Assessing the effect of elevated carbon dioxide on soil carbon: a comparison of four meta-analyses


Bruce A. Hungate, tel. +1 928 523 0925, fax +1 928 523 7500, e-mail:


Soil is the largest reservoir of organic carbon (C) in the terrestrial biosphere and soil C has a relatively long mean residence time. Rising atmospheric carbon dioxide (CO2) concentrations generally increase plant growth and C input to soil, suggesting that soil might help mitigate atmospheric CO2 rise and global warming. But to what extent mitigation will occur is unclear. The large size of the soil C pool not only makes it a potential buffer against rising atmospheric CO2, but also makes it difficult to measure changes amid the existing background. Meta-analysis is one tool that can overcome the limited power of single studies. Four recent meta-analyses addressed this issue but reached somewhat different conclusions about the effect of elevated CO2 on soil C accumulation, especially regarding the role of nitrogen (N) inputs. Here, we assess the extent of differences between these conclusions and propose a new analysis of the data. The four meta-analyses included different studies, derived different effect size estimates from common studies, used different weighting functions and metrics of effect size, and used different approaches to address nonindependence of effect sizes. Although all factors influenced the mean effect size estimates and subsequent inferences, the approach to independence had the largest influence. We recommend that meta-analysts critically assess and report choices about effect size metrics and weighting functions, and criteria for study selection and independence. Such decisions need to be justified carefully because they affect the basis for inference. Our new analysis, with a combined data set, confirms that the effect of elevated CO2 on net soil C accumulation increases with the addition of N fertilizers. Although the effect at low N inputs was not significant, statistical power to detect biogeochemically important effect sizes at low N is limited, even with meta-analysis, suggesting the continued need for long-term experiments.