Testing the hierarchy of predictability in grassland restoration across a gradient of environmental severity

Authors: BERTUOL-GARCIA, DIANA - University Of Victoria; BRUDVIG, LARS - Michigan State University; CURRAN, MICHAEL - Abnova Ecological Solutions; Davies, Kirk; LADOUCEUR, EMMA - Martin Luther University; LAUGHLIN, DANIEL - University Of Wyoming; MUNSON, SETH - Us Geological Survey (USGS); Svejcar, Lauren; SHACKELFORD, NANCY - University Of Victoria

Submitted to: Ecological Applications
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 8/18/2023
Publication Date: N/A
Citation: N/A

Interpretive Summary: Ecological restoration is a critical tool for promoting recovery in degraded ecosystems but is challenged by variable success and lack of predictability. We analysed data from 52 grassland sites to test how predictable grassland seed-based restoration can be. We found that the level of predicatability of restoration efforts highly depended on the metric being measured. Similarly, dominant species had higher levels of predictability than rare species.

Technical Abstract: Ecological restoration is a critical tool for promoting recovery in degraded ecosystems but is challenged by variable success and lack of predictability. Understanding which outcomes are more predictable and less variable following restoration could help practitioners set realistic goals and improve restoration effectiveness. A recent hypothesis states the predictability of outcomes would follow an order from most to least predictable based on metrics being nested from coarse to fine community properties (physical structure > taxonomic diversity > functional composition > taxonomic composition), and that predictability would increase with more severe environmental conditions that constrain variation by reducing species establishment. We tested this “hierarchy of predictability” hypothesis by synthesizing outcomes from 12 grassland restoration projects across the United States located along an aridity gradient. We used 1829 vegetation monitoring plots from 227 different restoration treatments, spread across 52 sites. We fit generalized linear mixed-effects models to predict outcomes as a function of restoration characteristics and used the variance explained by models as an indicator of their predictability. To test how aridity impacts predictability, we calculated absolute differences between predicted and observed values, and modelled these differences as a function of aridity. We found mixed support for the hierarchy of predictability hypothesis. Physical structure was among the most predictable outcomes when measured as relative abundance of grasses, but unpredictable when measured as total canopy cover. Similarly, while one dimension of taxonomic composition expressing variation in species identities was unpredictable, another dimension of taxonomic composition related to broad-scale patterns of dominant species was highly predictable. Predictability also did not increase consistently with aridity. The identity of species in restored communities was more predictable with increasing aridity, while functional composition became less predictable and other outcomes showed no significant trend. Overall, restoration outcomes related to dominant species were most predictable, whereas indicators involving rare species were harder to predict, possibly due to influences of stochastic processes on rare species and the large spatial extent of our study. By relating variation to specific metrics and site conditions, our findings can assist goal setting, guide assessment, and further the development of predictive capacities in restoration ecology.