Traits-based global change predictions
Of the 822 studies we reviewed, a small portion (23%) focused on applying traits in the context of global change, and even fewer (3%) applied traits to generate predictions about patterns of diversity, abundance, or distribution beyond the data used for the initial analysis. Global change drivers, in order of decreasing frequency within the studies we evaluated, included habitat degradation (8%), climate change (6%), biological invasion (5%), multiple/non-specified drivers (3%), and exploitation (1%). Predictive studies emerged primarily within the last 10 years (82%); half of all predictive studies were published in the last five years (Figure 2F). More than 30% of predictive studies focused on applying plant morphological traits to predict the outcomes of abiotic environmental filtering in terrestrial ecosystems. Crucially, studies that generated trait-based predictions of global change (the main focus of this review) represent fewer than 3% (22) of all studies. Of these, more than half (12) focus on ecological prediction in a climate change context, three on biological invasions, and a single study each on the consequences of habitat degradation and exploitation. Five studies used traits to predict the outcomes of multiple global change drivers (two marine [Jacob et al. 2011, Eklof et al. 2015]; three terrestrial [Cardillo et al. 2004; Dury et al. 2018; Knott et al. 2019]).
The narrowest subset of journal articles (and subsequently, traits) corresponded to studies that were both predictive and investigated global change impacts to ecosystems (nstudies = 22). Accordingly, the assemblage of traits used by papers that assessed drivers of global change and performed predictive ecological modelling were highly nested within the broader suite of traits used across descriptive and non-global change studies, and thus differed statistically with traits used more broadly (Figures 6E/F, Figures S1E/F, Figure S2E/F). Habitat associations and life history were the most important suites of traits in studies assessing climate change impacts, as well as undertaking predictive analyses, while morphological traits were most important to studies that investigated habitat degradation and biological invasions (Figure S3&S4, Tables S4-8). While size is also the most common single trait type in predictive studies, in general physiological traits related to resource acquisition and requirements, such as thermal tolerance, and life history traits are more often applied within predictive studies compared with descriptive trait-based work (Figures S5&S6, Tables S8-11).
More closely examining predictive global change studies identified in this review highlights a range of methodological approaches that require data inputs at varied spatio-temporal scales, and therefore resulting in predictions at a range of resolutions (Figure 7). Each method has strengths and weaknesses; trait-based experiments offer opportunities to generate and test fine-scale predictions about response to global change drivers, yet insights gained through experimentation are most relevant under the set of conditions under which the study takes place (e.g. Eckloff et al. 2015; eelgrass communities under climate change and grazer loss; Figure 7). The results of experiments and environmental correlations can be synthesized via meta-analyses to generalize effect sizes for trait types that recur across taxa and ecosystem type (e.g. Cattano et al. 2018’s synthesis of acidification effects on marine fishes; Figure 7). Process models offer an opportunity to examine the effect of more complex interactions on ecological phenomena under global change; however, insights gained from this approach may not be at a resolution needed for conservation and management decision-making (e.g. Jacob et al. 2011’s trait-based polar sea food web model; Figure 7). Traits-based distribution models—the most common approach to global change predictions identified by our review—have generally been applied to forecasting range and abundances under future abiotic (climate) conditions (e.g. Fordham et al. 2012’s and Whitney et al. 2017’s distribution models of trees and freshwater fish, respectively; Figure 7). Spatial projections generated from distribution models can be intuitively applied to place-based biodiversity conservation and natural resource management, but generally omit biotic interactions and feedbacks that further refine species’ ranges and abundances across the landscape (Figure 1). Spatially explicit process models offer a means to generate range and abundance projections that account for multiple environmental filtering processes simultaneously (e.g. Edmunds et al. 2014’s spatial process model of coral survival and growth under competition and climate warming scenarios; Figure 7). However, this approach has not yet been widely applied to global change prediction, perhaps due to the complexity and scale of the required trait data inputs.