Application and improvement of soil spatial distribution mapping using advanced modelling techniques

Abstract

The main purpose of this contribution is to develop  realistic prediction digital soil maps in order to increase their visuality, and to evaluate and compare the performance of different modeling techniques: a) Kriging, b) Artificial Neural Network – Multilayer Perceptron (ANN-MLP) and c) Multiple Polynomial Regressions (MPR). The following  criteria were used to determine selection of the testing site for the modeling: (1) intensive metal ore mining and metallurgical processing; (2) geomorphological natural features; (3) regular geological setting, and (4) the remaining minefields. The success of Digital Soil Mapping and the plausibility of prediction maps increases with the availability of spatial data, the availability of computing power for processing data, the development of data-mining tools, geographical information systems (GIS) and numerous applications beyond geostatistics. Advanced prediction modeling techniques, ANN-MLP and MPR include geospatial parameters sourced from Digital Elevation Models (DEM), land use and remote sensing, applied in combination with costly and time-consuming soil measurements, developed and finally incorporated into the models of spatial distribution in the form of 2D or 3D maps. Innovative approaches to modeling assist us in the reconstruction of different processes that impact the entire study area, simultaneously. This holistic approach represents a novelty in contamination mapping and develops prediction models to help in the reconstruction of main distribution pathways, to assess the real size of the affected area as well as improving the data interpretation.