1. Introduction

Our planet is undergoing rapid changes, including global climate modification, large scale habitat conversion, overexploitation of animal populations and unprecedented levels of pollution [1]. All these changes have two major outcomes on the demographics of living organisms: the decline of common species and the extinction of rarest ones. For example, the Living Planet Index (LPI) tracks over 16,700 animal populations across the world, many of which are exploited populations. The 2018 LPI report emphasizes the decline of demographic indices by 60% on average across species over the last 50 years ([2], but see [3] for a more nuanced analysis). Similarly, Rosenberg et al. [4] estimated that since the 1970s, the North American avifauna has lost over 3 billion birds, which corresponds to a decrease of 29% of its overall abundance, predominantly in common species. At the other end of the dominance-diversity spectrum, the IUCN Red List has been developed to monitor global trends in species extinctions, while acknowledging the increasing efforts deployed in species assessment [5]. Global analysis of this list shows that, in all taxonomic groups, the risk of rare species extinction is constantly increasing since the 1980s [e.g. 6, 7]. The rate of biodiversity erosion is so high that many scientists are now calling the current situation “the sixth mass extinction” [8].

Although both population decline and species extinction are ultimately rooted in similar causes, practitioners often face trade-offs when choosing where to prioritize resources. For instance, a large-scale study quantified the opportunity costs of conserving rare species of migratory fishes in the context of removing weirs and increasing habitat connectivity across thousands of tributaries [9]. Prioritizing projects to maximize benefits for the rarest species led to the poorest average habitat gains for other species, especially common ones [9]. Another study using abundance data of 144 bird species reported that European bird populations declined by 50% between 1980 and 2009, while the abundance of less-dominant species increased over the same period, presumably as the result of targeted conservation programs [10]. These counterintuitive results suggest the existence of a dominance-diversity trade-off in some situations, where the conservation of rare species parallels the decline of common ones and vice versa.

The dominance-diversity trade-off is also pervasive in the structure of ecological communities. Species assemblages share a common structure characterized by the presence of a few dominant and many rare species [11]. One simple way of representing this structure is to rank species according to their relative abundance in the community (rank abundance distributions; RADs). Over the last decades, ecologists have proposed several abundance distribution models [see 11], which range from resource apportionment models (e.g., dominance pre-emption and MacArthur random fraction) to purely semi-parametric distributions (e.g., log- and geometric-series, lognormal and Zipf-Mandelbrot). Many studies have attempted to identify which of these models best captures the structure of ecological communities, often with mitigated success [12–14]. Although abundance distribution models are tailored to reflect different ecological processes, their shared mathematical structure suggests the existence of a conservation compromise, where managing for abundant species and species richness cannot be achieved at once.

The objective of this study is to explicitly demonstrate how RAD models are mathematical representations of the dominance-diversity dilemma. Specifically, we evaluate whether RAD models can successfully predict animal species richness from the relative dominance of the most abundant species in thousands of communities across a range of taxonomic groups. We then discuss the implications of the dominance-diversity dilemma for stock management and biodiversity conservation.

2. Methods

2.2 Dataset preparation

The BioTIME dataset [23] contains a large collection of ecological communities across the globe and aims at evaluating temporal trends of biodiversity in ecological assemblages. To explore the idea of predicting the species richness of a community through the relative dominance of its most common species, we downloaded the full BioTIME dataset on December 12 2019 and filtered it to keep only observations from animal communities (see Table 1 for a per taxa breakdown). We excluded plants because Nmax and Ntot are ill-defined when biomass or cover are used as abundance measures. Similarly, we eliminated communities in which Nmin and Nmax were equal (i.e. maximum evenness) because the GS model is not properly defined for these limit cases.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Per-taxa breakdown of community data and bias assessment metrics.

https://doi.org/10.1371/journal.pone.0283439.t001

RAD models assume that species interact and share a common set of resources. To prevent the inclusion of communities that are mainly structured by environmental heterogeneity rather than competition for resources, we restricted our analyses to studies conducted with a spatial grain less than 1 km². We acknowledge that this cut-off is large and that processes other than resource partitioning could still cause some of the observed variation in species richness. Nevertheless, we assume that the authors of the original studies identified the proper spatial grain for their target organisms. Finally, we used the most recent data point of each survey, resulting in a final dataset of 4,375 independent local animal communities. We hereby adopted an inclusive definition of a “local community”, which is a collection of individuals from different species that share (i.e. are sampled in) the same area, over the same period, using a common sampling protocol. For each community, we calculated the relative dominance of the most common species and used it in combination with total abundance to predict SR as per Eq 2.

