Soil water dynamic change is significant to water cycle research and agricultural production. The least square support vector machine and the meteorological factors were used to train,test,and simulate soil water dynamic change in red soil region. Results showed that the least square support vector machine had more reliabilities and advantages of simulatio n perfor mance and mathematical meaning than the neural netw orks. Therefore, soil water dynamic change was simulated by the least square support vector machine and its trend was extracted by bior 3.3 with five layers of wavelet decomposition. The trend of soil water dynamic change can be divided into four stages which can provide a scientific basis for the water utilization and soil moisture prediction in the study region.