Sampling challenges posed by continental-scale soil landscape modeling.

Abstract

Limited research exists to guide the development of continental-scale field sampling approaches. We developed a method to compare different sampling schemes designed for continental-scale environmental analysis. Using simulated annealing, we produced hypothetical soil property surfaces for the conterminous United States. The surfaces exhibited varying degrees of polygonality and conditioning by existing soil datasets. Differences in mean between a gridded sample and a polygon-centered sample were evaluated using randomization and other statistical tests. We saw a weak evidence in support of the hypothesis that if a polygonal structure exists in a surface being sampled, then it should be utilized during the actual sampling routine. We saw a somewhat stronger evidence for the hypothesis that differences in local sample density across a continental trend surface can lead to significant sampling bias. Ultimately, the success of different sampling strategies designed for use in continental-scale environmental analysis will largely depend on the ability to anticipate the spatial variability of the variable being measured.