Biodiversity loss has accelerated in the Anthropocene – an epoch when human actions become the major driver of global environmental changes. A general consensus has emerged that biodiversity loss is altering fundamental processes that underlie goods and services ecosystems provide to humanity. Yet such consensus has been largely built from experiments performed at small spatial-scales for short time-frames. It remains unclear whether and how effects can scale up to real-world landscapes relevant for management and conservation policy. In this study, we conducted a data synthesis of > 600 published experiments and asked: (1) How do effects of biodiversity on ecosystem services vary across spatial and temporal scales? (2) To what extent do results of small-scale experimental studies over- or under-estimate biodiversity effects in real-world landscapes? Our initial analysis quantified effects of terrestrial plant species richness on biomass production in two forms: net diversity effects as log ratio (LRnet) and power function parameter (b) from curving fitting. Spatial scale was standardized as experimental unit size divided by mean body size, and temporal scales as experiment duration divided by mean generation time of organisms. General linear-mixed effects model was performed to test how biodiversity effects vary across spatial and temporal scales.
Spatial and temporal scales interacted to alter magnitude and functional form of plant diversity effects on biomass. Specifically, as spatial scale increased, effects of temporal scale on LRnet increased from near zero to positive in a nonlinear and concave-down manner. On the other hand, when temporal scales increased, effects of spatial scales on LRnet increased from negative to positive, and reached to a saturation point at the maximum of temporal scale in the dataset. Similar spatial- and temporal-scale interactions were found for analyses that considered the functional form of plant diversity effects, quantified as scaling parameter, b, of a power function. Within the range of dataset, LRnet and b increased by a factor of 1.68 and by 0.27, respectively for each 10-fold increases in the number of generations, and increased by a factor of 1.10 and by 0.04, respectively for each 10-fold increases in spatial scales. Based on our statistical models, we further predicted effects of plant species richness on biomass at scales of real-world landscapes. Our research indicated that adopting results from small experiments might underestimate actual biodiversity effects at human-dominated landscapes, and highlighted importance of conserving large ecosystems for long time for effectively sustaining biodiversity and ecosystem services.