What's hot and what's not: Making sense of biodiversity 'hotspots'

Abstract

Conserving biogeographic regions with especially high biodiversity, known as biodiversity 'hotspots', is intuitive because finite resources can be focussed towards manageable units. Yet, biodiversity, environmental conditions and their relationship are more complex with multidimensional properties. Assessments which ignore this risk failing to detect change, identify its direction or gauge the scale of appropriate intervention. Conflicting concepts which assume assemblages as either sharply delineated communities or loosely collected species have also hampered progress in the way we assess and conserve biodiversity. We focus on the marine benthos where delineating manageable areas for conservation is an attractive prospect because it holds most marine species and constitutes the largest single ecosystem on earth by area. Using two large UK marine benthic faunal datasets, we present a spatially gridded data sampling design to account for survey effects which would otherwise be the principal drivers of diversity estimates. We then assess γ-diversity (regional richness) with diversity partitioned between α (local richness) and β (dissimilarity), and their change in relation to covariates to test whether defining and conserving biodiversity hotspots is an effective conservation strategy in light of the prevailing forces structuring those assemblages. α-, β- and γ-diversity hotspots were largely inconsistent with each metric relating uniquely to the covariates, and loosely collected species generally prevailed with relatively few distinct assemblages. Hotspots could therefore be an unreliable means to direct conservation efforts if based on only a component part of diversity. When assessed alongside environmental gradients, α-, β- and γ-diversity provide a multidimensional but still intuitive perspective of biodiversity change that can direct conservation towards key drivers and the appropriate scale for intervention. Our study also highlights possible temporal declines in species richness over 30 years and thus the need for future integrated monitoring to reveal the causal drivers of biodiversity change.

Keywords: Random Forest analysis; biodiversity; biodiversity hotspot; conservation; diversity partitioning; marine benthic fauna; rarefaction and extrapolation; species richness.

FIGURE 1

The top 10 marine benthic ‘hotspots’ (large points) plotted over spatial estimates of γ‐, β‐ and α‐diversity across the UK EEZ using the combined dataset. Values are based on means where multiple temporal observations exist

FIGURE 2

Top panels: rarefaction of marine benthic assemblages used to estimate γ‐diversity for years 1990, 2000 and 2010. Crossing curves demonstrate that the sites rank‐order based on richness is not conserved as the number of sample cells (i.e. spatial units) increases, and this feature is consistent over time (n = 50 cells were randomly selected in 2000 and 2010 where there were data for >50). We highlight areas which may have relatively low to intermediate species richness based on a low number of sample cells (black lines) compared with other areas (e.g. blue dotted and green dashed lines) but, because of higher dissimilarity between local assemblages, tend to have some of the highest species richness values at larger scales. Based on these results, we would draw contrasting conclusions about which area was most diverse depending on whether we looked at 1, 20 and 60 sites within a region. Bottom panels: maps show differences in spatial data distribution between respective years (red = data collected in that year, blue = all data)

FIGURE 3

Variable importance based on node impurity ordered along the y‐axis from most important (top) to least important (bottom) following Random Forest analysis on the combined data. The suffix ‘df’ represents a variable's heterogeneity based on mean pairwise differences across selected sample cells within a 25 km radius

FIGURE 4

Partial dependence plots showing model predictions (red line) of γ‐, β‐ and α‐diversity (y‐axis) using the combined dataset in response to the six most important covariates (x‐axis) as determined by node impurity (Figure 3) while keeping other variables fixed at their average values. The suffix ‘df’ represents a variable's heterogeneity based on mean pairwise differences across selected sample cells within a 25 km radius

FIGURE 5

Observed versus partial dependence plots for full model and survey effects model estimates of temporal diversity. Change was partly captured by our environmental and survey covariates, revealed by the more limited change related to ‘Year’ in the full models, followed by the survey effects models, with most variation in our observed values

FIGURE 6

Density plots showing the distribution of pairwise and 25 km turnover estimates (i.e. β‐diversity) between cells sampled within the same year using the combined and BM datasets, respectively. Values were mostly intermediate, rather than bimodally distributed between extremes, indicating that many species were shared between unique multi‐species observations and thus individualistic processes were more prominent in general. Data have been split across UK regions (Figure S1) to show that this pattern was largely conserved through space