- Top of page
- Materials and methods
Recognition that loss of species may negatively affect ecosystem processes such as primary production or nutrient cycling has led to a rapidly growing literature on the effects of diversity loss over the last decade. In particular, a large number of studies have examined the effects of diversity loss using experimental systems in which species diversity could easily be manipulated (Schläpfer & Schmid 1999; Loreau 2000; Schmid, Joshi & Schläpfer 2002; Symstad et al. 2003). These studies have been criticized because they were based on the simplifying assumption of random species loss (Aarssen 1997; Huston 1997; Wardle 1999; Huston et al. 2000; Schwartz et al. 2000; Schmid et al. 2002; Schmid, Joshi & Schläpfer 2002; Smith & Knapp 2003; Symstad et al. 2003). Species extinctions in a local habitat are often non-random because not all species are equally likely to become extinct (Vitousek et al. 1997; Grime 1998, 2002; Stöcklin & Fischer 1999; Petchey & Gaston 2002). In natural or managed ecosystems, species loss is mainly related to habitat destruction, eutrophication, invasive species, climate change and harvesting (Chapin et al. 1997; Vitousek et al. 1997), the effects of which are likely to be species specific. Even stochastic extinction factors (Pimm, Jones & Diamond 1988; Hubbell 2001), such as habitat fragmentation, leading to small, isolated and therefore extinction-prone populations, may affect species differentially (Fischer & Stöcklin 1997). A particularly well-known case of non-random extinction is represented by previously species-rich mesic European grasslands, where intensified management has led to the removal of many competitively inferior species, while a set of high-yielding species has persisted (Fuller 1986).
Experimentally assessing the effects of non-random species loss resulting from land management, drought or other factors involves two distinct steps. First, the extinction pressure of concern must be applied to determine the post-extinction species assemblage. Secondly, ecosystem processes driven by the post-extinction assemblage must be compared with those of the pre-extinction assemblage in a common (post-extinction) environment. The latter could be achieved by examining the effect of recolonization into the impoverished assemblages or by replanting the two assemblages in a new common environment.
This approach is different from many biodiversity experiments, in which random species loss is assumed to have occurred already as communities consisting of the reduced species sets are established (Gonzalez & Chaneton 2002). It also differs from experiments in which a species richness gradient is induced by management or other environmental factors but resulting assemblages are not thereafter compared in common post-extinction environments (McNaughton 1977; Tilman & Downing 1994).
Random deletion of species, as simulated in previous biodiversity experiments, may thus over- or underestimate the effect of species loss on ecosystem functioning, depending on the relationship between extinction proneness and post-extinction performance of the species in a given extinction scenario and environment. In particular, under the sampling mechanism, if there is no correlation between species persistence and species performance in a specific ecosystem context, random extinction experiments will on average yield valid predictions. If the correlation is positive, effects of species loss will be overestimated. If the correlation is negative, effects will be underestimated. The magnitude of the effects of species loss because of a particular environmental factor can only be judged when extinction scenarios with their specific extinction sequences are imposed by applying this environmental factor. Subsequently, variables of ecosystem functioning can be examined in the post-extinction communities and environments. Although several recent studies factorially crossed biodiversity with nutrient–environment manipulations (Reich et al. 2001; He, Bazzaz & Schmid 2002; Craine et al. 2003; Fridley 2003), no study has so far attempted to estimate the effects of species loss induced by a transient management-related factor.
As previous biodiversity experiments have often been carried out in grassland systems and measured productivity as the variable of ecosystem functioning (Tilman, Wedin & Knops 1996; Hector et al. 1999; Pfisterer et al. 2004), we used this model system. We imposed an actual extinction scenario by simulating intensified grassland management in mesocosms. This is also relevant from an agricultural perspective. After several decades of intensification of grassland management across much of Europe, resulting in considerable species loss at the local scale (Fuller 1986), there is now a trend towards lower-input grassland systems. Therefore, the conditions affecting depauperate grasslands will no longer correspond with those that exerted the extinction pressure, while recolonization from a regional species pool may be limited (Loreau & Mouquet 1999; Schmid 2002). In our model, we therefore followed extinction management with ‘restoration’ management, as is now prescribed by agri-environment schemes in Europe (Ovenden, Swash & Smallshire 1998; Kleijn & Sutherland 2003). We investigated how non-random extinction processes affect plant production and whether these effects are smaller than those described for similar ecosystems in experiments with simulated random extinction.
We asked the following questions. Regarding characterization of the extinction scenario (a) are there species that are particularly likely to become extinct; (b) is extinction dependent on management intensity; and (c) is there an interactive effect of species identity and management intensity on extinction probability? Is extinction probability negatively related to productivity in monoculture? Regarding species loss effects, how does the biomass of species mixtures surviving after non-random extinction compare with the biomass of random mixtures of the same species richness drawn from the same original species pool? How does the management intensity in neighbouring communities affect recolonization and productivity during the restoration phase?
- Top of page
- Materials and methods
Transitory changes of land management in naturally assembling grassland ecosystems may lead to species extinctions, whereas recolonization may not always be effective at a short to medium time scale (Loreau & Mouquet 1999; Smith et al. 2000). This raises the question of whether the absence of any of the extinct species in a post-extinction environment affects ecosystem functioning, e.g. community biomass production. Frequently, the effect of species loss will depend on which of the original species disappear during the extinction phase.
