Drivers of Soil Organic Carbon Spatial Distribution in the Southern Ural Mountains: A Machine Learning Approach

• Published in: Eurasian soil science Vol. 57; no. 11; pp. 1942 - 1949
• Main Authors: Suleymanov, A. R., Suleymanov, R. R., Belan, L. N., Asylbaev, I. G., Tuktarova, I. O., Shagaliev, R. D., Bogdan, E. A., Fairuzov, I. I., Mirsayapov, R. R., Davydychev, A. N.
• Format: Journal Article
• Language: English
• Published: Moscow Pleiades Publishing 01.11.2024; Springer Nature B.V
• Subjects: Abies sibirica; Application of Modeling to Assess and Forecast Changes in Soil Carbon Stocks; Betula pendula; Bioclimatology; climate; Cloud cover; Coniferous trees; data collection; Decomposing organic matter; Digital Elevation Models; Earth and Environmental Science; Earth Sciences; Elevation; Environmental factors; forests; Geotechnical Engineering & Applied Earth Sciences; Humus; Land use; Learning algorithms; Leptosols; Machine learning; Mountains; Organic carbon; Phaeozems; Picea obovata; Pine trees; Pinus sylvestris; Precipitation; Prediction models; Random variables; Remote sensing; Root-mean-square errors; Soil; Soil formation; Soil layers; soil organic carbon; Soil temperature; Soil types; Spatial data; Spatial distribution; Surface temperature; Vertical distribution; Ural Mountains; digital soil mapping; remote sensing data; random forest; climate; relief
• ISSN: 1064-2293; 1556-195X
• DOI: 10.1134/S1064229324602014

This study aims to assess the relationships between SOC content and main soil-forming factors and identify key factors explaining the spatial distribution of SOC. The research was conducted in the Southern Ural Mountains throughout 420 km from north to south in the Republic of Bashkortostan. The predominant soil types are mountainous gray forest (Eutric Retisols (Loamic, Cutanic, Humic)), dark gray forest (Luvic Retic Greyzemic Someric Phaeozems (Loamic)) soils, and gray-humus lithozems (Eutric Leptosols (Loamic, Humic)). Forest stands are mainly composed of birch ( Betula pendula ), pine ( Pinus sylvestris ), spruce ( Picea obovata Ledeb.), and fir ( Abies sibirica Ledeb.). A data set of 306 soil samples taken from the top layer (0–20 cm) was studied using the “random forest” machine learning method. Ninety four spatial environmental covariates were used as explanatory variables, including remote sensing data, climate (temperature, precipitation, cloudiness, etc.), digital elevation model and its derivatives, land uses, bioclimatic zones, etc. The results showed that the SOC content varied widely from 0.8 to 32%. The random forest predictive model explained 55% of SOC variation ( R 2 ) with a root mean squared error (RMSE) of 1.35%. Key variables included surface temperature, absolute elevation, precipitation, and cloudiness, which together reflect the Dokuchaev vertical and horizontal zonality laws. The findings emphasize the importance of considering multiple environmental factors in subsequent research focused on assessing the spatial distribution of SOC.