Toward prediction in the restoration of biodiversity

By assisting the recovery of damaged or destroyed ecosystems, ecological restoration has been hailed for its promise to combat major causes of biodiversity decline (e.g. habitat loss, invasive species; Vitousek et al. 1997; Foley et al. 2005; Newbold et al. 2015) and is widely viewed as a critical component of successful conservation and land management strategies (e.g. Dobson, Bradshaw & Baker 1997; Hobbs & Harris 2001; Suding 2011; Menz, Dixon & Hobbs 2013; Perring et al. 2015). The relevance of restoration has only increased in recent years, as a tool to combat climate change (e.g. Harris et al. 2006), promote ecosystem services (Rey Benayas et al. 2009; Montoya, Rogers & Memmott 2012) and facilitate the emerging “restoration economy”—economic benefits of conducting restoration activities, which are now valued at ~$25 billion annually in the United States (BenDor et al. 2015). Ongoing or planned restoration projects range from local scale efforts, to high profile commitments to restore hundreds of millions of hectares of forest, such as the Bonn Challenge (Suding et al. 2015; Holl 2017). Successful restoration, in all of these regards, hinges on the capability to reliably achieve stated goals (Hobbs 2007). In practice, however, restoration outcomes are notoriously unpredictable and range from success to failure (Suding 2011). Seemingly similar restoration practices can result in substantial variation in outcomes, such as community diversity or composition (e.g. Fagan et al. 2008; Matthews et al. 2009; Grman, Bassett & Brudvig 2013), and in many cases, we lack understanding of the processes that lead to this variation (Brudvig 2011; Suding 2011). To reach its ecological and socioeconomic potential, restoration ecology must mature from its site-specific and ad hoc roots (Hobbs & Norton 1996) into a truly predictive science. Predictive restoration science would entail the capacity to robustly explain the variation in outcomes among existing restoration projects and, moreover, the ability to forecast outcomes prior to project initiation, based on clearly articulated goals, planned restoration actions and knowledge of conditions at a particular location (Fig. 1). This Special Feature confronts the issue of prediction in restoration by considering variation in restoration outcomes and how ecological theory can assist with interpreting this variation. The papers focus on restoration of biodiversity, which is a core consideration in restoration research and practice (Society for Ecological Restoration International Science & Policy Working Group 2004; Ruiz-Jaen & Aide 2005; Rey Benayas et al. 2009; Brudvig 2011) and an aspect of ecology to which substantial conceptual and theoretical attention has been paid, including in the field of restoration ecology (e.g. Hobbs & Norton 1996; Palmer, Ambrose & Poff 1997; Suding, Gross & Houseman 2004; Temperton et al. 2004; Funk et al. 2008; Brudvig 2011). The application of theory holds an important—some might argue central—role in environmental problem solving, by contributing to effective solutions that are transferable among contexts and locations (Cadotte et al. 2017). The value of theory for guiding restoration practice has long been recognized (e.g. Hobbs & Norton 1996; Palmer, Ambrose & Poff 1997; Young, Chase & Huddleston 2001; Suding, Gross & Houseman 2004; Török & Helm 2017), although the application of many relevant bodies of theory has lagged behind their conceptual consideration. Through strong empirical tests, papers in this Special Feature evaluate how a variety of ecological theories apply to restoration, focusing on theory in the sense of framing expectations or hypotheses around a priori concepts. Among these are theories about priority effects (Stuble, Fick & Young 2017), plant–soil feedbacks (Bauer et al. 2017), ecological resilience (Larios, Hallett & Suding 2017), community phylogenetics (Barak et al. 2017), and the roles of seed dispersal and plant functional traits for community assembly (Holl et al. 2017; Laughlin et al. 2017; Turley et al. 2017; Zirbel et al. 2017). In each paper, theories drive the development of research studies and the results of these studies inform restoration practice and, at the same time, feed back as tests of theory (Fig. 1). A common message linking the papers in this Special Feature is that variation in restoration outcomes is a result of how, where, and when restoration is practised (Brudvig et al. 2017). In doing so, these papers illustrate mechanisms underpinning variation among restoration efforts and demonstrate that, while restoration outcomes may be idiosyncratic, it is possible to make sense of this idiosyncrasy. Uncovering the mechanisms underlying variation in outcomes represents an important step toward prediction in restoration and points the field in the direction of predictive forecasting (Fig. 1). Papers in the Special Feature explore how the details of restoration practice influence variation in restoration outcomes. For example, differences in tree planting regimes influence subsequent tree recruitment patterns in tropical forests (Holl et al. 2017). The diversity of seeds in seed mixes influences the diversity of plant species in resulting restored prairie grassland plant communities (Barak et al. 2017). Tree thinning in longleaf pine savannas affects rates of recruitment of species sown by seed (Turley et al. 2017). And, variation in prescribed fire management influences the assembly of ponderosa pine forest understorey and prairie grassland plant communities (Laughlin et al. 2017; Zirbel et al. 2017). Thus, decisions that land managers make have important consequences