Recent analyses have found that, despite high and increasing levels of community turnover, there is no clear overall trend in local-scale species richness1,2,3,4. However, it remains unclear how this result translates into functional changes. One of the most fundamental functional traits of a species is its size5,6 and there is an expectation that a warming climate will lead to a shift towards smaller species7,8,9,10,11, drawing upon metabolic theory12 and the observed distributional patterns described by Bergmann’s rule13,14. Temperature-driven shifts towards smaller species have been observed in tundra plant communities15 and some7,9,16, but not all11, aquatic systems. Furthermore, larger species have been more extinction prone during some previous mass extinctions17,18 and are more likely to show strong recent population declines19. Although relationships are threat dependent20,21, larger species tend to be assessed at a higher risk of extinction due to longer generational intervals and increased threat from habitat loss, fragmentation and hunting22.
One might therefore expect a detectable signal of shifts in community trait values beneath the apparent underlying consistency in taxonomic diversity. To examine this, we tested whether the size of a species is correlated with the change in abundance through time using the publicly available BioTIME database23. This database is the largest collection of time series of ecological communities and, despite considerable biases that we discuss below, has wide geographic and taxonomic scope23. It consists of ‘studies’ defined by a consistent sampling methodology and taxonomic focus. After cleaning and standardizing the names associated with the records, we linked six fundamental ‘size’ traits from four openly accessible trait databases representing four broad guilds: adult body mass from a database of amniote life history traits24, adult body length and qualitative body size of marine species from the World Register of Marine Species (WoRMS) database25, plant maximum height and seed mass from the TRY database26 and maximum body length of fish from a compilation27 based on data in the FishBase repository28.
Observations from single-location studies were combined, whilst widely dispersed studies were separately binned into a global grid of cells, each approximately 10 km wide, and data from each study and cell were treated as discrete assemblages, following previous analyses1,29. Selecting only assemblages with quantitative observations of ≥10 species, over ≥5 years and with ≥40% of the species having records for at least one size trait, we generated 12,956 assemblage time series from 144 studies (Fig. 1). This filtered dataset represented 2,109,593 observations of 10,286 species, of which 7,234 could be linked to at least one size trait (representing 84.02% of observations). Equally weighting studies, the average time series length was 18.2 years (range 5–71.8 years), and the average number of species per included assemblage was 65.4 (range 10–337). The log10 ratio between the largest and smallest species in each study averaged 2.49 (range 0.55–6.73) across the ‘mass’ traits and 1.06 (range 0.3–3.15) across the ‘length’ traits.
The aquatic realm is principally marine but includes three freshwater studies. Note that the locations are shown as the centre point of each study, which can cause oceanic studies to be ‘located’ on land. See Extended Data Fig. 1 for full details of study-level results.
For each trait and community assemblage time series for which there were sufficient data, we first square-root transformed and standardized each time series following previous approaches3 and calculated βi, the slope of a regression of abundance of species i against time. We then calculated, for each assemblage, τ (the Kendall rank correlation coefficient between the trait in question) and β, across the species for which we had trait data. This gives a non-parametric measure of whether larger species are more or less likely than smaller species to have increased through time and, importantly, can be calculated where trait values for only a fraction of the observed species are available. To weight each study within BioTIME equally, where there were multiple assemblages per study, these were averaged to generate a τ value for each possible study–trait combination. To provide a reference distribution against which to evaluate the statistical significance of this multistage analysis, we repeated the procedure with 10,000 trait randomizations within each assemblage.
Certain individual studies showed significant relationships between size traits and population trends (coloured dots in Fig. 2 and Extended Data Fig. 1). However, for five of the six tested size traits, the overall mean τ values did not differ significantly from the null model (Fig. 2). For one trait (amniote body mass, Fig. 2d) we found a marginally significant (unadjusted for multiple comparisons) overall average positive relationship between size and the slope of population trends (β). Alternative population data transformations gave highly concordant results (Extended Data Figs. 2 and 3). Possible confounding factors for the value of τ associated with each study, namely the total span of the time series, the number of sample points, the species richness, the range of traits in the assemblage, the average size trait completeness, the number of assemblages within the study, the grain of the study and the absolute latitude, did not consistently predict either τ or τ2 (Extended Data Figs. 4 and 5 and Supplementary Tables 1 and 2). Further, the likelihood of an individual species showing either a statistically significant positive or negative population trend was not linked to its relative size trait value within the assemblage (all P > 0.05; Extended Data Fig. 6 and Supplementary Table 3).
a–f, Distribution of Kendall rank correlation coefficient between body-size traits for body length (a) and qualitative body size (b) of marine species, maximum length of fish (c), adult body mass of amniotes (d) and seed mass (e) and maximum height (f) of plants versus changes in abundance through time. Each dot represents one study, averaging across the constituent assembly time series for studies of large spatial extent. Study-level results are binned into classes 0.05 units of τ wide. Coloured dots highlight studies that were individually identified as showing a significant trend (yellow for negative, blue for positive; see Extended Data Fig. 1 for study-level intervals). The error bar below each plot displays the distribution (central 95% and 66%) of mean τ values over 10,000 permutations of the size trait data, whilst the red line indicates the observed mean τ value within that panel. Displayed P values are calculated from permutation tests. Equivalent results using alternative approaches to transforming the community data are given in Extended Data Fig. 3.
