The use of diversity indices for local assessment of marine sediment quality
Our analyses of marine invertebrate communities at a regional scale and at two local sites revealed that taxonomic density (i.e. species density) was a sensitive index of marine sediment quality. However, although Hill–Simpson diversity and Pielou evenness were shown to respond to sediment variability in the regional dataset, they could be insensitive or respond falsely when a low number of individuals was observed, and when more than one community co-existed at a local site. These results from two local sites should serve as a point of caution when using diversity indices. Although these indices can provide a good understanding of how communities respond to sediment degradation, it is important to understand how these indices collapse when there is a small number of individuals observed or when the data span multiple co-existing communities. This emphasises the need for better strategies for the ecological assessment of sediment quality based on diversity indices at the local scale in marine areas.The analyses of the regional dataset show that WC had a larger impact than other variables on the taxonomic density of benthic invertebrate communities, although grain size and organic matter content are also thought to affect benthic invertebrate richnesse.g.17,18. The high contribution of WC likely reflects its physical effects on sediment structure. The optimal range of WC for the burrowing activity of benthic invertebrates is between around 25% WC at the densest (i.e. hardest) and around 40% WC at the loosest (i.e. softest) packing of sediment19,20. A WC value exceeding the upper limit of this optimal range could indicate sediment that is too soft for the burrowing activity of benthic invertebrates; this may explain the negative effect of WC on taxonomic density observed in this study.The relatively high standard deviation of random effects as compared against the effect size of WC in GLMMs suggests that unmeasured variables had a strong effect on taxonomic density. Salinity21 and anthropogenic impacts, such as dredging and trawling13, are well-known factors that could affect the diversity of benthic invertebrates. However, they were not considered in the regional dataset, because these factors are site- and sampling-location specific, and therefore it was impossible to identify which factors needed to be measured prior to investigation. Our study highlights one advantage of GLMMs, which is the ability to show the effects of these unmeasured factors. The effect size of WC was almost as large as that of the random effects in the low-frequency group (Table 1), which suggests that rare invertebrates were more sensitive to sediment degradation in this group, and that this sensitivity contributed to the overall response of taxonomic density.The analyses of the regional dataset also showed that an increase in WC caused a decrease in Hill–Simpson diversity and Pielou evenness in the reliable data. This result is consistent with previously identified responses to sediment degradation6. Because the low values of these indices occurred in communities with a few dominant species, this suggests that the benthic community was dominated by a few species in soft sediments (i.e. where WC was high).Similarly, increasing WC was associated with a significant decline in species density at the two local sites (Table 2), and the trend was significant for both reliable and unreliable data. This suggests that WC can be an indicator of benthic invertebrate species density at the local scale. However, it is likely that the trend in species density was not only caused by the effect of sediment softness (as is suggested by our analysis of the regional dataset) but also by other factors. One such factor is anoxia, which has been observed from August to October in the water column above the sediment in Nagoya Port at locations where no individuals were sampled (i.e., N5, N9, N10, and N12)22. Similarly, high organic-carbon and trace-metal concentrations have been reported in our study area23. These factors could have co-occurred with high WC, and thereby contributed to the decline in species density observed in our study. Because spatial correlations between variables tend to occur at local scales24, it is difficult to identify factors that affect species density at this scale. Species density is itself a sensitive indicator; however, if alternatives are needed, parameters that explain variations in species density, such as WC, are recommended for use as a representative variable in local assessment.WC did not always have a significant negative effect on Hill–Simpson diversity or Pielou evenness at the local scale (Table 2). The significant negative effect of WC on Hill–Simpson diversity identified in the reliable data from Nagoya Port indicates that community structure was dominated by a few species at higher WC, which mirrors the results obtained from the regional dataset. However, WC had no significant effect on Pielou evenness, and even the effect on Hill–Simpson diversity was only significant once locations that have different coexisting community structures (i.e., the Fujimae tidal flat, N8, and N20) were excluded from the analysis. These results mean that these diversity indices are not as sensitive to changes in WC as species density. Conversely, we found questionable significant negative effects of WC on both Hill–Simpson diversity and Pielou evenness when unreliable data were included in the analysis (Table 2). The low values of these indices obtained at high WC likely reflect artefacts in the unreliable data (Fig. 5c, d).It is important to find and exclude coexisting communities when analysing the effects of sediment degradation on indices of community structure (i.e., Hill–Simpson diversity and Pielou evenness) in a target community. In Matsunaga Bay, the river-mouth community on the intertidal flat was found to have a distinct sediment-particle-size composition compared to other communities in the bay based on multivariate analysis (Fig. 3c). In addition, because the polychaete Simplisetia erythraeensis that dominated the river-mouth community can be found in brackish environments (WoRMS: http://www.marinespecies.org/), low salinity (which was not measured in this study) may be a distinguishing feature of this location. Therefore, environmental characteristics such as sediment particle size, salinity, and the location of the