Discordance between living and death assemblages as evidence for anthropogenic ecological change

Kidwell. 10.1073/pnas.0707194104

Supporting Information

Files in this Data Supplement:

Fig. 4. Scatterplots of 73 large (at least 20 live and 20 dead individuals) estuarine data sets, coded by AE, showing lack of variation in taxonomic similarity (Jaccard-Chao index; Left) and rank-order correlation of species abundances (Spearman rho; Right), as a function of (A and B) number of stations pooled to generate each habitat-level data set; (C and D) number of live species in the data set, which is proportional to the log of live individuals sampled, and is usually smaller than the number of dead species and individuals; (E and F) ratio of dead: live individuals; (G and H) mesh size in mm; and (I and J) bottom habitat type, coded as 1 = mud (<10% sand), 2 = sandy mud (<50% sand), 3 = muddy sand (<50% mud), 4 = clean sand (<~ 5% mud), 5 = shell or lithic gravel, gravelly sediments of any kind, 6 = grassbed. Data sets from AE1 and especially AE2 settings typically have smaller sample sizes and none were processed using mesh >1.5 mm, but AE0 data sets commonly have the same characteristics and yet yield on average higher fidelity.

Fig. 5. Scatterplot of raw proportional abundance of the opportunistic bivalve Mulinia lateralis in data sets from Gulf of Mexico lagoons and bays. Coarse-mesh data sets are from Tabasco and Veracruz, Mexico, and from Laguna Madre, southern Texas. Fine-mesh data sets are from northern Texas. Mulinia is relatively rare both alive and dead among pristine lagoons (all but one data set have proportional abundances <10%), with good live-dead agreement in proportional abundance. Mulinia ranges up to ~ 60% alive and 75% dead in AE ≥ 1 lagoons and is commonly among the most abundant or is the most abundant species, consistent with greater stress, such as from episodic hypoxia associated with AE. Death assemblages in AE ≥ 1 settings are not, however, consistently enriched in Mulinia compared to the living community, in part because of the dead-abundance of grass-dwellers and other relictual species.

Fig. 6. Same plot as Fig. 2, but for 34 open shelf data sets. Solid line connects mean values for each AE category with 95% confidence intervals on the standard error. (Upper) J-C declines significantly with increasing AE score, based both on comparison of means (solid line, 95% confidence intervals on the standard error of the mean) and medians (Kruskal-Wallis H = 14.0, P = 0.0009). The three pristine AE0 data sets having lowest J-C values are from the Amazon (fine-mesh, high nutrients from upwelling) and Yucatan (coarse-mesh, oligotrophic). The five AE1 data sets having highest J-C values are from three different shelves (fine-mesh Milazzo and Galveston, coarse-mesh Eddystone), and the five AE1 data sets having the lowest J-C values are from four different shelves (fine-mesh Milazzo, Brucoli; coarse-mesh Livorno, Eddystone). (Lower) Spearman rho declines significantly with severe AE2, based both on comparison of means and medians (Kruskal-Wallis H = 6.6, P = 0.0369). Summary statistics for pristine shelves (n = 18): mean J-C = 0.89 ± 0.06, with 95% confidence intervals calculated on the standard error; median = 0.92. The range of rank-order agreement is broad, but all correlations are positive and most are significantly so (61% of rhos positive at P < 0.05, 39% at P < 0.01, following sequential Bonferroni correction; mean rho = 0.41 ± 0.08, median = 0.38). For AE ≥ 1 shelves (n = 16): mean J-C = 0.51 ± 0.16, median = 0.42. The range of rank-order agreement is very broad, from -0.58 to 0.53, with 75% significant at P < 0.05 and 56% significant at P < 0.01; mean rho = 0.19, median = 0.33.

