Fig 1.
Distribution of biotic homogenization and differentiation across surveyed continental shelf regions.
(a) Map of temporal trends in spatial Jaccard dissimilarity by region. (b-d) Changes in spatial community composition over the study period are visualized using non-metric multidimensional scaling (NMDS) for example regions that experienced differentiation (b; West Coast South Island, New Zealand; first = 79 tow locations, last = 65 tow locations), homogenization (c; Southeast United States; first = 77 tow locations, last = 87 tow locations), and no significant trend (d; Iceland; first = 528 tow locations, last = 529 tow locations). Example regions are labeled in (a). Each point in subfigures (b), (c), and (d) represents the species composition of an individual sampling event, with points outlined in white for the first survey year and in black for the last survey year. The NMDS ordination arranges sampling events based on their species composition and Jaccard dissimilarity, such that points closer together represent more similar communities. To aid visualization of community shifts, ellipses enclose 95% of a multivariate t-distribution fitted to the first-year (dotted line) and last-year (solid line) samples. A contraction of the ellipse suggests homogenization, while an expansion suggests differentiation. Basemap from Natural Earth [71], and map rendered using the sf and ggplot2 packages in R [72–74].
Fig 2.
Trends in spatial beta diversity over time.
(a) Annual Jaccard dissimilarity for each region (colored points, n = 705) with generalized additive model (GAM) smoothers for each region (colored lines) and 95% confidence intervals (colored ribbons). A decrease in dissimilarity represents homogenization (yellow); an increase represents differentiation (pink). A lack of significant trend is shown in blue. The average linear trend across surveys (black line with 95% confidence interval in gray) is also plotted from a linear mixed effect model with a random slope and intercept for survey. (b) Coefficients and associated standard error of dissimilarity versus time for each survey from a linear model (LM) with a fixed effect interaction between survey and time in ascending order by coefficient value. Point size represents the length of the survey period. Asterisks mark surveys for which dissimilarity through time was better described by a non-linear GAM than a LM.
Fig 3.
Average linear model coefficients predicting annual Jaccard dissimilarity for all regions (n
= 32 surveys). Coefficients were allowed to vary by region for temperature (a) and relative fishing catch (b), but not for other characteristics (c). All variables were centered and scaled across all observations except for fishing catch, which was centered and scaled within each region. Therefore, coefficients are in units of dissimilarity over units of standard deviation of the covariate. Coefficients for which the standard error did not cross zero are in black and others in gray. Season is not displayed as it was included in the model as a factor.