Guiding soil sampling strategies using classical and spatial statistics: A review

Soil analysis is a key practice to increase the efficiency of nutrient management in agriculture. Since the early 20th century, increasingly sophisticated methods have been developed to describe and manipulate the inherent spatial variability in soil chemical properties within the realms of classical and spatial statistics. In this paper, we reviewed design‐based (classical) and model‐based (geostatistical) sampling to suggest field‐scale sampling strategies consistent with common agronomic management goals in annual crop production systems. To assess the relevance of common sampling methods in relation to practice, current extension recommendations across the United States were compared with results from peer‐reviewed literature. Despite decades of research, specific recommendations for sample sizes, sampling depths, numbers of soil cores, and layouts were highly variable for classical and geostatistical approaches. Mobile nutrients, such as NO3, are frequently lacking in spatial structure and rarely are recommended for site‐specific management. Nonmobile nutrients, such as P, are more spatially dependent and exhibit nested spatial structures that are inconsistent across fields. For these reasons, we recommend design‐based sampling in most situations for simplicity, cost, and objectivity. The common design‐based sampling protocol prescribes collection of individual cores in a zig‐zag pattern that are combined to produce a composite sample. This protocol should be amended because it is not sufficiently randomized and is inadequate for log‐normally distributed variables. To facilitate site‐specific management, we recommend structured approaches for delineating management zones or strata and for researchers to systematically enumerate confounding variables while explicitly defining the scope of inference for future soil sampling studies.