Testing the hierarchy of predictability across a gradient of climate severity: a synthesis of seed-based grassland restoration projects in the United States.

Background/Question/Methods

Ecological restoration is key to confronting pressing environmental problems but faces a lack of success and predictability in outcomes. This predictability likely depends on the outcome and the number of factors constraining it. Outcomes achieved by multiple combinations of species abundances such as physical structure and richness are expected to be more predictable than species composition. Additionally, severe climatic conditions constrain species establishment, potentially making restoration less variable and more predictable. Here, we explore drivers of variation in restoration by testing how the predictability varies across outcomes and climate severity. We synthesized data from 12 seed-based grassland restoration projects in the United States, ranging from arid to hyper-humid climates, from the Global Restore Project database. We used 2378 plant monitoring plots from 229 treatments. We fit generalized linear mixed-effects models to predict structure, richness, functional composition, and species composition as a function of restoration project characteristics. Variance explained by models was used as an indicator of predictability. To test how climate impacts predictability, we calculated differences between predicted and observed values. We considered two scales: observed values for plots and average values for treatments. If environmental filters in arid sites increase predictability, we expect smaller absolute differences with increasing aridity.

Results/Conclusions

The characteristics of restoration projects and the identity of sites and treatments explained 87% of total variation in species richness, one of the most predictable outcomes. Restoration projects that used richer seed mixes resulted in richer plant communities. The richness of restored communities declined with time since restoration, an effect only half as strong as the effect of seed mixes. The unexplained variability of richness observed in monitoring plots decreased with increasing aridity, but this effect was only seen in plots smaller than 1m2. Plots of larger area showed the opposite trend. This suggests that the predictability of richness might also depend on scale. In fact, the unexplained variability of treatment-level average richness increased with aridity. Similar patterns were found for structure, functional composition, and species composition, with predictability decreasing across metrics.We encountered differences in predictability across outcomes as expected but the effect of environmental severity did not fully support our hypothesis. Results indicate that predicting the average results of restoration at larger scales is harder under more severe conditions. In these situations where resources are limited, even small changes in resource availability caused by stochastic weather variations could impact species establishment and hence produce less predictable restoration outcomes.