Temporal changes in local species richness: going up, going down, going straight

A recent heated exchange emerged on the dynamic of local species richness. In other words if you track the number of species, say plants, in a specific meadow over 20 years, does this species number go up, go down or stay the same?

Until recently we had no information on these dynamics, most studies to this date have been using so-called space-for-time design to mimic the temporal dynamic. This can be illustrated as comparing the species richness of a protected meadow vs the species richness of a nearby city that was previously a meadow. The key points there is in finding proper reference points for local habitats that experienced changes or are currently under pressure. Two recent influential meta-analaysis (Vellend 2013 and Dornelas 2014) looked instead for studies that tracked the temporal dynamics of species richness. Both studies compiled a large number of sites worldwide and found no detectable changes, on average, in local species richness over time, this goes against some of the current crisis narrative of biological conservation and so sparked quite some heated debate.

In this (long) post I would like to retrace the main points of the papers in the chronological order of their publications and end up with some of my opinions on the topic.

A paper timeline:

Vellend et al 2013; in this meta-analysis Vellend and co-authors gathered a dataset of over 16’000 unmanipulated vegetation plots surveyed at least two time with a minimum timespan of 5 years. From this they computed log response ratio ie log (richness for the last record / richness for the first record) standardized by the study duration. Their main result is that across all these studies local plant species richness did not show any trend. Aware that this result could be miss-interpreted by some, Vellend et al state that it does not contradict concerns of global loss of species. They discuss one of the implication of this result, that is, using small-scale diversity experiments showing a relation between species richness and ecosystem functioning to justify conservation actions is contradicted by their results. It is not so much that such experiments are pointless rather that the conclusions from them have little applied and/or conservation relevance (but see Srivastava and Vellend 2006).

Dornelas et al 2014; in this meta-analysis Dornelas and co-authors gathered 100 time-series of local communities of plants and animals in terrestrial and marine system measured in at least three different years. Their main analysis is fitting separate regression lines for each time-series linking richness or Jaccard similarity with time. Their main result is that despite showing quite some variation in species richness trends, on average no overall trend was present in the data. On the other hand, Jaccard similarity declined with time. Basically species richness stay the same but which species are present in each local communities is shifting with time. Again the authors state that these results do not contradict global pressure on species and ecosystems that leads to a global loss of species, they highlight the need to study temporal variation in species composition.

McGill et al 2015; in this review McGill and co-authors argue that behind the simple narrative of biodiversity decline the empirical evidence are more complex, diversity taking up different forms and being scale-dependent. They review the empirical evidence behind 15 forms of biodiversity trends, such as the global decline of species richness, or the increase in temporal turnover at local scale. They conclude with a plea to go out there and collect more data as little informations is available on the trends of most of the biodiversity forms. But also to develop quality values for biodiversity: “more rats and jellyfish and fewer lynxes and tuna would be consider by many as a decrease in quality”, which will be an tricky exercise as many value system are in place or can be developed and the merits of one over some other will be subjective.

Gonzalez et al 2016; in this critique Gonzalez and co-authors raise three main issues with the analysis and conclusions reached in the Vellend 2013 and Dornelas 2014 meta-analysis. First of all, they argue that the studies used in the two meta-analysis are biased towards certain regions of the globes that are not representatives of global species richness and of the intensity of human pressure on natural systems. They argued that these meta-analysis are biased towards areas that are recovering after human disturbances where increase in diversity or compositional shifts are to be expected. Also, they argue, contrary to arguments developed in Vellend 2013, that tracking species richness is relevant even in sites that underwent drastic conversion, such as a forest being converted into a garden. The second critic raised in this paper is that most of the time-series included in the analysis are of rather short duration and they showed with simulations that short time series can miss known trend in species richness. In addition, regressing the log response ratio against study duration, they found that longer time-series were more likely to show decline in species richness than shorter time series. The final critique raised in this paper is that estimates of biodiversity trends need appropriate historical baseline. Without reference site log-response ratio will be positive in the phase of a recovery from a disturbance while still being lower compared to the original conditions. Re-analyzing the 2013 data with two modifications they showed that if time-series that included recovery were eliminated significant decline of species richness between 1 and 11% were detected. Gonzales and colleagues finish the paper by cautioning against overextending results in scientific studies and by outlining the need for systematic biodiversity monitoring especially in areas where little information is available.

