Arguments for biodiversity conservation are often based on ecosystem services, but it remains unclear whether few8 or many9 species are needed to maintain ecosystem services. Determining how many species provide ecosystem services will require a synthesis of several areas of biodiversity research (Fig. 1). Biodiversity–ecosystem functioning studies have often considered a single functional context and found that multiple, but not all, study species promoted ecosystem functioning5,6,7 (Fig. 1a). We define a functional context (henceforth context) as the measurement of one function, at one time and place, under one environmental change scenario. Several related biodiversity studies have explored whether more species promote ecosystem functioning when more than one context is considered. For example, biodiversity–ecosystem stability (that is, the invariability of productivity) studies have found that more species are needed to provide ecosystem functioning at larger spatio-temporal scales because different species promote productivity at different times10,11,12,13,14 (Fig. 1b) or places15,16. Biodiversity–ecosystem multifunctionality studies have found that more species are needed to provide multiple functions because different species promote different functions14,17,18 (Fig. 1c). Biodiversity–global change studies have found that more species are needed to provide ecosystem functioning in a changing world because different species promote ecosystem functioning under different environmental change scenarios12. Here, for the first time to our knowledge, we consider all of these relationships together.

Figure 1: Some of the ways that biodiversity can be important for ecosystem functioning. Each of the eight symbols represents a species. Species shown in the bivariate plots are those that promoted ecosystem functioning within each functional context (for example, context A1 might be above-ground biomass measured during 2001). Although this figure defines contexts in two dimensions for simplicity, we considered four dimensions (Fig. 3). Previous studies have considered (a) one context, or (b, c) one dimension of contexts (b, ecosystem stability studies; c, ecosystem multifunctionality studies). Figure 2 tests the question in d, by comparing the one-dimensional overlap (for example, between A1 and A2) with the two-dimensional overlap (for example, between A1 and B2) and three-dimensional overlap (that is, a pair of contexts that differs in three ways; not shown). Our results reject each of the null hypotheses shown on the left in a–d. Full size image Download PowerPoint slide

We included data from 17 grassland biodiversity experiments that considered multiple times, places, functions or environmental change scenarios (Supplementary Table 1 and Supplementary Data). To test whether different species promoted ecosystem functioning during different years, we included studies that planted replicate plots (same species compositions) during consecutive years23 or made repeated measurements of ecosystem functions across years17,19,21,22,24,25. To test whether different species promoted ecosystem functioning at different places, we included studies that planted replicate plots (same species compositions, with one exception25) at multiple sites across Europe26 or multiple spatial blocks within a site17,23,25. To test whether different species promoted different functions, we included studies that measured several functions14,17,22, such as biomass production and nutrient uptake. Many of these functions are considered to be supporting ecosystem services because other types of ecosystem services depend on them4,27. To test whether different species promoted ecosystem functioning under different environmental change scenarios, we included studies that applied environmental change treatments, such as nutrient and CO 2 enrichment19, precipitation changes21 or land use changes such as livestock grazing22 and haying20.

We began by identifying the sets of study species that influenced ecosystem functioning within each context. Species were considered to promote ecosystem functioning in a particular context if they had effects in the direction that would usually be considered desirable from an ecosystem services perspective17. Positive effects were considered desirable for all functions except for soil inorganic nitrogen and light availability at ground level, where negative effects are consistent with lower levels of unconsumed resources17. We did not use separate definitions of desirable effects for different species (for example, positive effects of legumes on soil nitrogen might be considered desirable) to be consistent with previous studies17,18, to be conservative and because it may not be possible to manage simultaneously for both positive and negative effects. We found that approximately 27% of the study species promoted ecosystem functioning within any particular context, regardless of the size of the study species pool (Fig. 2a). Note that if many species were functionally redundant, or if only the most common species promoted ecosystem functioning, then we would expect a saturating relationship in Fig. 2a. Instead, our results suggest that even rare species can promote ecosystem functioning.

Figure 2: Sets of study species that promoted ecosystem functioning. a, The mean number of species that promoted ecosystem functioning within each context increased linearly (t = 16.40, P < 0.001, R2 = 0.944) with the size of the species pool, such that approximately 27% (mean, 95% confidence intervals for slope: 0.27, 0.24–0.30) of the study species promoted ecosystem functioning within each context. Error bars for each study indicate 95% generalized linear model confidence intervals. b, Different sets of species promoted ecosystem functioning in different contexts (overlap < 1), and overlap between pairs of contexts decreased as the number of differences between contexts increased (see Fig. 1). Symbols indicate means for each specific type of overlap; horizontal dotted lines show ± 95% permutation test confidence intervals; error bars for symbols and bars indicate 95% bootstrap confidence intervals. Supplementary Data indicates numbers of contexts for each study. Full size image Download PowerPoint slide

After identifying the sets of species that promoted ecosystem functioning in each context, we tested whether different sets of species promoted ecosystem functioning in different contexts. We used Sørensen’s similarity index to quantify overlap between species sets17. All comparisons were made within studies so that differences between pairs of contexts were not due to sampling from multiple species pools. First, we quantified one-dimensional overlap between all pairs of contexts that differed in only one way (Fig. 1b, c). For example, multi-year overlap was quantified between each pair of contexts that differed only in which year ecosystem functioning was measured (that is, same place, function and environmental change scenario in both contexts). A multi-year overlap value of one or zero would respectively indicate that completely identical or completely unique sets of species promoted ecosystem functioning during different years, independent of the other sources of variation. We found overlap values between these two extremes, which indicates that somewhat different sets of species promoted ecosystem functioning during different years, at different places, for different functions and under different environmental change scenarios (Fig. 2b).

