Biodiversity, Ecosystem Functioning, and Human Wellbeing An Ecological and Economic Perspective EDITED BY Shahid Naeem, Daniel E. Bunker, Andy Hector, Michel Loreau, and Charles Perrings 1 CHAPTER 2 Consequences of species loss for ecosystem functioning: meta-analyses of data from biodiversity experiments Bernhard Schmid, Patricia Balvanera, Bradley J. Cardinale, Jasmin Godbold, Andrea B. Pfisterer, David Raffaelli, Martin Solan, and Diane S. Srivastava 2.1 Introduction and a representative response at the ecosystem, community, or population level, were significantly 2.1.1 Two meta-analyses of biodiversity fl fi studies published in 2006 in uenced by several factors; the speci cs of exper- imental designs, the type of system studied, and the The study of patterns in the distribution and category of response measured. For example, biodi- abundance of species in relation to environmental versity effects were particularly strong when variables in nature (e.g. Whittaker 1975), and to the experimental designs included high-diversity species interactions (Krebs 1972), has had a long mixtures (>20 species) and in well-controlled sys- tradition in ecology. With increasing concern tems (i.e. laboratory mesocosm facilities). about the consequences of environmental change A second meta-analysis was conducted by for species extinctions, researchers started to assess Cardinale et al. (2006a) which focused on experi- the potential of a reversed causation: does a ments, published from 1985–2005, where species change in species diversity affect environmental richness was manipulated at a focal trophic level factors and species interactions, such as soil fer- and either standing stock (abundance or biomass) tility or species invasion? Manipulative experi- at that same trophic level, or resource depletion ments that explicitly tested the new paradigm (nutrients or biomass) at the level ‘below’ the focal started in the early 1990s and since then the level was measured. Cardinale et al. (2006a) used number of such studies has been increasing expo- log ratios of responses to characterize biodiversity nentially (Balvanera et al. 2006, Chapter 3). effects. Their analyses showed that species-rich In 2006, two meta-analysis papers were pub- communities achieved higher stocks and depleted lished which together provided the most compre- resources more fully than species-poor communi- hensive quantitative assessment of the overall ties, but that diverse communities did not neces- trends observed in manipulative biodiversity sarily capture more resources or achieve more experiments to date. Both studies showed that, on biomass than the most productive species in average, random reductions in diversity resulted monoculture. Cardinale et al. (2006a) also fitted data in reductions of ecosystem functions, but differed in from experiments to a variety of functional rela- the covariates examined. First, Balvanera et al. tionships, and found that experiments were usually (2006) analyzed studies published from 1974–2004. best approximated by a saturating function. The This meta-analysis showed that biodiversity effects, results from both meta-analyses were remarkably measured as correlation coefficients between some consistent across different trophic levels and measure of biodiversity (usually species richness) between terrestrial and aquatic ecosystems. In this 14 CONSEQUENCES OF SPECIES LOSS FOR ECOSYSTEM FUNCTIONING 15 chapter we present further analyses of the two low to high, in most cases a biodiversity effect can be meta-data sets, in parallel, and attempt a joint more specifically defined as a positive or negative interpretation. relationship between variations in biodiversity as the explanatory variable and a function as response variable. Thus, a positive diversity effect occurs when 2.1.2 The two meta-data sets used a relationship is positive and a negative biodiversity in this chapter effect occurs when a relationship is negative. The two meta-data sets assembled by Balvanera et al. (2006) and Cardinale et al. (2006a) are hereafter 2.1.3 Hypotheses referred to as B and C, respectively. Together, the two databases contain more than 900 published The goal of meta-analyses of biodiversity–ecosystem effects of biodiversity on ecosystem functioning functioning experiments is to assess to what extent (Schmid et al. 2009, Cardinale et al. 2009). In B, these biodiversity effects reported in single studies can be effects were extracted directly from the publications generalized across different design variables, sys- and therefore rely on the analysis (assumed to be tem types, and response categories. Ideally, correctly executed) carried out by the original hypotheses about variation between studies should authors. In more than half of the cases, the