Analysis of Soil Carbon Stock Dynamics by Machine Learning—Polish Case Study
Artur Łopatka,
Grzegorz Siebielec,
Radosław Kaczyński,
Tomasz Stuczyński
Affiliations
Artur Łopatka
Department of Soil Science Erosion and Land Protection, Institute of Soil Science and Plant Cultivation—State Research Institute, Czartoryskich 8, 24-100 Pulawy, Poland
Grzegorz Siebielec
Department of Soil Science Erosion and Land Protection, Institute of Soil Science and Plant Cultivation—State Research Institute, Czartoryskich 8, 24-100 Pulawy, Poland
Radosław Kaczyński
Department of Soil Science Erosion and Land Protection, Institute of Soil Science and Plant Cultivation—State Research Institute, Czartoryskich 8, 24-100 Pulawy, Poland
Tomasz Stuczyński
Faculty of Science and Health, The John Paul II Catholic University of Lublin, Konstantynów 1 H, 20-708 Lublin, Poland
A simplified differential equation for the dynamics of soil organic carbon (SOC) that describes the rate of SOC change (dSOC/dt) was constructed using the LASSO regression—a regularized linear regression machine learning method. This method selects the best predefined explanatory variables and empirically evaluates the relevant parameters of the equation. The result, converted into a formula for the long-term equilibrium level of soil carbon, indicates the existence of carbon sequestration potential in the studied regions of Poland. In particular, the model predicts high SOC content in regions with a high Topographic Wetness Index (TWI), such as river valleys or areas with high cattle density, as expected.
