The outcomes of ecological restoration can vary dramatically across different sites and different years, even when management practices are similar, leading to the widespread belief that the outcomes of ecological restoration are essentially unpredictable. However, this assertion is rarely tested. Testing and developing predictive ability would benefit restoration practice by allowing practitioners to more reliably meet restoration goals, and measuring prediction accuracy would serve as a test of ecological theory by quantifying our understanding of community assembly. A first step toward meeting this goal is to explain variation in restoration outcomes. Then, we can use these explanatory models to forecast future site conditions and extrapolate our results to new sites. Here we focus on predictions of species richness, and we test our ability to explain, forecast, and extrapolate predictions of species richness in restored tallgrass prairies, including 21 sites that have been resampled three times over six years and 99 sites that were sampled during a single year. At each site we collected data on plant species richness and predictor variables, including site age, soil texture and nutrients, fire history, and seed mixes used to initiate the restoration.
Results/Conclusions
Using linear mixed models, we can explain a large proportion of the variation in species richness at sites surveyed in 2011 and 2013 (r2 = 0.90). Using the parameters from our explanatory models, we were able to forecast species richness observed during re-surveys in 2016. These predictions accounted for approximately half of the variation in species richness (r2 = 0.52). For both the explanatory models and forecasts, a similar proportion of variation was explained by random effects, including site identity, and fixed effects, including soil variables, fire history, and site age. Together, these results demonstrate that ecological restoration may be more predictable than generally thought, since we can generate good explanations for variation in species richness and use this understanding to forecast future conditions. Despite successes at forecasting species richness, our models were unable to extrapolate our predictions to new sites (r2 = 0.03). Overall, despite ecological restoration’s reputation for unpredictability, we were able to explain and forecast species richness for 21 sites that we have repeatedly sampled. However, our inability to extrapolate results to new sites highlights the need to train models under the conditions for which the models will be used to make predictions.