Search

Accurate mapping of soil thickness is crucial for understanding soil characteristics, erosion processes, and land management decisions. In this study, we estimated the spatial variation in the thickness of humus-accumulative genetic horizons (A and A + AB) and produced the map of erosion degree using the threshold values and the generated thickness maps (a model-based approach). The study area was a forest-steppe zone in Russia and encompassed four major soil types: Haplic and Luvic Chernozems, Grey-Luvic Phaeozems, and Fluvisols. Our approach incorporated machine learning techniques with environmental variables, such as relief, remote sensing data, climate, soil, and geology maps. We tested two spatial resolutions of covariates (30 and 250 m) to determine the modelling performance. The findings indicated that the average thickness of horizon A and A + AB was 37 and 50 cm, respectively, and exhibited variation in the following sequence: Fluvisols > Chernozems > Phaeozems soils. The analysis showed that soil type and topographic variables significantly contributed to the prediction of soil thickness, whereas 250 m resolution performed slightly better accuracy. The generated map of erosion degree revealed two dominant classes: non-eroded and weakly/slightly eroded soils. Our results provide valuable insights into the factors influencing the spatial prediction of soil thickness and highlight the significance of relief and legacy soil maps. The proposed approach for assessing soil erosion degree can become an alternative to the conventional field-based methods, which are time-consuming and expensive. Moreover, for the expeditiously-detected eroded areas, it is possible to plan or provide preventive soil and water conservation measures.