Applying an Artificial Neural Network (ANN) to Assess Soil Salinity and Temperature Variability in Agricultural Areas of a Mountain Catchment

Spatial analysis is currently a popular research tool, particularly in studies that focus on soil properties, and it is important for a comprehensive presentation of results by means of spatial statistics techniques. Spatial autocorrelation determines a degree of relationship between variables for two specific spatial units (locations). This relationship is reflected by spatial dependence of investigated soil properties. Moran's I was used as a measure of spatial autocorrelation. Positive spatial autocorrelation was determined for soil salinity (electrical conductivity) and temperature. Thus, the aim of the study was to identify the factors affecting spatial correlation of electrical conductivity (EC) and temperature in farmland and forest-covered areas. A model of artificial neural network was based on salinity, as salinity reduces the amount of nutrients and soil temperature, thus inhibiting plant root growth. Our study revealed that the most effective parameters determining soil temperature were EC and moisture content. The best results in the EC model were achieved for soil moisture content, temperature, and soil texture. Both soil parameters were impacted by catchment land use. Spatial analysis of soil properties and identification of factors affecting their diversity may be helpful in determining proper land use - particularly of sustainable agricultural practices in mountain areas.