Quantifying Contributions of Climate and Rice Phenological Changes to SOC Dynamics Based on the Integration of Biogeochemical Model and Machine Learning Algorithm

Abstract

Climate and rice phenological change notably impact the dynamic changes in soil organic carbon (SOC) in arable land, but few studies have focused on quantitatively analyzing the impacts of these two factors. In this study, spatiotemporal dynamics of paddy SOC along the southeastern coast of Fujian Province in China were simulated by integrating the random forest (RF) algorithm and the DeNitrification–DeComposition (DNDC) model. Subsequently, based on the simulated raster time series of SOC, the contributions of climatic factors and rice phenological parameters to the dynamics of paddy SOC were quantified via a linear regression-based factor analysis method. The results showed that the integration of RF and DNDC led to a 53% accuracy improvement in R2 for predicting SOC spatial variations and a 162% improvement in R2 for predicting SOC temporal dynamics. Under the influence of climate change, 75.2% of the paddy fields served as carbon sinks from 2008 to 2021, with an average SOC density increase rate of 0.03 kg·m-2·a-1. The contribution of climatic factors and rice phenological parameters to the SOC dynamics accounted for 27.04% and 40.26%, respectively. The contribution of preseason climate to SOC dynamics was generally greater than that of midseason climate. Greater contributions of rice phenology to paddy SOC dynamics were primarily distributed at higher altitudes. In comparison, greater contributions of climate change to paddy SOC dynamics occurred mainly at lower altitudes. The results of this study could provide decision support for the formulation of farmland soil carbon sequestration and emission reduction measures.

Keywords: subtropical climate, seasonal differences, Machine learning, biogeochemical model, Digital soil mapping