A soil sampling design for arable land quality observation by using <scp>SPCOSA–CLHS</scp> hybrid approach

Arable land quality has been evaluated through weighted average of indicators related to soil properties and tillage technics to present the most basic function of cropland: food production potential. A hybrid sampling method spatial coverage sampling and random sampling–conditioned Latin hypercube sampling (SPCOSA–CLHS) was designed for arable land quality observation network in this paper. The SPCOSA–CLHS integrates the uniform spatial partitioning results generated by the spatial coverage sampling and random sampling (SPCOSA) into the conditioned Latin hypercube sampling (CLHS) method along with other arable land quality indicators such as field slope, soil bulk density, organic matter content, thickness of plough layer and irrigation. Then, SPCOSA, CLHS, SPCOSA–CLHS, CLHS with x and y coordinates as covariates (XY‐CLHS), random sampling method (RSM) were compared using the example of Heilongjiang Province. The sample population covers 12,147,008 grid cells and 17 arable land quality indicators. Five parameters: information entropy, Kullback–Leibler divergence, similarity distance, expression ability to local spatial heterogeneity of arable land quality and distribution homogeneity of sampling results were used to compare the applicability of these methods for overall arable land quality. The SPCOSA–CLHS can better trade off sampling result's representative ability to the population and spatial heterogeneity of arable land quality, as well as its samples have advantages of spatially uniform distribution. When the sample size is between 5000 and 20,000, all methods show good applicability. When the sample size is below 5000, however, the differences among these methods become significant. SPCOSA and random sampling method offset most dramatically. Based on this detailed comparison of the five sampling strategy approaches, we strongly recommend to use SPCOSA–CLHS to design arable land quality observation.