This paper armed to optimize soil moisture monitoring by taking hill slopes in a tea garden and a bamboo forest located in Gaochun district of Nanjing City as examples and monitoring soil moisture in long-term. Based on the temporal stability and factor analysis, representative sampling sites were selected to predict soil water contents for other sampling sites by building stepwise regression models, and then checked the predication accuracy. The results showed: when only monitoring soil water content at 7 representative sampling sites in tea garden, RMSE of prediction was ≤1.5 cm³/cm³. In addition, while only monitoring soil water content at 5 representative sampling sites in bamboo forest, RMSE was ≤1.7 cm³/ cm³. This method can reduce the number of soil moisture monitoring sites in predicting soil water content on hill slopes with limit observations. In addition, land use type and soil depth can affect soil moisture characteristics. Bamboo forest had stronger temporal stability and higher spatial autocorrelation in soil moisture than tea garden, however, the performance of regression models for bamboo forest was worse than those for tea garden. It is noted that the spatial structure of soil moisture at 30 cm depth was more stable than that at 10 cm depth, implying the performance of regression models is better at 30 cm than at 10 cm depth.characteristics .The temporal stability of soil moisture in bamboo forest is stronger than that in tea garden as well as the relation of spatial. However, the performance of regression models for bamboo forest is generally worse than those for tea garden. It is noted that the spatial structure of soil moisture at the depth of 30cm is more stable than that at 10cm, which implies the performance of regression models of 30cm is better than those of 10cm depth.