Human activities alter communities around the world. Experiments to assess these impacts are more powerful when replicated at multiple sites, but sensitivity to treatment effects should differ among sites depending on how much variation occurs within sites and how that variation is partitioned. For example, it would be easier to detect treatment effects at a site with low spatial variation and high temporal variation. Furthermore, conclusions may depend on how variation is expressed. We used a large grassland plant community dataset to i) quantify the magnitude of compositional variation within sites, using multiple metrics, ii) relate magnitude of compositional variation to site-level context, and iii) partition compositional variation among sources. Data were collected at 50 grassland sites on 6 continents. Each site had an identical experimental and sampling design: 8 nutrient addition treatments x 3 blocks x 4 years (pre-treatment plus 3 years post-treatment), for a total of 96 plot-years. Site-level context was expressed by biotic and climate-related predictors. For each site, compositional variation was expressed as abundance-based (Bray-Curtis; B-C) and incidence-based (Sorensen) dissimilarities and by the percentage of dissimilarity due to turnover. Compositional variation was partitioned to quantify the relative importance of spatial, temporal, and treatment-related sources.
Results/Conclusions
Compositional variation was large: across sites, the average B-C dissimilarity between two plot-years at a site was 0.55, Sorensen dissimilarity was 0.38, and species turnover accounted for 69% of compositional variation. However, there was wide variation around these averages: plot-years were similar in composition at some sites (B-C = 0.23) but very different (B-C = 0.78) at other sites. Site-level patterns in magnitude of compositional variation were related to the size of the local species pool, climate-related variables, productivity, and local variation in productivity. Treatment accounted for much less of the compositional variation than was accounted for by spatial and temporal sources. We highlight four implications from this work. First, substantial compositional variation is the norm in grassland communities around the globe. Second, there is a degree of predictability to the magnitude of compositional variation at a site, relating to the size of the local species pool and to environmental context. Third, using multiple compositional metrics can provide unique ecological insights. For example, some sites exhibit high variation in abundance-based dissimilarity while others exhibit high variation in incidence-based dissimilarity. Finally, monitoring and analysis programs need to account for both spatial and temporal variation in composition.