Manipulative field experiments are being conducted around the globe to understand ecosystem impacts of global change factors such as rising atmospheric CO2, warming, drought, and nitrogen deposition. Meta-analysis techniques enable quantitative synthesis of results from such experiments to draw general conclusions about the impacts of global change factors. However, traditional meta-analyses are challenged by several key issues inherent to synthesizing experimental results, including: (1) potential for study- and site-specific factors to influence responses; (2) multiple response variables are often simultaneously reported by individual studies, yet they are often treated independently; (3) responses measured in multi-factor experiments are often binned by single treatment factors and analyzed separately for each factor; and (4) incomplete reporting of responses, covariates, sample sizes, or measures of uncertainty. The goal of this study was to develop a hierarchical Bayesian, multi-variate (HBMV) meta-analysis approach that addresses these issues. We applied a HBMV model to 80 trivariate records of aboveground biomass (AB), belowground biomass (BB), and soil respiration (SR) obtained from elevated CO2 (eCO2) and/or warming experiments to understand how these factors impact ecosystem functioning. We accounted for site history effects (study duration and climate) and potential covariation among study-level AB, BB, and SR responses.
Results/Conclusions
The HBMV meta-analysis is a flexible and powerful synthesis tool, and in this study, it revealed the importance of simultaneously synthesizing multiple response variables (e.g., AB and BB responses to eCO2 were correlated, r = 0.61). The HBMV model also imputed missing records for AB, BB, and SR (~16% of the records reported all three variables), duration (98% reporting), and climate (90% reporting). Site history can govern how eCO2 and warming impact plant and ecosystem functioning; e.g., the effects of eCO2 and warming on BB and AB, respectively, decreased with longer study durations (p = 0.09 and 0.02, respectively). Moreover, the effects of eCO2, warming, or their combination on AB, BB, and SR varied depending on a site’s mean annual precipitation (MAP) and/or mean annual temperature (MAT). Given these climatic effects, model predictions identified biomes that may be particularly sensitive to eCO2 or warming. For example, warming is expected to stimulate SR in subtropical deserts and boreal forests, but reduce SR in savannas and tropical seasonal forests; yet, only ~7% of the records represent these biomes. Additional experiments are needed in these potentially “sensitive” biomes to develop a better understanding of how global change factors impact ecosystem function globally.