Vellend et al 2017; in this paper Vellend and co-authors respond to the critics raised by the Gonzalez paper. First they show that shorter time-series do not provide, in general, biased estimates of log-response ratios, correcting for overlap and duration reveal that variance is indeed higher for log-response ratios estimates in shorter time-series but no systematic bias is present. Vellend and colleagues attribute the findings of Gonzalez to two distinct errors in the computation and in the simulation of that paper. Then the authors showed that the reported decline in log-response ratio with study duration in both the Vellend 2013 and Dornelas 2014 dataset was depending on assuming non-zero log-response ratio at the beginning of the study, setting the intercept of the regression to 0 led to no-significant results. Also, including new data or removing an outlier data point removed the relation. In this paper Vellend and colleagues also show that removing studies tracking communities under recovery did not lead to decline in local species richness when removing the modifications introduced by Gonzalez. They also argue that disturbances do not always lead to a decline in species richness as shown by other studies focusing on disturbances (such as Supp and Ernest 2014). On the historical baseline, Vellend and colleagues recognize that local richness might have increased or declined before the start of the data collection, but as they have no information in this regard they argue that with the data currently at hand it is not possible to further account for these effects. Finally, they also recognize that the dataset in the two original publications are spatially biased, and that future analysis with more complete geographical coverage might provide different answers. Yet, they point to the fact that this spatial bias will affect all global meta-analysis.

Hillebrand et al 2018; the main point raised by Hillebrand and co-authors in this papers is that inadequate tools that have been used to quantify anthropogenic impacts on natural systems. They argue that species richness is a poor biodiversity metric with numerous statistical and ecological issues, a point that has been repeatedly raised by different authors. They propose to focus on compositional shifts and introduce two new metrics to measure it. They explore richness and compositional turnover trends to three datasets and found little changes in species richness despite important colonization and local extinction dynamics. They also found an accumulation of compositional differences with increasing temporal distances between samples in two out of the three datasets. They argue that species richness trends provide little information for assessing biodiversity changes and offer three recommendations for future monitoring programs: (i) measure multiple aspects of diversity, (ii) develop coordinated monitoring programs that measure biodiversity at multiple locations in multiple times, (iii) focus monitoring program to ensure long-term consistency.

Cardinale et al 2018; in this paper Cardinale and co-authors aim at providing a summary of the debate about temporal trends in local species richness towards conservation practitioners. They focused on four papers: Vellend 2013, Dornelas 2014, Gonzalez 2016 and Vellend 2017. They repeated some of the main critics raised in Gonzalez 2016: geographical bias and missing historical baselines. They argued further that the original meta-analysis, by collating very different datasets collected with different purposes and designs but also from systems having experienced different human disturbances in different years in the past, will lead to flawed conclusions. Basically the meta-analysis are averaging over stuff that should not be averaged over.

My take on the debate:

Each of these individuals papers could make a blog post on its own, I will sum up my take on the current state of the debate in 4 points: (i) a timely discussion, (ii) the issue of missing baselines, (iii) the BEF pandora box, (iv) double-standards and scientific honesty.

A timely discussion

One can have different opinions on the Vellend 2013 and Dornelas 2014 meta-analysis, yet all could agree that these papers sparked very important discussions within the ecological community forcing us to realize that species richness trends are scale-dependent. In the wake of these papers, the McGill 2015 review is great in ordering biodiversity trends into 4 metrics (biomass / abundance, richness, temporal and spatial turnover) and exploring the evidence currently available on the trends of these different metrics at different spatial scales. The result is sobering: we know very little about current biodiversity trends and constant species richness can hide dramatic changes in species composition. This last point is taken up in the Hillebrand 2018 paper where the authors repeat some of the long-standing arguments on the limits of species richness as a diversity metric. So scientists should communicate this gap to monitoring agencies. Several science-policy initiatives could take up that challenge, I think of the GEO BON and IPBES for instance which uses broader definitions of biodiversity than community ecologists (they include species distributions or ecosystem structure in there) but are the right actors to transmit these gaps and needs. To sum up, Vellend 2013 and Dornelas 2014 meta-analysis may be perfectible but they started an important discussion within the scientific community on the scale-dependence of biodiversity trends and the very thin evidence that we currently have.

The issue of missing baselines

To my mind, the main critical point of the Vellend 2013 paper is the absence of relevant historical baseline to derive the log-response ratios. In the meta-analysis the authors used the first year of records as the reference point, if this was a “good” year for plants the resulting log-response ratio might be significantly different than if that first year was a “bad” year. I would argue against using log-response ratios for exploring temporal trends, this approach is well-suited for experimental data where there is a clear control and a clear treatment data point, but in the case of temporal richness dynamic for observational plots, what should be the control is very hard to determine. I would rather see something like weather forecaster do, average diversity values over a period of reference that might be fixed (the onset of widespread fertilizer use in agriculture) or that might be moving (the last 2 decades excluding the current one ie 1991 – 2010). Then compare yearly values to this reference as in the picture below.

