is used as a proxy variable of productivity. Data on the y-axis are residuals from regression between mean link density and year of study (R.Y., residual of year). The x-axis is precipitation of the year centred on the mean. Px2 indicates the probability from t-test for the partial regression coefficients for the quadratic value of the x-axis.
Download figure to PowerPoint
With regard to the calculation of mean link density, observations based on fewer than 20 individual prey did not affect the observed pattern (Fig. 1). However, the power of the test was reduced by the increase in data dispersion. The model was not significant (F2,12 = 1·53, P < 0·26), same as for precipitation (t12 = 1·47; P = 0·17) or for year (t12 = 0·4; P = 0·69). It should be noted that in addition to the lack of statistical association, the coefficients of variation (standard deviation/mean) of the mean number of links per predator were small, both considering years with fewer than 20 individuals observed (CV = 0·17) and without considering these years (CV = 0·11).
On an individual basis, predator species presented an idiosyncratic pattern of response to variation in productivity (Figs 2, 3 and 4). In most regressions the models suggested by the stepwise procedure were reduced by the exclusion of some variables. This was the case for Speotyto cunicularia, Bubo magellanicus, Glaucidium nanum and Falco sparverius, where it was possible to obtain a good enough model (significant and with a high r2) with fewer variables (Neter et al. 1996). In all cases the response function was not different when observations with fewer than 20 individuals prey were included. However, probabilities were affected by the addition of data points, usually increasing the variance of the response variable (but see Fig. 2).
Figure 2. Association between link richness and precipitation for Speotyto cunicularia in Aucó, Chile. Px indicates the probability associated to the t-test for the partial regression coefficients between x-axis and y-axis; Px2 indicates the probability associated with the t-test for the partial regression coefficients between the quadratic value of x-axis and y-axis; t, t − 1, and t − 2 indicate no time lag, 1-year lag, or 2-year lag, respectively, in reference to precipitation. R2 in part (a) represents the explained variance by multiple regression.
Download figure to PowerPoint
Figure 3. Association between link richness and precipitation for Bubo magellanicus in Aucó, Chile. t − 1, and t − 2, same as in Fig. 2. R.Y. is the residual of the regression between year of study and link density. R. t − 1 is the residual of the regression between link density and precipitation with a time lag of 1 year.
Download figure to PowerPoint
Figure 4. Association between link density and precipitation for the remaining four top predators in Aucó, Chile. t, t − 1, t − 2 and R.Y. same as in Fig. 2 and 3. Work with residuals allows to visualize the functional association of variables. R2, Px, and Px2, as in previous figure. Pseudalopex culpaeus always had more than 20 individual prey.
Download figure to PowerPoint
The richness of trophic connections of Speotyto cunicularia was positively associated with precipitation of the year and of the previous two years. The stepwise procedure selected precipitation of the year, its quadratic value, precipitation of the previous year, that of the two preceding years and their quadratic value. Because this model involves many of the variables available, we searched for one with fewer variables. A model with the quadratic value of precipitation, and precipitation of the previous two years was good enough (F3,10 = 8·1; P = 0·0049; Fig. 2). This model suggests that S. cunicularia diet richness positively responds to productivity at all time lags analysed and that the significance of the association declines with the extent of the lag. That is, without lag, the association was highly significant (t10 = 4·6; P = 0·00099), with a one-year lag it was marginally significant (t10 = 2·22; P = 0·051), and with a 2-year lag it became insignificant (t10 = 1·75; P = 0·11). This model was improved (F3,11 = 9·73; P = 0·002; Fig. 2) by the addition of one richness estimation based on fewer than 20 individuals (year 1996 in Table 1).
Prey richness in the diet of Bubo magellanicus presented a time lag in the functional relation to precipitation. The general model was statistically significant (F1,8 = 6·35; P = 0·036) and suggested a U-shaped functional relationship between prey richness in the diet and precipitation accumulated 2 years before (t8 = 2·5; P = 0·036). The same pattern was observed when five years with invertebrate prey represented by fewer than 20 individuals were included (Table 1 and Fig. 3). Addition of these data yielded a marginally significant model (F1,13 = 4·46; P = 0·055), with the same U-shaped functional relationship (Fig. 3).
Prey richness in the diet of Glaucidium nanum presented a U-shaped functional relationship vs. productivity without time lag (F2,5 = 6·51; P = 0·04; Fig. 4a). The stepwise procedure included year of study, precipitation, precipitation of the previous year, and the quadratic value of precipitation. We searched for a model with fewer variables, and found that a regression model with precipitation (t5 = 1·68; P = 0·15) and its quadratic value (t5 = 3·23; P = 0·023) as independent variables was good enough. Although only the quadratic term was significant, combination of both variables better showed the shape of the functional relationship (Fig. 4a). Addition of two years with fewer than 20 individuals (1991 and 1995) did not change the response function (Fig. 4A) but the model was not as significant (F2,7 = 0·97; P = 0·42). It should be noted that the two data points added to this model are based on a single owl pellet.
