The current rapid decline in biological diversity has spurred a large body of research investigating linkages between species richness and ecosystem function. With few exceptions, these experiments randomly manipulate species richness, which may limit their applicability because not all species within an ecosystem are equally susceptible to decline or extinction. Rapid biodiversity losses at both local and global scales involve species with particular evolutionary histories and ecological traits. Hence, our understanding of the functional consequences of non-random species loss is limited. We compared the effects of random vs. ordered species losses on aboveground net primary productivity (ANPP) using a pool of sixteen native serpentine grassland plant species planted in serpentine field plots at Kirby Canyon Reserve near San Jose, California. Our experiment consists of 90 plots (29 cm radius) in nine blocks varying in soil depth. Each plot contains a mixture of 2, 5, 8, 12, or 16 species either drawn at random or adhering to a realistic loss order based on nested subset analysis of 19 years of plant community composition data from our experiment site. We estimated ANPP from dry aboveground biomass.
Species richness did not affect ANPP in the randomized species loss treatments (F = 0.75, P = 0.56). In contrast, ANPP declined strongly with species richness in the ordered loss treatment plots (F = 18.98, P < 0.001), by an average of ~ 45% from the 16-species treatment to the 2-species treatment. Although deeper soils in this resource-limited ecosystem are typically associated with higher nutrient and moisture availability, soil depth did not affect ANPP in our plots (F = 0.999, P = 0.37). Our findings illustrate that the order of species losses from a community can strongly influence the effects of declining species richness on ecosystem processes. In this ecosystem, a biodiversity-productivity study based solely on randomized assemblages would have predicted a very different response of ANPP than the one that occurred under more likely, realistic biodiversity losses. We argue that the contrast between the effects of randomized and realistic species losses results because the latter involves concentrated loss of particular traits rather than a random loss of traits from across a spectrum.