Simulated extinction showed that management intensity increased extinction probability, such that species with greater biomass production had lower extinction probabilities. There was also species-specific variation in extinction probability, and there was a significant interaction between management intensity and species identity. This suggests that, after management-induced extinction, the remaining species are particularly productive ones, and this may reduce the impact of species extinction on ecosystem functioning. However, this argument would require a positive correlation between species persistence during the extinction treatment and species performance in the post-extinction environment, or at least an above-average biomass of a few highly persistent species, as we found (Fig. 5). Thus, because of a few important remaining species (Alopecurus pratensis and Trisetum flavescens; Fig. 5b), the simulated species loss would have been less damaging to ecosystem functioning in the restoration environment than in the random extinction scenarios. Based on the random extinction assumption, species loss from communities originally containing 15 species led to a reduction in predicted yield by 26–54% when four species remained, whereas in the non-random extinction scenario it only reduced predicted yield by 2–35% (the yield predictions for the case of one remaining species were 54% and 42–49%, respectively) (Fig. 4). These biodiversity–ecosystem functioning relationships had to be predicted, because the duration of our experiment was too short to actually assemble the post-extinction communities as derived from the estimates of extinction probabilities. What we could analyse within the duration of the experiment was the relationship between the observed monoculture yield and the observed yield for the total mixture of all 15 species (Fig. 2).
Figure 5. Persistence of the 15 species in the extinction environment (high-intensity neighbouring plots) and monoculture biomass (MC plots; relative to mixture) in (a) the extinction environment (correlation coefficient 0·169, P= 0·55) and (b) the restoration environment (correlation coefficient 0·312, P= 0·26; for species abbreviations see Fig. 2).
Download figure to PowerPoint
How do the observed and simulated effects of non-random species loss relate to ecosystem management? We distinguish between purposefully sown, tilled grassland and permanent grasslands subject to natural community assembly and ‘disassembly’ (Ostfeld & LoGiudice 2003). From the perspective of tilled grassland management, the potential for a yield higher than that of the highest-yielding monoculture may determine the choice of species mixtures. In our experiment, mixtures established from 15 species transgressed the mean yield of the best monoculture, Festuca pratensis, by 2% (in the restoration phase). From the perspective of the management of permanent, naturally assembling grasslands with shifts in management, however, an additional relevant effect of species loss concerns the lack of best-performing species (or species combinations) after management-induced extinctions. Persistence in the face of an extinction-inducing factor may be closely linked with high monoculture biomass in the extinction environment, but less so to high monoculture biomass in a changed post-extinction environment (Lolium perenne in Fig. 5). After the projected management-induced (non-random) extinction of all but one species and a subsequent shift to restoration management, the remnant species, Dactylis glomerata, would have yielded 13·1 g m−2, 49% less biomass than the best extinct species, Festuca pratensis, would have done in the same post-extinction environment (Fig. 5b). In a context with a transient extinction pressure therefore, non-random species loss may still have quite severe consequences.
The present study demonstrates that, in the context of species limitation in naturally assembling communities, non-random ‘sampling’ of species through transient environmental factors may be highly biologically relevant. Similar conclusions are also emerging for the diversity–productivity relationship in microbial communities (Hodgson, Rainey & Buckling 2002) and for effects of aquatic plant diversity on ecosystem variables in a wetland system (Engelhardt & Ritchie 2002).
The choice of management regimes is critical for subsequent correlations between species traits, the non-random sequence of species loss and the relation between species richness and yield for any given species assemblage (Petchey & Gaston 2002). We are only able to report a particular biodiversity–productivity relationship for a grassland system where the extinction pressure is fertilizer application and frequent cutting, followed by restoration management. Establishment of recolonization treatments through differential management of neighbouring communities, however, proved to be difficult. In spite of shallow soils and major treatment differences, the response of species richness was relatively inert, remaining above four species under all managements. Our impression is that the reason for this was not uncontrolled dispersal between experimental units. On the contrary, minimal colonization on bare plots suggests that experimental time scales well beyond the 3 years of the present study would be required to allow measurable effects of colonization on focus plot biomass.
Because of the limitations of our small-scale and short-period experimental approach, we are cautious in offering conclusions for grassland management based on the persistence–productivity relationships found in the experimental plots. Implications concern the restoration of European species-rich grasslands in the context of agri-environment schemes (Smith et al. 2000, 2002, 2003; Pywell et al. 2002, 2003). Recent studies have found that removal of extinction pressures is not always sufficient to enhance biodiversity in these grasslands (Pywell et al. 2003; Smith et al. 2003). Our results indicate that active assistance in plant recolonization may also be necessary to enhance plant production after shifts in grassland management. When, as in our experiment, mesic grasslands of the Swiss Plateau classified as Lolio perennis–Arrhenateretum elatioris (Dietl 1995) are turned into species-poor communities through fertilization and frequent cutting, and later subjected to low-intensity management, Festuca pratensis is a key ‘missing’ species that needs to be reintroduced if maximum production is an objective. Similar conclusions may emerge for other ecosystems and species that are subject to different extinction pressures and shifts in management and environmental conditions.
The more general conclusion concerns the differences between random and non-random species-loss effects on ecosystem functioning. Our results demonstrate that biodiversity–productivity relationships critically depend on the specific extinction process. The negative effects of the present scenario, of management-induced, non-random species loss, were less severe than would be predicted from random species loss. However, we do not know how robust this result is regarding other extinction-inducing factors, such as drought and heavy grazing.
To be relevant for ecosystem management, future research on the biodiversity–ecosystem functioning relationship needs to address further specific real-world extinction scenarios, post-extinction environments and species limitation–recolonization contexts (Gonzalez & Chaneton 2002; Pywell et al. 2002; Schmid et al. 2002; Smith & Knapp 2003; Pfisterer et al. 2004). Experimental studies on the correlation of species persistence in a variety of extinction contexts and species contributions to ecosystem functioning in post-extinction environments should become an important research focus. These studies should clarify the links between biodiversity and ecosystem functioning in the context of managed, naturally assembling communities. Random species loss represents a useful null model against which to contrast more realistic non-random species loss scenarios in future experiments.