A new approach to clustering soil profile data using the modified distance matrix

The application of different data mining methods for large soil profile datasets can be very useful for many agricultural and natural resource management applications, ranging from crop modeling to soil taxonomy. Distance or dissimilarity measures are key features of these methods. The proximity measure or the distance between vectors is calculated when they have the same dimensions. In the case of the soil profile data, the corresponding matrices representing different soils and their horizon properties usually have different dimensions. The objectives of this study were to explore a new approach for creating a semi-metric based on adjustment of the soil profile horizons, implement it in the computer application and to apply the modified distance matrix calculation to maximize the use of soil horizon properties for soil data mining. We assume that each soil horizon is homogenous while a vertical heterogeneity of soil profile is expressed through different soil horizons. Therefore, any sublayers of a horizon are characterized with the same values for its attributes as the horizon itself. In our approach, we developed matrices with the same dimension for soil profiles and calculated the proximity measure. The algorithm was implemented as an easy to use Fortran application that can calculate the modified distance matrix (MDM) for low- and high-dimensional soil profiles data. The proposed approach was shown to be effective when using existing reliable datasets, such as WISE Version 3.1, a global soil profile database developed by ISRIC. Hierarchical clustering was performed using the MDM based on the original algorithm of soil profile horizons adjustment with further integration into R. The principal finding shows that a proposed modified distance matrix can be used with different clustering methods for soil profile data clustering on a horizon-by-horizon basis. This study established a new methodology for using the modified distance matrix calculation and applying it with different clustering algorithms to large sets of soil profile data obtained from detailed soil surveys.