After considering these sources of variation independently, we quantified multi-dimensional overlap between pairs of contexts that differed in two or three ways (Fig. 1d). Again, all comparisons were made within studies. We found that the average overlap between pairs of contexts decreased as the number of differences between contexts increased (Fig. 2b and Supplementary Fig. 1). This means that, for example, the identities of the additional species needed to provide one function during multiple years were not the same as the identities of the additional species needed to provide multiple functions during one year (Fig. 1d). Additionally, species sets did not simply vary independently of context attributes (permutation test P < 0.001 for one-, two- and three-dimensional overlap) (Fig. 2b). Thus, our results indicate that even more species will be needed to maintain ecosystem functioning and services than previously suggested by studies that have either (1) considered only the number of species needed to promote one function under one set of environmental conditions, or (2) separately considered the importance of biodiversity for providing ecosystem functioning across multiple years10,11,12,13,14, places15,16, functions14,17,18 or environmental change scenarios12,19,20,21,22. Future studies could more completely consider the consequences of biodiversity declines for ecosystem functioning and services by similarly considering the multidimensionality of ecosystem functioning both in experimental and natural communities.

Next, we quantified the extent to which the number of species promoting ecosystem functioning increased as more years, places, functions or environmental change scenarios were considered within each study. In other words, we quantified the accumulation of species across each of the four dimensions of contexts that we considered (Fig. 1). We found that a greater proportion of species promoted ecosystem functioning when more years, places, functions or environmental change scenarios were considered (Fig. 3). These relationships result from different species promoting ecosystem functioning in different contexts (Fig. 2b). Note that if the one-dimensional overlap values corresponding to each panel in Fig. 3 were one or zero, then these relationships would be horizontal or linearly increasing, respectively17. Our results are between these two extremes.

Figure 3: The proportion of study species that promoted ecosystem functioning increased when more (a) years, (b) places, (c) ecosystem functions and (d) environmental change scenarios were independently considered. Solid blue lines indicate generalized linear model fits for each study; dashed red line indicates grand mean generalized linear model fitted across all studies. Box plots summarize observed data: black band, median; bottom and top of boxes respectively correspond to lower and upper quartiles; error bars, ± 1.5 times the interquartile range. See Supplementary Data for the specific years, places, functions and environmental change scenarios considered in each study. Full size image Download PowerPoint slide

After comparing contexts within studies, data from all studies were combined to consider how the total number of species that promoted ecosystem functioning increased with the total number of contexts. We quantified the number of species that promoted ecosystem functioning in a random subset of all possible combinations of our observed contexts (that is, 100 pairs, 100 groups of three, etc.). The large increase in the number of species that promoted ecosystem functioning as more contexts were considered (Fig. 4) is the result of different species promoting ecosystem functioning during different years, at different places, for different functions, under different environmental change scenarios and in different species pools. Considering all of these factors together suggests that many species will be needed to maintain ecosystem multifunctionality at large spatio-temporal scales in a changing world. Consequently, the extinction (or decreased local occurrence) of almost any of these species is expected to decrease ecosystem functioning and services in at least some contexts. Further study is needed to identify the processes that explain why different species promoted ecosystem functioning in different contexts. The specific mechanisms involved probably differed across contexts, but previous results from these and other biodiversity experiments28 suggest that complementarity (in time, space, functional effect traits and functional response traits) is a general explanation for this pattern.

Figure 4: The number of study species that promoted ecosystem functioning increased with the number of contexts considered across all studies. The points are the number of species that promoted ecosystem functioning when 1–557 contexts were sampled from all 557 contexts. The dashed line indicates the total number of studied species (147), which restricts the upper limit for these values. The x axis includes variation across years, places, functions, environmental change scenarios and species pools. Full size image Download PowerPoint slide

Our results reveal new opportunities and challenges for prioritizing conservation efforts and predicting consequences of biodiversity declines. According to the precautionary principle, all species should be conserved because we cannot be certain which species actually provide ecosystem services29. Our results offer further support for the precautionary principle because most of the studied species were important at least once, and species exhibited context-dependent effects that will be difficult to predict. If it is impossible or impractical to conserve all species, then future studies could additionally consider how often (Supplementary Fig. 2 and Supplementary Data) and how much species influenced ecosystem functioning to determine which species are most important for maintaining ecosystem functioning and services. This will require careful consideration of many contexts, because it is not possible to make general predictions or conclusions by considering a few context-dependent phenomena30.

Future studies could determine whether some species consistently promote ecosystem functioning under environmental conditions that are currently common, or under environmental change scenarios that will probably become increasingly common. Note that even species that have small effects could be important for maintaining ecosystem functioning and services if they have a large cumulative desirable effect across many contexts. For example, Eriochloa sericea had the smallest desirable effect on above-ground biomass in the irrigated plots at the MEND Irrigation experiment during 2009, but promoted ecosystem functioning in 75% of the contexts in which it was included. Future studies could also determine which species promote ecosystem functioning in particular contexts that are highly valued by stakeholders. Note that even species that rarely promote (or often decrease) ecosystem functioning could be most important for maintaining ecosystem functioning and services within some contexts. For example, although Pascopyrum smithii only promoted ecosystem functioning in 2% of the contexts in which it was included, it promoted soil carbon more than any other species in the Cedar Creek Biodiversity experiment during 2004. Furthermore, note that declines in local diversity, which are far more common than global extinctions, will also decrease ecosystem functioning and services within some contexts. Finally, even the few species that never promoted ecosystem functioning in these studies (Supplementary Fig. 2) could promote ecosystem functioning in other contexts, or be a conservation priority for other (for example, ethical, aesthetic) reasons. Therefore, we encourage careful consideration of many contexts when making conservation decisions and predicting the consequences of biodiversity declines.