3. Results

The general dominance-diversity dilemma is illustrated by simulating SR from Eq 2 using different values of MaxRel and Ntot (Fig 1). When the number of sampled individuals (Ntot) is high, SR decreases with increasing MaxRel. At lower values of Ntot, the relationship between SR and MaxRel becomes increasingly flat, indicating that both the community size and the sampling effort affects the prediction of SR (Fig 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Theoretical relationship between species richness and dominance in a community.

SR is the species richness value, MaxRel represents the relative abundance of the most common species and Ntot represents the total number of individuals in the community. Simulations were conducted using Eq 2 and a grid of parameter values where MaxRel ranged from 0.05 to 0.95 and Ntot ranged from 5 to 100 individuals. Communities where dominant species had less than 2 individuals were removed because these communities could not be dominated per se.

https://doi.org/10.1371/journal.pone.0283439.g001

We predicted the SR of 4,375 local communities from around the world using only the relative dominance (MaxRel) and the total number of sampled individuals (Ntot). The relative dominance across all datasets and communities ranged from 0.061 to 0.995, with the median at 0.448. The inverted GS model successfully predicted SR in actual communities, as assessed by the squared correlation between log-predicted and log-observed SR (pseudo-r² = 0.69; Fig 2A). By comparison, the pseudo-r² of the log-log empirical relationship between MaxRel and observed SR was much lower at 0.20. Although our predictive approach presented a reasonable fit to the observed data, the global error magnitude is nonetheless relatively high, with a MAE of 1.66. Using a comparable dataset (i.e., with only communities containing at least one singleton species, n = 3,665), the pseudo-r² of the LS model is 0.82. The pseudo-r² of the log-log empirical relationship between the number of singleton species in a community (Fisher’s α) and observed SR was also much lower at 0.60. Inversion of GS or LS models always produced better approximations of SR.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Relationship between the observed species richness and model predictions in 4,375 animal communities.

With the raw predictions from a geometric-series model (a) and after applying an empirical correction factor (b). Dashed line is the 1:1 relationship.

https://doi.org/10.1371/journal.pone.0283439.g002

As expected from its steeply declining nature, the GS model fit was systematically biased (Fig 2A). On average, the inverted RAD model underestimated SR (MBE = 1.56), especially for species-rich communities. We found that multiplying log (SR) by a correction factor (cf) of 1.20 (+/- 0.004 SE) both reduced error magnitude (MAE* = 1.45) and overall cancelled the systematic bias (MBE* = 1.04) (Fig 2B, see Table 1 for a per taxa breakdown). This empirical cf was obtained at no extra cost of input variables and can be factored in Eq 2 directly. However, systematic bias > 30% could still be detected for fish and benthos, even after applying the cf.

4. Discussion

RAD models are frequently used to describe the structure of ecological communities, but their mathematical underpinnings hide additional ecological insights. We showed that the inversion of the simple GS model can be used to approximate SR with a minimum of information on the size of the community and the relative abundance of the dominant species. Consequently, our results suggest the existence of an intrinsic trade-off between dominance and diversity (i.e. the long tail of locally rare species in RADs), irrespective of taxa or sampling protocols. Hints about the existence of that trade-off have been reported before, either directly for fish and lizard communities [25, 26], or indirectly for foraminifera [27]. However, in all the above studies, dominance was treated as a reciprocal measure of biodiversity, and not as a driving force, such as implied by the structure of inverted RAD models.

