Soil layer models created with profile cone penetrometer data

In creating soil layer models for our study site, we were challenged (i) to express vagueness of our soil data, while at the same time maintaining adherence to systematic classification principles, and (ii) to describe continuously the spatial distribution of soil materials and layers in three dimensions. We developed a method to create 3-dimensional (3-D) continuous soil layer models describing the distribution of soil materials, reworked loess vs. glacial till. Soil attribute data such as texture, bulk density and water content, in combination with penetration resistance obtained with a profile cone penetrometer on a 10-m grid, were used to describe soil materials and layers. We compared crisp hierarchical clustering with fuzzy k-mean classification in creating soil layer models for a 2.73-ha site in southern Wisconsin. The continuous 3-D soil layer models were developed using horizontal ordinary kriging and vertical linear interpolation. Validation proved that the crisp 3-D soil layer model predicted soil layers more accurately than the fuzzy 3-D soil layer model. We conclude that at the working scale, the crisp classification is superior to the fuzzy classification.