The ability to accurately forecast ecosystem dynamics is a gold standard of ecological understanding and would facilitate the capacity to restore damaged ecosystems. Yet, the accuracy of forecasts may vary depending on the focal ecological property (species, traits, functions) and/or environmental conditions. Due to the idiosyncrasies of species turnover during community assembly, biodiversity measures that consider traits or species identities may be less predictable than measures that ignore identities (e.g., species diversity) or consider ecosystem properties (e.g., biomass). Site conditions that limit species membership, such as unproductive soils or competitive community members, may constrain assembly outcomes and lead to more accurate forecasts. We evaluated forecasting potential for multiple plant biodiversity measures (species diversity, trait diversity, biomass) in 20 tallgrass prairies undergoing restoration. We developed forecasting models through surveys of each site in 2011 and 2013, along with information about soils, competitive species abundance, and management factors; we then evaluated forecast accuracy with a third survey in 2016. We asked: What aspects of biodiversity are most predictable during restoration? And, what ecological factors influence the accuracy of these forecasts?
Results/Conclusions
Forecast accuracy was influenced by both assessment metric and site properties. As we hypothesized, forecasts of Simpson's diversity, a metric which ignores species and trait identity, were the most accurate, predicting 40% of species diversity in the third survey period. Simpson's diversity forecasts were least accurate at sites supporting intermediate abundance of a dominant competitor (big bluestem), particularly when soil resources were limited. As we expected, forecasts of functional trait diversity were less accurate than the Simpson's diversity model, predicting 33% of variation at the third survey period. Neither big bluestem abundance nor soil resources influenced forecast accuracy for the trait model. Surprisingly, forecasts of biomass were the least accurate, predicting only 8% of biomass in the third survey period. The biomass model was most accurate at sites supporting high big bluestem abundance. In sum, we find mixed support for our hypotheses. Metrics ignoring species and trait identity were both the most (Simpson's diversity) and least (biomass) accurately forecasted. Dominant species abundance and harsh soil conditions influenced forecast accuracy, but only in some models and not always in the direction we hypothesized. Our findings demonstrate the potential for forecasting restoration outcomes and the importance of considering site conditions and choice of evaluation metric when doing so.