Size–Density Relationships: a Cross-Community Approach to Benthic Macroinvertebrates in Mediterranean and Black Sea Lagoons

Abstract

Cross-community scaling relationships (CCSRs), which result from individual density scaling with average individual body size at guild and community levels, enable investigation of energy constraints at high levels of the ecological hierarchy. Here, we studied cross-community scaling relationships in benthic macroinvertebrate guilds in 15 Mediterranean and Black Sea lagoon ecosystems characterized by strong habitat heterogeneity and high energy density, using data already available in the LifeWatch-Italy data portal. The study sought to describe CCSR patterns in lagoon ecosystems, analyzing their variability across habitat and ecosystem types and evaluating the relative influence on individual body size, macroinvertebrate guild density, or both, of proxies of ecosystem properties, including physiographic characteristics and external disturbance, acting as potential drivers. Significant CCSRs were observed in benthic macroinvertebrate guilds in Mediterranean and Black Sea lagoons. They were characterized by high internal variability and slopes less negative than the metabolic scaling theory expectation (b = −0.75), ranging between b = −0.27 and b = −0.50. Lagoon ecosystem typology, inter-lagoon variation, and ecosystem properties explained part of the variation in internal CCSRs, while habitat variation and intra-ecosystem habitat heterogeneity did not show any influence. CCSR intercepts expressing macroinvertebrate-specific densities showed patterns of variation that were consistent with those of proxies of ecosystem energetics and parsimony, such as eutrophication, chemical and physical disturbances, and openness. These relationships highlight the relevance of CCSRs, which enable inferences on the properties, functioning, and ecological status of ecosystems from simple analyses of community structure.

Appendix

Simulation Test for the Dichotomous Analysis of Abiotic Parameters’ Influence on CCSR Shape Components

In order to test the relevance of various selected abiotic parameters to the shape of CCSRs, we used a dichotomous analysis, clustering field sites into HIGH and LOW value groups (see the “Material and Methods”—“Statistical Data Analysis” section). This analysis served first to account for potential non-linearities in the response of the size–density ratio to each single abiotic gradient and, second, to set aside a basic assumption of linear multivariate regression analysis (i.e., that different relationships have separate intercepts but the same slope). However, the OLS estimator used to assess intercepts and slopes (and the significance of their differences across HIGH and LOW levels) is potentially biased due to heteroskedasticity in the residuals distributions.

To assess whether our observations are robust with respect to potential biases due to data heteroskedasticity, we performed a null model simulation test. The test aimed to check whether significant differences in estimates between LOW and HIGH groups are likely to arise from stochastic effects, rather than from the deterministic influence of the abiotic gradient on the size–density ratio, as we assume.

Rank Randomization

Randomization techniques are often used in statistics to generate null distributions with which the estimated values can be compared (Good 2005). The distribution of the test statistic under the null hypothesis was obtained by calculating all possible values of the test statistic under rearrangements of the labels on the observed data points. This makes it possible to assess whether the observed patterns arise from stochastic effects (i.e., the estimated coefficients are not significantly different from the average of the values obtained by randomizing the dataset), rather than from deterministic influences (i.e., the estimated coefficients are significantly different from the average of the values obtained by randomizing the dataset).

In our case, we assume that the rank of a site (i.e., HIGH or LOW) along certain abiotic gradients has a deterministic influence on the size–density ratio, so that size–density relationships fitted across the HIGH and LOW categories should significantly differ in terms of intercepts or slope. To test whether these observations are biased due to stochastic effects arising from sensitivity of the data to heteroskedasticity of the residuals distribution, we performed a randomization test on the ranking of the size–density observations along every abiotic parameter gradient. Keeping the HIGH and LOW categories of the abiotic parameter gradient fixed, we

• 1. randomized the order of the size–density observations along the gradient;

• 2. fitted the linear regression model for the HIGH and LOW size–density relationships thereby obtained;

• 3. ran the ANCOVA statistics to compare intercept and slope between the HIGH and LOW size–density relationships;

• 4. iterated the procedure 9999 times;

• 5. built the probability distribution of intercepts and slopes of the HIGH and LOW size density relationships linked to every abiotic parameter (Fig. 9); and,

• 6. built the probability distribution of significant differences between HIGH and LOW cluster intercepts and slopes, for every abiotic parameter (Table 4).

Table 4 Probability (%) of estimating significant differences in intercepts and slopes between the size and density relationships fitted for the HIGH and LOW levels of abiotic parameter gradient, when size–density observations are randomized along the gradient (9999 bootstrap iterations)

The density distributions (ln n. of cases) of intercepts and slopes of the HIGH and LOW clusters for every abiotic parameter are shown in Fig. 9, together with the observed intercept and slope values. The likelihood that the differences in intercepts and slopes would be observed once the size–density observations ranking has been randomized along the abiotic parameter gradient is very low (respectively <1.7, <3.3, <0.3%; Table 4). Thus, we rejected the null hypothesis (that differences arise simply from stochasticity in data distribution) and accepted the alternative (that significant differences in intercepts, slopes, or both across the HIGH and LOW categories are related to the deterministic influence of the ranking along the abiotic gradient).