Of course this require standardized data sampling schemes which is totally absent for biodiversity these days, but I hope that scientists, governments and civil society will take up this challenge. Also I am not a fan of the approach used in the Dornelas 2014 paper of assuming a linear temporal trend in biodiversity. The linear approximation might be a first good guess and make sense to summarize many dataset but it is super stiff (only allow for straight lines). I would rather see some kind of hierarchical generalized additive model comparing the average trend with the variability present between the individual time-series.

The BEF pandora box

I think that the Vellend 2013 paper was particularly controversial because of its targeted criticism against the field of Biodiversity and Ecosystem Function research. The main argument in the paper is that since overall local species richness is neither going up nor down, the fact that BEF experiments show a link between local richness and ecosystem function is irrelevant for justifying conservation. This is not saying that such experiment are worthless, rather that they do not correspond to the species richness dynamic observed at the scale at which they manipulate species richness. In other words, the fact that there is mechanistic links between species richness and ecosystem functioning cannot be used to justify across all system and all species, conservation measures because species richness is temporally and locally stable. Therefore, based on BEF results ecosystem functions are also temporally and locally stable. The intention of the authors, as I get it, is to promote new BEF experiments manipulating not species richness but rather spatial and temporal turnover but also specific shifts in species traits via the introduction of new species. It seems to me that most of the critics of the Vellend paper on that particular point missed this argumentation. For instance, in Gonzalez 2016 and later Cardinale 2018 or Hillebrand 2018, the authors seem to have the impression that Vellend argumentation invalidate BEF experiments. This is not the point of Vellend argument, to my mind the point is rather that manipulating plant species richness might have led to important theoretical advances on the impact of biotic communities on ecosystem functions, yet to have practical relevance BEF experiment should manipulate some aspect of diversity that is indeed showing some temporal dynamic. The results from Vellend 2013 have many implications, it is the liberty of the authors to have chosen this focus on the practical relevance of BEF results. I think that it definitively contributed to the onslaught on the Vellend 2013 and Dornelas 2014 papers and to the actors of this onslaught. I hope that as a result of these papers future BEF experiment will target relevant scales for metrics that are indeed under changes due to human pressures.

Double-standards and scientific honesty

One of the main critic of Gonzalez 2016 on the Vellend 2013 and Dornelas 2014 papers is that the datasets included are spatially biased leading to an unrepresentative sample. Gonzalez 2016 argue that in such cases extra care must be taken as to avoid miss-interpretation in general newspaper. So basically since these results go so much against what we previously thought they should only be made public and visible once we have rock-solid evidence to support it. Vellend 2017 respond that this limitation apply equally to all ecological studies also to those fitting in the biodiversity crisis narrative and that current results are based on the best data available that will be perfected by further data collection. But until we see these new data from these undersampled places, the evidence is that local species richness show no temporal trends. Finally, Cardinale 2018 responded that Vellend 2013 and Dornelas 2014 gathered an heterogeneous set of studies basically comparing apple with oranges (or fat and skinny mice as they put it). This is a different critic that the one on the spatial bias, it is rather questioning the choice of studies made in the two papers and the control for potential sources of heterogeneity. Yet, I think that the results from the two incriminated meta-analysis are to the best of our current knowledge, future (coordinated) sampling will tell us if covering the current data gaps will lead to different outcomes. Plus the heterogeneity present in the dataset is a desirable one (why that is so would require a separate post). I am more concerned by some of the very harsh, unbalanced and sometime unjustified criticism that were raised against the two incriminated meta-analysis especially in Cardinale 2014 and Cardinale 2018. Reasonable doubt against unexpected results is a scientific strength but firmly holding on beliefs that are not backed up by any evidence (biodiversity is declining at all scale) is unethical. I might go a bit further, as we are called to take a more and more active role in communicating with the general public and policy-maker as advocates or experts, we should be careful not to fall into double-standard of gratefully accepting new evidence that go in the sense of the narratives that we are developing but not believing or dismissing evidence going against it. Ecology is complex, let’s try to convey that complexity and avoid over-simplification, one answer that will (almost) always be true in ecology is: “It depends”.

Further readings: