Predicting the Spatial Distributions of Elements in Former Military Operation Area Using Linear and Nonlinear Methods Across the Stavnja Valley, Bosnia and Herzegovina

Abstract

This study has the purpose of developing a realistic soil prediction maps of the spatial distribution of elements by evaluating and comparing different modelling techniques: Kriging, artificial neural network-multilayer perceptron (ANN-MLP) and multiple polynomial regressions (MPR). The Stavnja Valley was selected as a test area due to the following reasons: (1) intensive metal ore mining and metallurgical processing; (2) peculiar geomorphological natural features; (3) regular geological setting, and (4) the remaining minefields. Geospatial parameters from digital elevation models (DEM) are used as an input to advanced prediction modelling techniques: ANN-MLP and MPR. Soil measurements, land use data, and remote sensing are applied, developed, and finally incorporated into the models of spatial distribution in the form of 2D or 3D maps. In order to reconstruct the different processes that influenced the entire study area simultaneously, we used novel approaches to modelling. This comprehensive approach not only represents an innovation in contamination mapping, but developed prediction models also help in the reconstruction of main distribution pathways, assess the real size of the affected area, and improve the data interpretation.

Investigations of Chemical Element Distributions in Soil, North Macedonia—A Review

“…In addition, MLP can be used to prove the classification of linear, inseparable patterns. MLP represents feedforward neural networks (FNNs), with multiple layers of units between the input and output layers [150]. Compared to other commercial and freely available software packages that allow users to implement neural networks relatively easily by offering only a limited number of training algorithms, the MLP offers an unlimited number of algorithms to achieve better results [153].…”

“…This could possibly be solved by increasing the number of sampling sites or using a denser sampling grid. [67,149,150] The Artificial Neural Network-Multilayer Perceptron (ANN-MLP) was used to improve the reconstruction of the main distribution pathways, assess the real size of the affected area and improve the interpretation of the data at two selected sites. This advanced predictive modelling technique combined soil measurements with data from DEM, land cover data and remote sensing developed by Alijagić [149,150].…”

Research on Remote Sensing Geochemical Modeling Based on Surface to Surface Model

“…The surface geochemical exploration sampling data in the study area is a small range of point data, which is about 20 cm × 20 cm squares with an interval of 100 m × 40 m. Compared with ground sampling points, remote sensing data can be thought as surface data with continuous surface features (Figure 2) [16]. Simply using the ground sampling point data to represent the surface data of remote sensing image IOP Publishing doi:10.1088/1742-6596/2597/1/012013 4 will cause large deviations and reduce the precision of the model.…”