extrac- be derived, a priori, from underlying mathematical ted biodiversity effects were correlation coefficients theory about mechanisms responsible for biodiver- (Balvanera et al. 2006). For these, and for additional sity effects. In practice, however, it is often only cases, significance, direction, and shape of the possible to look for patterns in variation of biodi- relationship between biodiversity and each versity effects and then develop explanatory response variable could be extracted. In C, the hypotheses in retrospect. This is primarily due to mean values of response variables were available the fact that the majority of biodiversity experi- for each level of species richness. This allowed the ments included in our meta-databases focused on authors to decide whether a linear, log-linear, or demonstrating biodiversity effects rather than saturating curve (Michaelis–Menten) was the best attempting to test specific mechanistic hypotheses fitting relationship (see Cardinale et al. 2006a). (for an exception, see e.g. Dimitrakopoulos and For ease of comparison with B, the correlation Schmid 2004). The hypotheses presented in this coefficients obtained using the log-linear fitinC chapter are derived from patterns found in the pre- are used for this chapter. These were very closely vious meta-analyses of B and C. To avoid repetition correlated with the correlation coefficients on the of results reported in Balvanera et al. (2006), we omit Michaelis–Menten scale (r ¼ 0.99, n ¼ 105). The hypotheses relating to the influence of specific significance was not assessed in C because the experimental designs. Instead, we consider several relationships were calculated from means. new hypotheses (see below). We also consider the If the same response variable was measured shape of the relationship between biodiversity and repeatedly in an experiment, it was only entered specific response variables. once in each of the two meta-databases: B focused on Our first hypothesis is that biodiversity effects the first date on which measurements were taken in differ among ecosystem types (Hooper et al. 2005). a study (excluding establishment phases of experi- Differences in biodiversity effects among ecosys- ments) while C selected the last date of published tems could arise, for example, from variation in the measurements. Although about half of the mea- ratios of producer/consumer stocks, or the size, surements contained in C are also in B, the two data generation times, or growth rates of dominant sets were kept separate for our new analyses organisms. For example, Giller et al. (2004) sug- because of the different ways in which biodiversity gested that biodiversity–ecosystem functioning effects were initially extracted or calculated. relationships differ between aquatic and terrestrial We speak of a ‘biodiversity effect’ if a function ecosystems because of more rapid turnover of varies among different levels of biodiversity. Because material and individuals in aquatic systems. How- different levels of biodiversity can be ordered from ever, despite the often expressed concern that 16 BIODIVERSITY, ECOSYSTEM FUNCTIONING, AND HUMAN WELLBEING extrapolation from one ecosystem type to another is ness manipulations than rates (or depletion of unwarranted (Hooper et al. 2005, Balvanera et al. resources). However, as with differences between 2006), we were unable to find specific predictions ecosystem types, it is difficult to predict the direc- about the direction of differences in biodiversity tion of the differences. Using the argument made effects between ecosystem types. above that, for example, community size (as a We distinguish between population-level func- measure of standing stock) may have upper limits tions, recorded for individual target species, such due to the total availability of resources in the as density, cover or biomass; community-level environment, whereas rates of change in commu- functions, recorded for multi-species assemblages, nity size should not be restricted in this way, it such as density, biomass, consumption, diversity; follows that rates should be affected more strongly and ecosystem-level functions, which could not be than stocks. This argument is used by researchers assigned to population- or community-level and who claim that plant species richness may well included abiotic components such as nutrients, increase plant productivity but not carbon storage second hypothesis waterorCO2/O2.Our then is (see e.g. Körner 2004). On the other hand, the the- that species richness enhances community (and ory developed by Michel Loreau (personal com- ecosystem) responses but affects population munication) predicts that stocks should be more responses negatively (Balvanera et al. 2006). responsive than rates. This follows from