for the outcomes of restoration. Yet, as papers in this Special Feature illustrate, variation in restoration outcomes also results from environmental factors that typically lie outside of managers’ control. These include the ways that localities differ in their underlying edaphic conditions (Barak et al. 2017; Zirbel et al. 2017), levels of eutrophication (Larios, Hallett & Suding 2017), site elevation (Holl et al. 2017), landscape context (Zirbel et al. 2017) and differences related to variation in land-use history (Bauer et al. 2017; Turley et al. 2017; Zirbel et al. 2017). Finally, Stuble, Fick and Young (2017) demonstrate that when and, in particular, in which year restoration takes place can impact its outcome. In their study, inter-annual variation in planting year conditions, likely related to precipitation and temperature, caused otherwise identically conducted California grassland restorations to assemble along divergent trajectories. Developing a predictive science of biodiversity restoration requires grappling with the rich concept of biodiversity. Biodiversity encompasses the variety of life from genes to landscapes; however, as a field, restoration ecology has largely focused narrowly on species as a unit of biodiversity (Brudvig 2011; Hipp et al. 2015), often summarized as species richness (Ruiz-Jaen & Aide 2005). Species will (and should) remain a staple currency of biodiversity restoration; yet, articles in this Special Feature illustrate the promise and importance of considering biodiversity beyond the taxonomic level. First, restoration may impact various metrics of biodiversity in different ways. Barak et al. (2017) illustrate this point by comparing plant biodiversity in remnant tallgrass prairies to plant biodiversity in tallgrass prairies that were restored by seed sowing onto former agricultural lands in Illinois. Restored prairies supported similar (taxonomic) species richness, but lower phylogenetic diversity and altered community composition, relative to remnants. A likely reason for this is that the seed mixes used during restoration were phylogenetically clustered, comprised of many closely related species, and lacked several clades characteristic of remnants. As a result, the authors call for the use of more phylogenetically diverse seed mixes in restoration. Second, consideration of biodiversity through the lens of functional traits can afford mechanistic insights into how communities assemble and ecosystems function (Lavorel & Garnier 2002). Zirbel et al. (2017) illustrate how this framework applies to restoration by showing how restored tallgrass prairie plant communities in Michigan have assembled along environmental gradients (e.g. soil moisture, fire frequency) based on plant functional traits (e.g. specific leaf area). They then relate aspects of ecosystem functioning, such as decomposition rates and the production of insect-supporting flowers, to the plant community trait composition, as well as environmental factors that were related (e.g. fire frequency) and unrelated (e.g. edaphic conditions) to restoration practice. Third, how we assess biodiversity has bearing on how predictable we deem restoration to be. Brudvig et al. (2017) argue that this is due to the nested nature of biodiversity responses. Whereas some aspects of biodiversity can result from many different combinations of functionally redundant species (e.g. vegetation structure or species richness), rendering them relatively more predictable, others result from fewer or only one particular species assemblage (e.g. taxonomic community composition), leaving them more prone to rare or chance events during restoration and, thus, less predictable. As a result, they propose a hierarchy of predictability: ecosystem structure (most predictable), taxonomic diversity, functional or phylogenetic diversity and taxonomic composition (least predictable). Laughlin et al. (2017) then confront this proposed hierarchy of predictability using herbaceous understorey data from a long-running ponderosa pine forest restoration experiment in Arizona, where plots were subjected to combinations of tree thinning and prescribed fire restoration treatments for 23 years. Their results are broadly congruent with Brudvig et al.'s (2017) proposed hierarchy. Laughlin et al.'s (2017) statistical models best predicted levels of herbaceous biomass (vegetation structure) and species richness, followed by functional trait-based measures of diversity and community composition (the most variable response). These findings are important for framing expectations during restoration and, like Barak et al. (2017), illustrate how responses to restoration depend on how we define biodiversity, even within a single trophic level. The papers in this Special Feature illustrate how progress toward prediction in restoration will require a combination of observational, experimental and modelling approaches. These approaches are complimentary and together afford insights into broad scale patterns, the mechanisms underlying these patterns and the capacity to forecast outcomes (Fig. 1). For example, Barak et al. (2017) and Zirbel et al. (2017) use observational approaches to compare among tallgrass prairie restorations, leading to insights into why restoration outcomes differ from conditions in reference (remnant) sites and why outcomes differ among restorations respectively. A variety of experimental approaches are used to understand mechanisms underpinning restoration outcomes. These include plant–soil feedback experiments (Bauer et al. 2017), seed sowing and planting experiments (Holl et al. 2017; Stuble, Fick & Young 2017; Turley et al. 2017), and large-scale manipulations of