Mapping of Rill Erosion of the Middle Volga (Russia) Region Using Deep Neural Network

Soil erosion worldwide is an intense, poorly controlled process. In many respects, this a consequence of the lack up-to-date high-resolution maps. All over world, problem insufficient information solved in different ways, mainly on point-by-point basis, within local areas. Extrapolation results obtained locally to more extensive territory produces inevitable uncertainties and errors. For anthropogenic-developed part Russia, especially urgent because assessment intensity processes, even with use models, does not reach necessary scale due all required global large-scale remote sensing data complexity considering regional features processes such vast This study aims propose new methodology for automated mapping rill networks based Sentinel-2 data. A LinkNet deep neural network DenseNet encoder was used solve mapping. The recognition area than 345,000 sq. km were summarized grid 3037 basins analyzed assess relationship main natural-anthropogenic factors. Generalized additive models (GAM) model dependency density explore complex relationships. nonlinear between topographic, meteorological, geomorphological, anthropogenic factors shown.