These results indicate that there is not yet evidence for widely pervasive within-assemblage trends in a core functional trait, size. Importantly however, this study should not be seen as a refutation or diminishment of the heightened threats faced by the very largest apex species30,31, which constitute only a minor component of the BioTIME database. Rather, against a background of considerable turnover2,3 across whole observed community assemblages, on average, species positions in communities are being taken up by species of comparable size. Our results suggest that previously identified shifts towards smaller species found in some aquatic systems9,16 may not be as universal as currently expected7,11 and align with the divergent changes in global body-size abundance distributions observed between mammal guilds32 and the apparent stability of trait diversity in North American birds despite declines in abundance33.
The tendency towards an overall positive association between body-size and population trends across the amniote studies could have a number of drivers that would benefit from further investigation. One putative explanation that has been put forward for positive size trends is that anthropogenic dispersal limitations (generally considered to act more strongly against smaller species) may be having a greater immediate impact than climate change34. There are also indications of differences between terrestrial and marine systems. Previous work with the same datasets1,29 has found greater species richness and abundance changes in marine than terrestrial systems, whilst here we see a signal of greater trait changes in the (largely terrestrial) amniotes.
In our dataset, the fish length trait studies displayed a particularly skewed distribution of τ values (Fig. 2c), with a modal peak of studies showing small negative values then a tail of strongly positive relationships. This guild is also the most likely to have experienced sustained anthropogenic pressure35, and many of the ‘fish’ datasets in BioTIME include data from surveys of actively fished and managed areas. Accurately quantifying marine community trends is a challenge36,37, but this pattern could reflect the imposition or relaxation of anthropogenic pressure across marine systems38,39. Positive τ values could represent recoveries from past pressures on larger species, and positive τ values were associated with shorter study durations in the fish studies (Extended Data Fig. 4).
Our analysis necessarily sacrifices fine resolution for global scale. Technically, BioTIME studies represent assemblages defined by taxonomy and sampling protocol rather than complete ecological communities. We must implicitly assume that the scope of each study within BioTIME strikes a reasonable balance between the need to include a sufficiently diverse set of species to be able to observe any potential impact of trait differences whilst maintaining meaningful comparability. Limitations to total time series lengths and the limited range of sizes recorded within each dataset inevitably constrain our capacity to detect gradual changes or subtle influences of size. Although the lack of consistent study-level drivers of τ suggests that the results are unlikely to be solely determined by the inevitable spatial and temporal limitations of the BioTIME database, future work should seek to improve the scope and resolution of available data to enable more strongly parametric analyses and examine additional measures of community change.
Whilst available trait databases of amniotes and fish are carefully curated, checked and taxonomically tidy, the other databases pose more problems in terms of taxonomic matching and consistency of trait measurements. Without direct correspondence between the sources of dynamics and trait data, it is necessary to take traits as fixed values for each species, despite known differences in traits in time8,40,41,42 and space43 that may themselves represent responses to global change. However, in Celtic Sea fish, within-species shifts have been shown to contribute less to community-level size shifts than changes in species composition44. We also note that ‘size’ traits for indeterminately growing plants have a less clear meaning than for animals. However, both seed size and maximum height are linked to environmental variables45,46, plant size is linked to life history47,48 and changes in community height driven by species turnover have been observed in tundra plants15.
Many of the criticisms and defences regarding earlier studies using the BioTIME dataset, and indeed other analyses of large collections of time series, also apply to this work49,50. The consistency between the alternative approaches we tested to determine population trends (Extended Data Fig. 3) demonstrates that our conclusions are not dependent on particular data transformation choices. However, a largely non-parametric statistical approach was necessitated by the unevenness of the available data, and it must be noted that it could lack the power and resolution to identify subtle changes. Biases in the underlying BioTIME database towards vertebrate taxa, particular biomes and temperate North American and European sites23 are further exaggerated when crossed with trait data availability (Fig. 1). One particularly concerning gap is the absence of any insect studies in our dataset due to a paucity of usable trait information. Observations suggest that there have been considerable changes in the structure of insect communities34,51,52. Developing comprehensive insect trait datasets, including using proxies and data imputation, will be crucial to address this deficit53,54,55.
In conclusion, despite necessary reservations, this global analysis suggests that examples of relative increases of larger species11,34 may in fact be as frequent as shifts towards smaller-sized species16. Community responses appear to be considerably more nuanced and localized than previously considered based on theoretical macro-ecological expectations7.
Source: Ecology - nature.com