Statistical Challenges in Soil Carbon and GHG Emission Analyses.

Measurements of soil carbon (C) and greenhouse gas (GHG) emissions from soil have gained a large amount of attention in an effort to reduce the impact of agriculture on concentrations of atmospheric GHG and to potentially increase its role in mitigation changing climate effects. The number of field experiments aiming at monitoring changes in soil C and GHG emissions in response to different land use and management practices has grown exponentially in the last decade. However, it has been only partially recognized that the nature of these data is such that requires advanced statistical techniques in order to fully and properly explore the influence of the studied treatments; moreover that for soil C and GHG data some experimental design approaches can enhance while others can weaken the study’s ability to detect treatment induced differences.  The objective of this presentation is to address several of the experimental design and statistical analysis issues in soil C and GHG emission data analyses. Specifically, I will discuss 1) pros and cons of linear vs. non-linear model fitting in GHG emission rate determination as well as benefits of the random coefficient regression approach; 2) importance of accounting for spatial and temporal correlation in soil C and GHG data via repeated measures approach; 3) heterogeneous variances in treatments and time and why it is crucial to account for those; 4) sources of variability in soil CO₂ and N₂O emission data and their implications for determining the numbers of experimental plots and measurement chambers (sub-samples); 5) particular importance of proper blocking in experiments measuring soil C and CO₂ emissions. Actual soil C and CO₂ and N₂O emission data collected from several experimental sites of Michigan State University will be used as illustrations.