Creating a Soil Map Using Digital Soil Mapping on the Example of the Diurtiulinsky Municipal District

Abstract

The study describes the use of machine learning methods, geostatistics, etc. in establishing soil properties depending on various classes of soil. The most commonly used data are information from spectral reflectance bands of satellite images and terrain models. Besides, there is also great potential for creating new data tiers. The study relies on a method known as SCORPAN-SSPFe, which assumes spatial error autocorrelation as a standalone function. This method is actively used in places where there is not enough information about soil data. Besides, four types of interpolation were compared using the SCORPAN method: multiple linear regression, cubistic model, cubistic model with kriging and random forest model, which use extensive but common values of soil properties associated with soil classes. The research result is obtained by applying the method to conduct large-scale soil surveys, which determines the purpose and relevance of our study.