fire and tree canopy structure (Laughlin et al. 2017; Turley et al. 2017). Holl et al. (2017) combine the strengths of observational and experimental approaches by replicating a tree planting experiment across landscapes that vary in surrounding forest cover (and, thus, natural propagule sources), showing that it is the planting regime and not surrounding landscape context that affects natural tree recruitment over the course of a decade in their plots. In turn, the papers in this Special Feature employ a variety of models to evaluate and interpret restoration outcomes. Brudvig et al. (2017) develop conceptual models to guide understanding of variation among restoration efforts, considering aspects of restoration approach, environmental conditions and the metric of biodiversity under consideration (e.g. species richness vs. composition). Papers employ an assortment of statistical modelling approaches. These include Zirbel et al.'s (2017) use of structural equation modelling to link environmental conditions, plant functional traits and aspects of ecosystem functioning, within a trait effect–response framework. Laughlin et al. (2017) use mixed effects models to consider the roles of experimental manipulations (fire, tree thinning) and unmanipulated site-to-site variation, in an effort to explain variation in herbaceous understorey responses to restoration treatments. Larios, Hallett and Suding (2017) employ a demographic modelling approach to explore the issue of resilience during restoration and forecast predictions of population dynamics under varying scenarios of grassland composition and nitrogen deposition. In summary, the papers in this Special Feature wield approaches that, together, will make progress toward prediction in restoration ecology (Fig. 1). Theory-driven experiments and observational studies are developed in response to stated restoration goals (e.g. to promote the recolonization of post-agricultural forests by native understorey herbs; Turley et al. 2017). The outcomes of these studies are evaluated using explanatory models (e.g. Laughlin et al. 2017; Zirbel et al. 2017), or used to parameterize forecasting models that make predictions about future system states (Larios, Hallett & Suding 2017). The outcomes of each study represent tests of underlying theories and inform the practice of restoration. Prediction remains a frontier in ecology (e.g. Doak et al. 2008; Evans, Norris & Benton 2012; Mouquet et al. 2015; Petchey et al. 2015; Houlahan et al. 2017) and restoration ecology is no exception. The papers in this Special Feature help to point the way forward, through strong connections to theory, tests of mechanisms underlying restoration outcomes, and explanation of the variation that results among restoration efforts. In this way, the papers in this Special Feature largely focus on “explanatory predictions” (as opposed to “anticipatory predictions”—see below; Mouquet et al. 2015), through the use of statistical models that explain variation in biodiversity based on measured or manipulated variables. For example, Laughlin et al. (2017) consider how models of restoration treatment factors and environmental variation predict herbaceous understorey responses in ponderosa pine forests. Such approaches to explanatory prediction are important for informing restoration practice (Fig. 1). Yet, in spite of these advances, further progress will be necessary to cultivate predictive forecasting capabilities in ecological restoration. In particular, there are few if any studies that have developed predictive models of restored ecosystems (e.g. determinants of species richness at restored sites) and then tested these predictions against independent datasets—a key to validating anticipatory predictions (Mouquet et al. 2015). Larios, Hallett and Suding (2017) make important headway on this front, by developing predictive demographic models that forecast population responses. An exciting avenue in restoration ecology research will be to confront models parameterized by data from existing restorations or experiments with data from independent restorations, or to predict and then evaluate the outcomes of restorations a priori (Fig. 1). There are clear pragmatic reasons for restoration ecology to develop into a predictive science and, at the same time, this development would position restoration to make important advances to predictive ecology broadly. Restored ecosystems have long been recognized for their capacity to serve as testing grounds of ecological theory (e.g. Bradshaw 1987; “acid test” metaphor). As restoration ecology has broadened from a strict focus on recreating past ecosystem states (Choi 2004, 2007), a new acid test emerges to make accurate explanatory and anticipatory (forecasted) predictions of restoration outcomes and for these outcomes to lie within the range of desired conditions (Hiers et al. 2016). Once we can reliably hit restoration targets based on a priori expectations, this will provide convincing evidence that our theoretical constructs are valid and our restoration practices are effective. Toward prediction in restoration! I thank Marc Cadotte and Alice Plane for their assistance throughout the development of this Special Feature and all of the authors for their excellent contributions. Many of the ideas in this Special Feature initiated at an Organized Oral Session entitled “Toward prediction in the restoration of biodiversity,” at the 2015 Ecological Society of American Annual Meeting. Jonathan Bauer, Tyler Bassett, Alex Peake and Chad Zirbel provided helpful feedback on this editorial. L.A.B. was supported by NSF DEB 1552197. Data have not been archived because this article does not contain data.