Dynamic modeling of topsoil organic carbon and its scenarios forecast in global Mollisols regions

Abstract

Global Mollisols regions play crucial roles in maintaining world food security, performing ecosystem services and global carbon cycling. Little is known about the spatial and temporal variability of soil organic carbon (SOC), future trends, and the driving forces of changes in each Mollisols region. The aim of this work is to trace the SOC dynamics during the 2000–2020 period and forecast its dynamics for 2020–2040 period (every 5 years) across the global Mollisols regions, highlighting the driving factors for these changes. We used bare soil images and environmental covariates to construct a robust prediction model of SOC using machine learning. The best model was used to forecast future changes in SOC content under different emission scenarios. Also, the role of different predictor variables in predicting and driving SOC changes is indicated with the help of the shapley value algorithm. Results show that SOC content in the current global Mollisols regions is 26.45 g kg−1 in Ukraine, 21.34 g kg−1 in Northeast China, 21.05 g kg−1 in the USA and 15.59 g kg−1 in Argentina. However, they have been decreasing since 2001 and showed a slowing trend, with the highest and lowest rates of 11.45% in Ukraine and 7.33% in the USA, respectively. The forecast under different climate scenarios shows a slight decrease in SOC content within a range of 0.46%–1.22%. Among them, SSP 245 is more consistent with the current trend of SOC content, while SSP585 indicates the greatest loss of SOC content. Results revealed that the driving factors for SOC prediction in descending order were meteorologic, parent material, terrain and vegetation. All soil and meteorology variables contributed by 71%, while soil texture fractions and temperature were the top 20% of these variables. The main factors driving SOC changes from high to low latitudes are temperature, precipitation, and vegetation. It can be concluded that the temporal transfer model is a powerful tool to trace and forecast the dynamics of SOC in Mollisols regions at the global scale, enabling a clear understanding and ranking the factors affecting this process under diverse climate conditions.