Carbon (C) and nitrogen (N) cycles of terrestrial ecosystems play key roles in global climate change and ecosystem sustainability. In recent decades, climate change has threatened the nutrient balance of dryland ecosystems. However, its impact on soil organic carbon (SOC) and soil total nitrogen (STN) in drylands of China are still unclear. In this study, the structural equation model (SEM) was used to explain the relationship between environmental variables used by the best model and SOC or STN. Then Adaptive Boosting Regressor (AdaBoost), Gradient Boosting Regression (GBRT), Extreme gradient boosting Regression (XGBoost) and Random Forest Regression (RF) were used to establish the prediction model of SOC and STN based on soil samples along with environmental variables. The performance of these models was assessed based on a 10-fold cross-validation method using three statistical indicators. Finally, we predicted the SOC and STN of soil samples from 2000 to 2019 based on the best model. Overall, the RF model performed better at predicting SOC and STN in drylands than the other three prediction models (AdaBoost, GBRT, XGBoost). Climate factors were the main factors affecting SOC and STN in the study area. In the Alashan, a dryland in northern China, the precipitation in the growing season increased from 2000 to 2019, at a rate of 12.9 mm/decade. During the same period, the annual sunshine duration significantly decreased by 66 h/decade. Along with interannual hydrothermal variability, SOC showed a fluctuating upward trend at a rate of 0.04 g/kg/decade, while STN exhibited a fluctuating downward trend at 0.003 g/kg/decade from 2000 to 2019. Due to the effects of climate change, dryland were considered as potential sites for carbon sequestration. However, due to the annual hydrothermal variance causing dynamic annual changes, it was deemed unstable. Moreover, it would cause STN loss, which might reduce soil fertility. More attention should be paid to STN monitoring in dryland in the future.