Statistical Evaluation of an Array of Soil Moisture Sensors.

The study of large experimental units may require decisions regarding the number and location of sensors for data collection in order to characterize the experimental unit. Sensors are often arrayed spatially based on the assumption that the variation is closely associated with landscape position. We studied the variation of soil moisture in space and time within a small (7.8 ha) catchment within the Georgia Piedmont of the Southeastern USA over 3 years. Statistical “distance” between all possible pairs of 12 sites was estimated using the Mahalanobis distance. A diagonal matrix of all possible paired differences was then tested with multidimensional scaling (MDS) to estimate the number of dimensions separating the 12 sample sites. The MDS analysis showed that there was only a single dimension that separated the 12 sites even though they were arrayed throughout the catchment. The single dimension was then shown to be related to the depth to the Bt (r=0.69) and the depth of the Ap (r=0.80) soil horizons. The position in the landscape was not related to the variation. Sample points that were physically closer together did not express variation in soil moisture that was statistically closer together. The use of Mahalanobis distance and MDS followed by correlation and/or regression analysis was an efficient and effective method of relating a set of highly variable observations to soil properties. This set of statistical procedures may also provide guidance in the design of a sampling array to adequately sample the spatial variation with fewer sample sites.