Fig. 7. Scatterplots of 34 large (at least 20 live and 20 dead individuals) data sets from open shelves, coded by AE, showing lack of variation in taxonomic similarity (Jaccard-Chao index; Left) and rank-order correlation of species abundances (Spearman rho; Right), as a function of (A and B) number of stations pooled to generate each habitat-level data set; (C and D) number of live species in the data set, which is proportional to the log of live individuals sampled, and is usually smaller than the number of dead species and individuals; (E and F) ratio of dead: live individuals; (G and H) mesh size in mm; and (I and J) bottom habitat type, coded as 1 = mud (<10% sand), 2 = sandy mud (<50% sand), 3 = muddy sand (<50% mud), 4 = clean sand (<~ 5% mud), 5 = shell or lithic gravel, gravelly sediments of any kind, 6 = grassbed. Data sets from AE1 and especially AE2 settings typically have smaller sample sizes and none were processed using mesh >1.5 mm, but AE0 data sets commonly have the same characteristics and yet yield on average higher fidelity. Study areas and sources listed in SI Table 2.

Fig. 8. Same scatterplots as Fig. 1, but with rectangles outlining ± 1 (solid line) and ± 2 (dotted line) standard deviations of the population mean of Jaccard-Chao and Spearman rho for pristine data sets. In A and B, the larger pair of rectangles indicate the SD1 and SD2 boxes if the outlying pristine data set is included, and the smaller pair of rectangles exclude the outlier. If the broader SD2 and SD1 rectangles are used, the false-negative rate for AE ≥ 1 estuarine data sets increases from the 43 and 29% values given in the text to 76 and 42% respectively. A. Among the pristine estuarine data sets, the low-fidelity outlier is an unvegetated silty sand in Smuggler's Bay, Virgin Islands (1). This area receives some allochthonous material from adjacent seagrass beds during major storms (2), perhaps contributing to its low taxonomic similarity, but an alternate or contributing factor might be taphonomic inertia to a recent loss of seagrass from natural or anthropogenic processes other than AE, or perhaps greater time-averaging owing to intense, deeply penetrating burrowing by callianassid shrimp (A. I. Miller, personal communication). (C) Among the pristine open-shelf data sets, the two data sets with lowest J-C and rho are from the Amazon shelf (relictual sand habitat) and Yucatan (open sea habitat), which are from opposite ends of the nutrient-level spectrum and mesh-size spectra (0.5 and 3 mm); both shelves are tropical, wide, and had no bottom-trawling. The third nearby data set (J-C~ 0.55, rho ~ 0.4) is a firm muddy-sand habitat from the Amazon shelf. These three data sets thus do not distinguish themselves from the other pristine data sets in terms of physical environmental or methodological variables. (D) Among open-shelf AE1 data sets, the low-J-C, low-rho outlier is from sandy patches within a seagrass-rich area (Livorno); the death assemblage contains abundant dead specimens of seagrass-dwellers, probably reflecting a combination of small-scale postmortem transportation of grass-dwellers and short-term stochastic variation in sand-grassbed boundaries. This portion of the cross-plot is otherwise occupied only by AE >1 data sets.

1. Miller AI (1988) Paleobiology 14: 91-103.

2. Miller AI, Llewellyn G, Parsons KM, Cummins H, Boardman MR, Greenstein BJ, Jacobs DK (1992) Geology 20:23-26.

Table 1. Estuaries and lagoons with studies reporting the quantitative composition of both living and dead shelled mollusks, organized by extent of anthropogenic eutrophication (AE) at the time of sampling (year)

Total number of habitat-level data sets, with number of large data sets in parentheses (at least 20 live and 20 dead individuals after pooling all samples for a given sedimentary bottom type in that study area). Mesh size in mm, with only the first author identified; for full references including unpublished data, see SI Text. Other human modification in the study area, active either at the time of sampling or in preceding decades, but not necessarily directly impacting all habitats are as follows: SFa, artisanal shellfishing; SFm, mechanical shellfishing or bottom-trawling; Sol, solid sediment runoff; Dr, dredging and/or spoil dumping; Tox, toxic pollution; Sal, salinity change from inlet cutting or pumping. U.S. and Mexican states: CA, California; TX, Texas; GA, Georgia; MD, Maryland; QR, Quintana Roo; TB, Tabasco; VZ, Veracruz.

Table 2. Open shelf study areas and habitat-level data sets used here