Potential responses of diversity indices.
The Polychaeta were the most diverse class of benthic invertebrates in the regional dataset, and Malacostraca, Bivalvia, and Gastropoda were the next most diverse classes overall after Polychaeta (Table 1). The composition of benthic invertebrate assemblages was relatively more diverse in the low-frequency group (see Materials and Methods) than in the other groups in the regional dataset, whereas Polychaeta were dominant in the high-frequency group. The average number of families observed at each location in the regional dataset was 12.4 ± 8.2 (mean ± standard deviation) with a range of 0–36 (Table 2). We set up low-, intermediate-, and high-frequency groups to have a similar number of families among groups.
The sediment characteristic water content (WC, %), ranging from 24.7% to 360%, had significant power to explain the variability in family density in the overall regional dataset (Fig. 2a; the result of averaged model based on Akaike information criterion (AIC), and see Supplementary Table S1 for the accepted candidate models). Water depth, latitude, and sample size were not significant. Also, the effect of WC was significantly negative in the low-frequency group (Supplementary Fig. S1). In this group, the ratio of the effect size of WC to the standard deviation of random effects was more than double in absolute value the values in the other groups (Table 2).
The inverse Simpson’s concentration index for the overall data from the regional dataset (N = 65) was 4.3 ± 3.1 (mean ± standard deviation). Pielou evenness was 0.71 ± 0.21, where Pielou evenness was defined (N = 60). In the unreliable data (see Materials and Methods for the definition of reliable and unreliable data), the inverse Simpson’s concentration index was not saturated (Supplementary Fig. S2), and Pielou evenness tended to be greater than in the reliable data at the same family density (Supplementary Fig. S2). The inverse Simpson’s concentration index and Pielou evenness became 4.8 ± 3.4 and 0.64 ± 0.18 (N = 36), respectively, in the reliable data, where WC ranged between 32.7% and 263%.
The effect size of WC became negative for the inverse Simpson’s concentration index in the analysis of reliable data (Fig. 2b). Among models with low AIC values, the effect size of WC was more negative in models that used only the reliable data than in models that used both reliable and unreliable data (Supplementary Table S2). The confidence interval for Pielou evenness extended farther into the negative range in the analysis of reliable data (Fig. 2c). This extension was produced because low-AIC models including WC were selected for the reliable data, whereas the explanatory variables varied in analyses with small AICs for the overall data (Supplementary Table S3). Although the effect of WC did not differ significantly from zero, its shifts toward the negative mean that there was a greater possibility of selecting a model including WC when using the reliable data. Sample coverage increased significantly from 0.79 ± 0.28 in the overall data to 0.94 ± 0.04 in the set of reliable data (N = 36; t = −4.1, P = 1.2×10−4; Welch two sample t-test), and from 0.86 ± 0.16 to 0.94 ± 0.04 in observations where Pielou evenness was defined (N = 36; t = −3.5, P = 7.8×10−4).
Assessments at two local sites.
Community structure in Matsunaga Bay.
In Matsunaga Bay, Polychaeta, Malacostraca, Bivalvia, and Gastropoda were the most diverse classes of benthic invertebrates (Table 1), which was the same as in the regional dataset. The ranges of values for other sediment variables were similar to those in the regional dataset except for sediment temperature (Supplementary Fig. S3). Because this local dataset was acquired in winter (see Appendix S1), sediment temperature (13.9 ± 1.9 °C; mean ± standard deviation) was lower than the temperature in the regional dataset (24.8 ± 3.5 °C; between 10.4 and 30.4 °C).
Cluster analysis based on Sørensen dissimilarity was unable to show any apparent separations for species compositions in Matsunaga Bay (Fig. 3a); species compositions were similar within the intertidal flat at the mouth of a small river (sample locations IF1–IF5) and within the one in the inner bay (IF6–IF10). By contrast, cluster analysis based on Euclidean distance after log(n + 1) transformation showed that the community structure at the mouth of the small river was quite different from that at other locations in the bay (Fig. 3b). Distance-based redundancy analysis (dbRDA) showed that the four explanatory variables selected explained 56.9% of total variance in the benthic invertebrate community. The variability in community structure correlated with median sediment particle size (D50; 31.6%, P = 0.0001) (Fig. 3c), which appears as a difference in structure between the community at the mouth of the small river and the other observations in the first dbRDA coordinate axis. The polychaete Simplisetia erythraeensis was the most dominant in the community at the mouth of the small river. The variation in community structure was correlated with temperature (13.0%, P = 0.0004), depth (8.0%, P = 0.0044), and TOC (4.4%, P = 0.042).