intertidal flat likely underlie the spatial variability of community structure in this bay.Whereas we were able to predict the spatial variability of community structure prior to field sampling in Matsunaga Bay, this was not true in Nagoya Port. Our a priori expectation was that the benthic community on the Fujimae intertidal flat would have a distinct structure because of its location; although this was borne out by the data, we were unable to predict that there would also be distinct community structures at N8 and N20 because of the complex spatial patterning of benthic communities in this area. The explanatory variables we selected (salinity, C/N, WC, and D50) explained less than 11% of the total variance in community structure. This weak explanatory power indicates that unmeasured environmental variables may underlie the complex spatial patterning of benthic communities observed in our study, which is typical of the complexity often found in urbanised marine areas13.Although our results demonstrate that excluding distinct coexisting communities from the overall data is important when analysing species density (Fig. 5b) and Hill–Simpson diversity (Fig. 5c), such communities can be difficult to distinguish prior to field sampling. Therefore, post-hoc multivariate analysis is needed to distinguish between a target community and other communities. In addition, because diversity indices are affected by both species composition and the proportions of individuals in each taxon, the use of multiple distances between sampling points is recommended to assess how communities differ across space.The unreliability of Hill–Simpson diversity and Pielou evenness values calculated from small sample sizes can be explained by a theoretical framework for the effective number of species9. The effective number of species, which reflects the number of dominant species14, is predicted to decline or remain unchanged in response to low species density in cases where taxonomic density has a sensitive negative response (Fig. 6a). However, the effective number of species can be underestimated when there is a small sample size (Fig. 6b). This suggests that the questionable negative responses of Hill–Simpson diversity and Pielou evenness (which is calculated from the Shannon index) (Table 2) likely do not reflect real changes in community structure in Nagoya Port, but instead are caused by an artefact that negatively correlates with sediment degradation. However, low Pielou evenness was rarely associated with unreliable data in our study (Appendix S2). Pielou evenness tended to be high, approaching 1.0, in unreliable data from the regional dataset (Supplementary Fig. S2) and Matsunaga Bay (Supplementary Fig. S4). This bias can be explained as a possible result of small sample size. Our results should serve as a warning that false or insensitive responses in Hill–Simpson diversity and Pielou evenness may occur if sample size is insufficient to estimate these indices accurately.Figure 6Two mechanisms that can affect the effective number of species (which can be estimated with Hill–Simpson diversity). (a) The effective number of species becomes lower at low species density with sufficient sample size. When the degradation of sediment quality (SQ degradation) affects species density but not the density of individuals, the effective number of species decreases as a real response to community structure. However, as shown in (b), the effective number of species also becomes lower at small sample size n. When SQ degradation affects the density of individuals, the effective number of species might not reflect a real response in community structure.Full size imageIn this study, we used a sample size of 50 individuals as the threshold between reliable and unreliable data. Although a sample-size threshold can be useful when judging whether a sample accurately reflects community structure, the specific value we used was not based on any scientific evidence. In fact, our datasets included several data points classified as “reliable” that were not sufficiently saturated in Hill–Simpson diversity (Supplementary Figs. S2a, S4a, and S5a). Sample coverage is an index that standardises the number of taxa observed by the completeness of the sample15,25. The sample coverage of the reliable data was close to 1.0 (complete) and greater than that of the unreliable data in all three datasets used in this study. Although the rarefaction curve is a more direct way to show the estimation accuracy of Hill–Simpson diversity, the simplicity of the sample coverage index (as compared to drawing a rarefaction curve) is an advantage when judging data reliability. In addition, sample coverage is useful when plotting the degree of accuracy in two-dimensional figures, as was done in this study.When the number of individuals observed, n, is not sufficient to estimate the indices of community structure accurately, we can use an extrapolation technique that provides more reliable estimates by doubling the number of individuals observed to 2n9. Although we did not use this technique in this study because our objective was to explore how small sample sizes affect assessments of marine sediment quality, this technique is a useful solution for practical assessment when the number of individuals observed is not sufficient.In conclusion, our results show that species density responds sensitively to sediment degradation. By contrast, indices of community structure (i.e. Hill–Simpson diversity and Pielou evenness) were insensitive at the local scale because of masking by multiple coexisting communities, and sometimes produced misleading results because of inaccuracies associated with small sample sizes. Because indices for community structure provide a good understanding of how communities respond to sediment degradation, which cannot be provided by species density, ecological approaches using these indices have merits for assessing sediment quality because they are more realistic under field conditions3 and because they reduce uncertainties26,27. The potential for misleading and insensitive results must be avoided to keep from diluting these merits. We recommend that these diversity indices for community structure be used in local assessments only if it is possible to obtain a sufficient sample size for accurate estimation, and if co-existing communities can be differentiated before field sampling or by post-hoc analysis through sampling at multiple distances. More
