The effectiveness of using changes in environmental conditions to explain the spatiotemporal variability in soil organic carbon (SOC) with digital soil mapping (DSM) requires investigation. In this study, temporal variables representing temporal patterns of climate, vegetation, and land cover factors were explored. Models to predict SOC stocks were developed using a random forest algorithm and data from China during two periods (the 1980s and 2010s). We forecasted and hindcasted the developed models and assessed their temporal projections against temporally independent data. Models were developed for both periods using different sets of variables (with/without temporal variables), and their temporal projections were compared. The important temporal variables were identified by applying the recursive feature elimination algorithm. The results showed that the performances of temporal projections for the 1980s and 2010s were improved by approximately 17% and 47%, respectively, when temporal variables were included in the models. Spatially, the maps of changes in SOC stocks derived from the models that included temporal variables presented stronger associations with temporal changes in climate, vegetation, and land cover than those derived from the models that did not include temporal variables. This work highlights that variation in SOC stocks can be linked to temporal patterns of environmental factors. The findings also provide evidence that the application of temporal patterns of environmental factors to DSM models can be useful for the large-scale prediction of changes in SOC.