Taken together, the cluster analysis and dbRDA for Matsunaga Bay indicate that there is a specific community at the mouth of the small river (hereafter, the “river-mouth” community) with more individuals than at the other locations. The data were reliable for this river-mouth community but unreliable at the other 25 locations.
Community structure in Nagoya Port.
The sediment variables in Nagoya Port had ranges similar to those in the regional dataset (Supplementary Fig. S3). The range of sediment temperatures was within that in the regional dataset because both datasets were obtained in summer (Supplementary Appendix S1).
The cluster analysis based on Sørensen dissimilarity was performed for all data from Nagoya Port except for sites N5, N9, N10, and N12, where there were no individuals sampled. The resulting tree showed that at the higher levels, the Fujimae intertidal flat and sites N6 and N17 could be separated from observations at other locations (Fig. 4a). However, the data for N6 and N17 were classified as unreliable. This means that the Fujimae intertidal flat was a habitat type that differed in species composition from other reliable data. The cluster analysis based on log-transformed abundance showed that N8 and N20 were separate specific habitat types in the community structure (Fig. 4b). The second outcome of this analysis is that the habitat type of the Fujimae intertidal flat can be separated from the other 11 locations with reliable data. The dbRDA coordination resulted in four explanatory variables explaining 38.4% of total variance in the benthic invertebrate community in Nagoya Port. Salinity separated the community structures of the Fujimae intertidal flat, N8, and N20 from other locations (Fig. 4c), however, this explained only 11.1% of the variation (P = 0.001). The other three explanatory variables also had relatively low contributions to the variability in community structure: the molar carbon to nitrogen ratio (C/N; 11.2%, P = 0.016), WC (9.2%, P = 0.0006), and D50 (6.8%, P = 0.039).
Taken together, these analyses suggest that sites N8 and N20 had specific community structures in terms of abundance, and that the Fujimae intertidal flat represents a distinct cluster in terms of species composition and abundance. These sites can be judged to include minor communities (see Materials and Methods). Conversely, the remaining reliable observations (data from 11 sites) were relatively isolated.
Responses of diversity indices at two local sites.
Because the analysis of the regional dataset showed that WC was the sediment variable with the most impact, we analysed trends in the diversity indices at the two local sites with WC as an explanatory variable.
Species density decreased significantly with increasing WC in all data types at the two local sites (Table 3). The slope was similar in the analyses for both overall and only test communities in Matsunaga Bay (Fig. 5a). In Nagoya Port, although the response to WC was significant for both data types, the slope was more moderate in species density in the overall data than in the data excluding minor communities (Fig. 5b).
Sample coverage was 0.96 ± 0.02 (mean ± standard deviation) in the river-mouth community and 0.61 ± 0.24 at the other 25 locations in Matsunaga Bay. The inverse Simpson’s concentration index was saturated in the river-mouth community but not saturated at the other 25 locations (Supplementary Fig. S4). At the other 25 locations, Pielou evenness tended to be high, approaching 1.0 (Supplementary Fig. S4). In Nagoya Port, the inverse Simpson’s concentration index was almost saturated for the reliable data (Supplementary Fig. S5), where sample coverage was 0.95 ± 0.04. Pielou evenness was lower at the two sites with unreliable data than at those with reliable data (Supplementary Fig. S5). We analysed the response to WC for the inverse Simpson’s concentration index and Pielou evenness in Nagoya Port. However, similar analyses were not performed in Matsunaga Bay because of the dearth of reliable data.
The trend of the inverse Simpson’s concentration index with WC was not significant in the analysis for reliable data with combined specific and test communities in Nagoya Port, but decreased significantly in the analysis for reliable data with only test communities (Table 3), which excluded the six sites with unreliable data that had a low index at high WC (Fig. 5c). Analyses including both reliable and unreliable data also showed significant negative responses to WC (Table 3). Pielou evenness showed no significant response in the sets of reliable data. However, in the analysis including unreliable data with test communities N6 and N17, which plotted at low Pielou evenness and high WC (Fig. 5d), Pielou evenness showed a questionable negative response to WC (t = −2.3, P = 0.044) (Table 3).