Geoderma (Oct 2024)
Unraveling the threshold and interaction effects of environmental variables on soil organic carbon mapping in plateau watershed
Abstract
Understanding the spatial distribution and mechanisms driving soil organic carbon (SOC) is crucial for assessing soil carbon stocks and implementing effective carbon sequestration strategies in agricultural landscapes. The linear and nonlinear relationships between environmental variables and SOC have been extensively documented, but the threshold and interaction effects among multiple covariates on SOC remain underexplored. This study focused on farmland within the Qilu Lake watershed in Yunnan Province, China, which is characterized by complex surface conditions shaped by both climate change and anthropogenic activities. Utilizing 216 soil samples from the watershed, this research aimed to investigate the threshold and interaction effects of environmental variables on SOC. To achieve this, gradient boosted decision tree (GBDT) combined with partial dependence analysis were employed to elucidate the spatial distribution of SOC and the intricate relationships between environmental factors and SOC. In order to enhance the accuracy of SOC prediction, we employed the landscape metrics as environmental variables, thereby facilitating a more comprehensive description of the landscape. The results indicated that GBDT (R2 = 0.47) outperformed random forest (R2 = 0.38), achieving higher accuracy and lower uncertainty, indicated by a narrower 90% prediction interval. The SOC distribution was predominantly influenced by soil moisture, elevation, and the contagion index (CONTAG), with threshold effects observed at relatively high soil moisture levels (>50%), CONTAG levels (>85%), and relatively low elevations (<1812 m). Moreover, the nonlinear relationship between one environmental variable and SOC could be influenced by another, suggesting interaction effects rather than a simple additive effect. Our findings suggest that combining GBDT modeling with partial dependence analysis provides an efficient and interpretable approach for SOC mapping. Knowledge of the threshold and interaction effects is critical for understanding the complex relationships between environmental variables and SOC, which has important implications for soil carbon management.
