Abstract: Accurate prediction of soil moisture content is essential for preventing geological disasters and protecting the environment. Recently,numerous artificial intelligence algorithms have been applied to simulate and predict soil moisture content,providing valuable insights into its distribution and migration patterns. In this study,a multiple adaptive regression spline (MARS) data-driven model was developed to predict soil moisture content at various depths for the next 12,24,and 36 hours,using meteorological data along with soil heat and moisture measurements. Long-term observations were conducted in Yanjiao,Hebei Province,where fiber-optic sensing technology was employed to collect measured data. Based on these observations,a soil moisture profile prediction model was established. The results demonstrate that the model performs effectively in predicting soil moisture content across different time frames and depths. Furthermore,analysis of the relative importance of influencing factors reveal that soil temperature has a more significant impact on soil moisture content than meteorological factors. Spatiotemporal analysis indicate that shallow soil moisture content is strongly influenced by rainfall and other factors during the monitoring period,exhibiting relatively poor temporal stability. The fluctuation range in shallow layers is considerably larger than in deeper layers,resulting in better prediction accuracy during autumn compared to winter. The findings of this study can serve as fundamental data for engineering geology applications and provide a novel methodological approach for soil moisture content prediction.