Accounting for extensive topographic and pedologic secondary information to improve soil mapping

Soils generally vary in a quite complex way. To assess and model spatial variability, several variables can be sampled in addition to the target used to quantify and describe the main phenomenon under study. One of the major advantages of geostatistics over simpler spatialization methods is that sparsely observations of the primary attribute can be complemented by secondary attributes that are more densely sampled. Two methods were applied to incorporate dense secondary information: multicollocated cokriging, which restricts the neighbourhood to the only secondary data collocated with the available data of the primary variable and simple cokriging with varying local means related to crisp classes. The objective of this paper was to find the method that best improves the estimation of primary attributes through dense secondary information for the study area, which was the province of Siena in central Italy, an area of about 3820 km2. Soil samples were taken at 742 locations and depth, sand and clay contents and available water capacity were determined.A multivariate geostatistical analysis was performed using the four soil properties and elevation from a digital elevation model of 20 × 20 m resolution as auxiliary variable. The variables were interpolated using three different approaches: cokriging, multicollocated cokriging, multicollocated simple kriging with local mean, estimated by averaging over the soil unit occurring at each node of the interpolation grid.These methods were compared in terms of precision, through cross-validation, and of accuracy, through a validation test, using an independent data set of 170 soil depth measurements. The results did not show clear differences among the methods.