Summary1. Ecological restoration is a global priority that holds great potential for benefiting natural ecosystems, but restoration outcomes are notoriously unpredictable. Resolving this unpredictability represents a major, but critical challenge to the science of restoration ecology. 2. In an effort to move restoration ecology toward a more predictive science, we consider the key issue of variability. Typically, restoration outcomes vary relative to goals (i.e. reference or desired future conditions) and with respect to the outcomes of other restoration efforts. The field of restoration ecology has largely considered only this first type of variation, often focusing on an oversimplified success vs. failure dichotomy. The causes of variation, particularly among restoration efforts, remain poorly understood for most systems. 3. Variation associated with restoration outcomes is a consequence of how, where and when restoration is conducted; variation is also influenced by how the outcome of restoration is measured. We propose that variation should decrease with the number of factors constraining restoration and increase with the specificity of the goal. When factors (e.g. harsh environmental conditions, limited species reintroductions) preclude most species, little variation will exist among restorations, particularly when goals are associated with metrics such as physical structure, where species may be broadly interchangeable. Conversely, when few constraints to species membership exist, substantial variation may result and this will be most pronounced when restoration is assessed by metrics such as taxonomic composition. 4. Synthesis and applications. The variability we observe during restoration results from both restoration context (how, where and when restoration is conducted) and how we evaluate restoration outcomes. To advance the predictive capacity of restoration, we outline a research agenda that considers metrics of restoration outcomes, the drivers of variation among existing restoration efforts, experiments to quantify and understand variation in restoration outcomes, and the development of models to organise, interpret and forecast restoration outcomes.

Summary Ecological restoration is critical for mitigating habitat loss and providing ecosystem services. However, restorations often have lower diversity than remnant, reference sites. Phylogenetic diversity is an important component of biodiversity and ecosystem function that has only recently been used to evaluate restoration outcomes. To move towards prediction in the restoration of biodiversity, it is necessary to understand how phylogenetic diversity of restorations compares with that of reference sites, and where deficits are found, to evaluate factors constraining phylogenetic diversity. We quantified plant taxonomic and phylogenetic diversity in eastern tallgrass prairie, one of the most endangered ecosystems on earth. We measured diversity at large (site) and small (plot) scales in 19 restored prairies and compared patterns with those from 41 remnant prairies. To evaluate how environmental conditions and management actions influence outcomes, we tested the effects of soil properties and seed mix composition on diversity of restorations. Restored prairies were less phylogenetically diverse than remnants at both spatial scales. On the other hand, the total species richness of remnant and restored prairies did not significantly differ, but remnants had higher native richness. Restored communities were taxonomically and phylogenetically distinct from remnants. Soil properties (moisture and pH) influenced phylogenetic diversity and composition. There were positive relationships between the taxonomic and phylogenetic diversity of seed mixes and resulting diversity of planted assemblages (excluding volunteer species). Species in seed mixes were more closely related than expected by chance, and several clades found in remnant prairies were missing from seed mixes. Synthesis and applications. Restored tallgrass prairies had lower phylogenetic diversity than remnant prairies, which may contribute to the widely observed phenomenon of restorations not being functionally equivalent to reference sites. It is encouraging for restoration efforts that seed mix phylogenetic diversity predicted phylogenetic diversity of planted assemblages. This indicates that designing phylogenetically diverse seed mixes for restoration is beneficial. In addition, clades found in reference sites that are missing from restoration seed mixes could be added to new or existing restorations to reduce gaps in phylogenetic diversity. Further work on the effects of management on phylogenetic diversity is needed to advance restoration of biodiversity.

No abstract