The outcomes of ecological restoration are notoriously unpredictable, but we have no general predictive understanding of this contingency. Management decisions can have strong effects on restoration outcomes, but in other cases may be overwhelmed by site characteristics (e.g. soil conditions), landscape context (e.g. abundance of similar habitat) or historical factors (e.g. priority effects). However, we generally cannot predict which of these four classes of drivers will affect restoration outcomes. Disparate aspects of restoration outcomes (e.g. species richness, beta diversity and community composition) and their unique responses further complicate our understanding. Finally, these four classes of drivers might differentially affect subsets of the restored community, where, for example, management might shape the abundance and distribution of species of the target community, while other species are more contingent on site, landscape or historical factors. 2. Here, we used variation partitioning to compare the relative importance of management, site, landscape and historical factors for determining the plant community outcomes of 27 prairie restorations in south-west Michigan. 3. We found that management, especially the composition, diversity and density of seed mixes applied, and history, especially site age, were the most important drivers of prairie restoration species richness, beta diversity and composition. Site and landscape factors were only rarely important for restoration outcomes. 4. Finally, we found that comparing the unique responses of sown and non-sown species typically increased our understanding of the dynamics contributing to community-wide restoration outcomes. 5. Synthesis and applications. This is, to our knowledge, the first quantitative comparison of how four major classes of drivers determine the outcome of restoration. Historical legacies and management decisions, but generally not landscape context or local site conditions, shaped plant communities at restored sites. These findings represent an important step towards developing a more predictive framework for understanding contingency in restoration outcomes.

Summary Recovering biological diversity and ecosystem functioning are primary objectives of ecological restoration, yet these outcomes are often unpredictable. Assessments based on functional traits may help with interpreting variability in both community composition and ecosystem functioning because of their mechanistic and generalizable nature. This promise remains poorly realized, however, because tests linking environmental conditions, functional traits, and ecosystem functioning in restoration are rare. Here, we provide such a test through what is to our knowledge the first empirical application of the ‘response–effect trait framework’ to restoration. This framework provides a trait‐based bridge between community assembly and ecosystem functioning by describing how species respond to environmental conditions based on traits and how the traits of species affect ecosystem functioning. Our study took place across 29 prairies restored from former agricultural fields in southwestern Michigan. We considered how environmental conditions affect ecosystem functioning through and independently of measured functional traits. To do so, we paired field‐collected trait data with data on plant community composition and measures of ecosystem functioning and used structural equation modelling to determine relationships between environmental conditions, community‐weighted means of functional traits and ecosystem functioning. Environmental conditions were predictive of trait composition. Sites restored directly from tillage (as opposed to those allowed to fallow) supported taller species with larger seeds and higher specific leaf area (SLA). Site age and fire frequency were both negatively related to SLA. We also found a positive relationship between soil moisture and SLA. Both trait composition and environmental conditions predicted ecosystem functioning, but these relationships varied among the measured functions. Pollination mode (animal pollination) increased and fire frequency decreased floral resource availability, seed mass had a negative effect on below‐ground biomass production, and vegetative height increased decomposition rate. Soil moisture and fire frequency both increased while site age decreased above‐ground biomass production, and site age and soil moisture both increased decomposition rate. Synthesis and applications. Our results suggest that both trait composition and environmental conditions play a role in shaping ecosystem function during restoration, and the importance of each is dependent on the function of interest. Because of this, environmental heterogeneity will be necessary to promote multiple ecosystem functions across restored landscapes. A trait‐based approach to restoration can aid interpretation of variable outcomes through insights into community assembly and ecosystem functioning.

Community assembly filters, which in theory determine the suite of species that arrive at and establish in a community, have tremendous conceptual relevance to restoration. However, the concept has remained largely theoretical, with a paucity of empirical tests. As such, the applicability of assembly filters theory to ecological restoration remains incompletely known. We tested the relative strengths of dispersal and establishment filters by comparing the plant species composition, measured by species' presence/absence, in 29 restored prairies with the seed mixes used to restore each prairie. We found that both establishment and dispersal filters limited prairie similarity to the seed mix. Sown species responded differentially to filters, with a few species limited only by dispersal (seed density), many others limited only by establishment conditions (i.e. organic matter and sand content of soils, land use history, and fire frequency), and others limited by both dispersal and establishment filters. A few species, typically those sown most often, were not restricted by dispersal or establishment filters, likely because they were sown in high enough densities and all sites had suitable environmental conditions. Finally, one group of species established poorly, but we could not attribute this to either dispersal or establishment filters. This information can help land managers select species likely to establish in restorations when sown at sufficient densities. These results illustrate that dispersal and establishment filters limit the establishment of species in restored communities and these filters are species-dependent. Identifying the most limiting filter(s) for species will inform strategies to increase their establishment success.