A Comparison of Three Field Sampling Methods to Estimate Soil Carbon Content
Abstract:Accurate estimates of soil carbon (C) contents over large spatial scales require extensive sampling and are susceptible to error associated with landscape variability. Various methods to minimize error have been posed, including conditioned Latin hypercube sampling (cLHS). The potential advantage of cLHS is that it uses existing ancillary landscape data in geographic information systems to select stratified random samples. Although the theoretical basis for cLHS has been demonstrated, few empirical evaluations have been performed. This study compared simple random, stratified random, and cLHS predictions of soil C content and their associated variability. A population of 903 gridded samples was collected and then subsampled based on stratification of curvature, slope, land cover, and soil type using 1, 5, 12, and 35% data set proportions. Random stratified and cLHS methods best approximated mean and variance of the population at sample sizes of 5 and 12%. Large advantages of cLHS relative to stratified random sampling were not apparent at this site, although cLHS consistently sampled the tails of the soil C content population distribution. Mapped soil C contents from the three sampling methods using regression kriging resulted in root mean square error (±SE) values of 11.1 ± 0.5, 13.4 ± 0.3, and 10.1 ± 0.2 Mg ha−1, respectively, for random, stratified, and cLHS sampling. Stratification sampling methods such as cLHS and stratified random sampling use existing ancillary information to provide an effective means for stratifying sampling locations within existing feature space and can offer improved spatial predictions of landscape-associated soil C contents.
Document Type: Research Article
Publication date: October 1, 2012
More about this publication?
- Membership Information
- ingentaconnect is not responsible for the content or availability of external websites