Comparison of Response Surface and Unsupervised Classification Sampling Approaches to Characterize Soil Spatial Variability with ECa-Directed Soil Sampling.

Characterizing spatial variability is an important consideration of any landscape-scale soil-related problem. Soil spatial variability influences agricultural management, solute transport processes in the vadose zone, and impacts of contaminants on surface and ground water resources, just to mention a few. Recently, scientists at the USDA-ARS U.S. Salinity Laboratory have developed protocols for characterizing the spatial variability of specific soil properties (e.g., salinity, water content, texture, bulk density, organic matter) using geospatial measurements of apparent soil electrical conductivity (ECa) to direct soil sampling. Statistical sampling designs used in the protocols can be either model based (e.g., response surface sample design) or design based (e.g., unsupervised classification). A comparison of response surface and unsupervised classification sampling designs is presented to better understand their strengths and weaknesses. Five fields near Sterling, CO, were used to make the comparison. Geospatial ECa measurements were taken using mobile electromagnetic induction equipment. Based on the ECa measurements, 11 sites were selected for each field using a response surface and 9 sites were selected using unsupervised classification. Six randomly sites in each field were selected and used to evaluate which sampling approach could better predict those soil properties at the random sites that correlate with ECa. Results show the response surface sample design generally predicts the random locations better than unsupervised classification, suggesting that a response surface sampling design will better characterize spatial variability. This information substantiates the use of a response surface sample design over design-based sampling schemes in characterizing spatial variability with ECa-directed soil sampling.