An Improved Land Use Methodology Based on Machine Learning for Predicting the Spatial Distribution of Contaminants in Urban Surface Sediments

One of the effective approaches to modeling the spatial distribution of an object is the land use regression (LUR) method, which allows obtaining estimates with high spatial resolution. The method involves constructing a mathematical model based on experimental data and geographic information systems (GIS) data. In this study, we propose using machine learning (multilayer perceptron) as a regression model. We also propose an improved methodology for constructing ring spatial variables around the studied geolocation. The data for the study were obtained during the screening of urban surface deposited sediments (USDS) in the city of Murmansk (Russia). The model using the improved methodology for constructing ring spatial variables turned out to be more accurate.