Moreover, inversion of the GS model better predicts SR than simply regressing it against dominance. Indeed, the pseudo-r² of the model inversion approach is 69% in comparison to 20% for the empirical regression between SR and dominance. However, with 45% error on SR estimates, inversion of the GS model does not provide the level of precision needed for surveying individual animal communities. For example, the SR of a 20-species community predicted by the inverted GS model (including the correction factor) could fall anywhere between 12 and 23 species (first and third quartiles for 113 communities of exactly 20 species; Fig 2B). Furthermore, considerable bias asymmetry persisted for some taxonomic groups like small mammals and some aquatic organisms, which would warrant a more in-depth analysis. Factors such as the type of sampling protocol, the regional species richness, or animal behaviour could strongly influence the predictive ability of the inverted GS model.

Approximation of SR through the LS model slightly outperforms the inverted and corrected GS model. This result was expected since the LS model requires more data to parameterize, as all the singleton species in a sample must be identified to predict SR. By comparison, the parameters of the GS model only require the relative abundance of the dominant species. From a conceptual standpoint, the GS model captures the iterative process of resource pre-emption by competing species, whereas the LS model is often regarded as a purely statistical representation of the community (but see [18] for a process-based interpretation of the LS model).

Inversion of the GS model suggests that factors affecting the size of the community, irrespective of dominance, also have an influence on SR. For a given dominance value, increasing the abundance of the entire community (e.g., through habitat restoration) should increase SR, whereas decreasing it should decrease SR (Fig 1). Our approach therefore decouples the effect of relative dominance among species from the effect of total abundance. The dominance-diversity trade-off reveals itself only by altering relative dominance per se. Consequently, one can predict that selective harvesting of dominant species should increase SR, whereas indiscriminate harvesting of individuals across the entire community should decrease it.

The dominance-diversity trade-off is in line with current knowledge of invasive species management. Release of exploitative pressure on invasive species upon their introduction in a new location may induce dominance (aka enemy-release hypothesis, [28]) and cascade negatively on the SR of host communities [29]. The inverted GS model supports this mechanism and provides testable predictions of species losses following the introduction of a dominant invader. Conversely, the approach could also be used to compare different management scenarios, in order to prioritize interventions in locations where eliminating or controlling an invasive species might have the largest positive effects on biodiversity.

Implications of the dominance-diversity trade-off for the exploitation of natural populations are less intuitive. For instance, halting the harvest of an abundant fish stock may help restore densities and increase the long-term viability of that population. Although one might expect that the whole community should benefit from such no-catch policies, the inversion of the GS model suggests that, overall, one might see a decrease in SR in those areas due to increased dominance. This could explain why some no-catch marine reserves reported negative effects on SR in comparison to nearby exploited communities, despite good protection levels and ongoing habitat alterations outside reserves ([30], negative effects are reported for 11 out of 39 studies in S1 Table therein).

Conservationists may be tempted to propose harvesting the most abundant species in a community to support SR in exploited ecosystems. And indeed, exploitation or experimental removal of abundant species have been shown to increase the diversity of local communities, especially if leftover resources become available to rare species [31–33]. However, in a hyper-connected world, there is a real possibility that the resources freed by such a harvest would allow another opportunistic species to enter the community and reinforce, or even worsen, the dominance pattern. Also, it is important to emphasize that most exploitation methods, especially commercial ones, usually have devastating consequences on habitats as well as on rare species through bycatches. For example, marine fishing operations often drastically reduce habitat complexity, alter productivity and remobilize contaminants and fine particulate matter [34, 35]. Therefore, the theoretical benefits of a selective harvesting program on animal SR are easily offset by the destruction of habitats and the precariousness of vulnerable populations. Nevertheless, the existence of the dominance-diversity trade-off highlights that gaining or losing individuals in a community will not have the same effects if these changes affect a few species only or the entire community.

The shared mathematical structure of RAD models could be worded as follows: “It is common to be rare”. The generality of this maxim in ecological communities, but also potentially in economic and social systems, has profound implications for the management of ecosystems. It underlines the presence of pervasive inequalities among resource users, which are only reshuffled by changing the environmental context, and thus, the rules of the apportionment game. It reminds us that strategies aiming solely at exploiting the natural capital, or at conserving species, are doomed to fail. Instead, protection of biodiversity could be promoted by encouraging a broad range of resource exploitation and socio-economical systems, each with its own set of rules.

Supporting information