Incorporating Spatial Dependence into Estimates of Soil Carbon Contents under Different Land Covers

Soil C sequestration is a potential method to reduce increasing atmospheric concentrations of CO2, a greenhouse gas. Soil C represents a significant portion of total C in a landscape, although spatial variation can complicate predictions of soil C contents and its change, particularly for C sequestration projects. Quantifying spatial variability for different landscapes may allow adjustment of sampling intensities to improve estimation efficiency, thus facilitating change detection. This study used semivariogram analysis to quantify spatial autocorrelation of soil C concentrations and contents across distances of 10 to 100 m in two 1-ha pine, hardwood, and pasture sites in the Georgia Piedmont USA. We hypothesized that land cover could help predict spatial autocorrelation and that spatial dependence would exist at greater distances in pasture than pine or hardwood. Average (±1 SD) major range was smaller for pasture total soil C (65 ± 48 m) than pine (77 ± 31 m) or hardwood (81 ± 25 m), which did not support our hypothesis. Two forested plots demonstrated major ranges of 98.8 m, indicating that spatial dependence probably occurred at a scale greater than our plots. Spatial structure was weak to moderate with nugget/sill ratios > 0.30 in all plots. Average C contents in 0 to 7.5 cm based on 64 point samples per plot were 29.9 ± 0.0, 19.8 ± 1.2, and 19.4 ± 4.0 Mg ha–1 for hardwood, pine, and pasture plots (n = 2), respectively. Kriged predictions estimated average C contents of 29.6 ± 0.4, 20.2 ± 0.9, and 19.5 ± 4.1 Mg ha–1, respectively. Incorporating spatial dependence can limit sampling redundancy and improve precision, but only where spatial structure is sufficiently strong.