The diet of Tyto alba was composed chiefly of vertebrate prey. Invertebrate prey was rare, 36 individuals having been observed among 9106 prey items. Further, a maximum of 16 invertebrate prey was observed in a single year (Table 1). Because of this, the analysis of prey richness for T. alba considered only vertebrate prey. With a one-year time lag T. alba linearly reduced its dietary prey richness as productivity rose (t12 = 2·73; P = 0·018; Fig. 4b). The addition of one observation with fewer than 20 individual prey (year 1995 in Table 1) yielded the same pattern (t13 = 2·18; P = 0·048; Fig. 4B).
The diet richness of Falco sparverius presented a positive linear response to precipitation without time lag (Fig. 4c). The general model was significant (F1,5 = 8·75; P = 0·031). However, when observations based on fewer than 20 individuals were included, the model was no longer significant (F1,10 = 1·27; P = 0·29; Fig. 4c), although retaining the same association pattern.
The diet richness of Pseudalopex culpaeus presented a humped functional relationship to precipitation and a significant association with year (Fig. 4d). The general model was statistically significant (F3,7 = 11·51; P = 0·005) and included year of study (t7 = 5·78; P = 0·0007), precipitation (t7 = 2·27; P = 0·058) and quadratic value of precipitation (t7 = 2·35; P = 0·051) as independent variables. Note that the probabilities that relates precipitation with prey richness was marginally significant.
- Top of page
The set of species analysed were top predators present throughout most of the 15 years of study. All six species responded to productivity but they showed idiosyncratic patterns of association between number of trophic connections and interannual variation in productivity (Figs 2, 3 and 4). These significant responses to productivity at the species level did not involve corresponding variations in food web structure among top predators as a whole (Fig. 1). Further, the constant value of the mean number of predator/prey links is likely to emerge from compensatory variations at the level of predatory species. To our knowledge, this is the first time this type of result is communicated. At least in our study system, food web structure displays a dualism of (i) variation at the level of predator species, simultaneous with (ii) constancy at the level of predators as a whole.
Individual species trends with productivity (Figs 2, 3 and 4) had to be mutually cancelled in order to produce the lack of association observed for the entire assembly of predators (Fig. 1). The open question is whether this is just a statistical consequence of combining independent trends, or whether there are ecological constraints on species connectance. We have evidence suggesting the involvement of purely ecological mechanisms. For instance, El Niño disturbances that produce large changes in productivity also affect predator membership to different feeding guilds (Jaksic et al. 1997). Perhaps, when one species reduces the number of prey in its diet, new resources become available to other predators from the relaxation of direct or indirect interactions; the opposite being expected when one dominant predator increases its diet richness (Schoener 1971, 1974; Case & Gilpin 1974). This mechanism could account for the complementary response in prey richness of predators at our study site. Nevertheless, we recognize that the contribution of purely statistically generated patterns in nature cannot be downplayed in ecological studies (Gotelli & Graves 1996; Arim & Barbosa 2002).
Recent attempts to relate species richness with link density have in most cases not found a significant association (Winemiller et al. 2001). Our work does not suffer from the noise ensuing when comparing different ecosystems, species or methodologies. Therefore, in our case it was possible to find an association between the number of trophic connections and productivity for individual predator species. This leads us to think that the lack of association reported here for the entire assembly is not a consequence of low statistical power or of too much variation, as in other studies (Winemiller et al. 2001).
The dynamic nature of links within food webs has been recognized elsewhere (Paine 1980; Pimm & Lawton 1980; Holt 1996; Tavares-Cromar & Williams 1996). However, methodological constraints have limited the number of studies on temporal variation of food web structure (see Pimm & Kitching 1987; Winemiller 1990, 1996; Schoenly & Cohen 1991; Closs & Lake 1994; Tavares-Cromar & Williams 1996, for qualified exceptions). In spite of the high-quality food webs recently compiled (Dunne et al. 2002a), not a single food web is yet available with comparable species composition, recorded at different times, or in similar environments with different productivity. The approach used here to analyse a community subset is promising in order to relate food web structure with environment and community attributes (Holt & Polis 1997; Winemiller et al. 2001). Our data base is good enough to find a significant pattern of food web structure, which at least among top predators is involved in the conservation of topological structure despite large variation in system conditions. This pattern may be conserved in other food web compartments or in different food webs, a point that could be analysed in other ongoing long-term studies.
Three owl species Speotyto cunicularia, Bubo magellanicus and Tyto alba displayed a time lag in the association between productivity and number of trophic connections (Figs 2, 3 and 4b). Time lags suggest the involvement of population processes in determining observed patterns. Among predator populations, changes in abundance often imply variation in the proportions of species with different feeding strategies (Sergio, Pedrini & Marchesi 2003). Lags in the response of prey populations imply lags in the variation of relative abundance and availability of resources that could promote changes in predator diet richness. Thus, the association between trophic structure and environmental conditions may be obscured by time lags in the response. Further, it implies that experimental and observational studies should be long enough to observe patterns that may take two years before realization (Figs 2 and 3).
The dynamic nature of food web structure has been little addressed. This study detected significant and idiosyncratic associations between the number of trophic connections and productivity separately for six top predators. However, no association was detected between the mean number of connections among all predators combined. Pure statistical and/or ecological mechanisms could be involved in these contrasting results. Nevertheless, conservation of food web topology, in spite of large variation in productivity, species composition and individual species diet richness, supports the view that connection pattern is a key attribute of food webs.