Poster Number 558
Wednesday, 8 October 2008
George R. Brown Convention Center, Exhibit Hall E
Soil carbon sequestration is a potential method to alleviate the effects of increasing atmospheric concentrations of CO2, a greenhouse gas. However, obtaining accurate estimates for soil C concentrations and contents often requires extensive sampling and is susceptible to error associated with landscape variability. This study analyzed the relative ability of four different sampling methods—simple random, systematic random, stratified random, and Latin hypercube sampling (LHS)—to accurately predict soil C content and its associated variability. A total of 903 samples were collected along a 10 x 10 m grid in the Piedmont of Georgia to create an exhaustive soil C population. Samples were selected or stratified based on a combination of four landscape variables—curvature, slope, landcover, and soil type—each of which affect soil C sequestration. Using subsamples representing 1%, 4%, 10%, 33%, and 55% of the entire population, the distributions of each of the four sampling methods were compared to the population soil C set. It was clear from the analysis that the stratified and LHS methods provided the best approximations of the mean and variability of the population data. However, further analysis for additional sampling sizes is required to determine the sampling intensity at which the approximation abilities of the two methods diverge. We hypothesize that there may be a threshold sampling intensity below which the LHS consistently provides a better approximation than the stratified method. The goal of this analysis is to provide recommendations for developing the most efficient sampling regime for soil C across variable landscapes.