Abstract: Soil moisture can be a very important indicator to monitor agricultural drought. The dynamics of water distribution in the soil can greatly contribute to the decision-making on the soil and water resources. This study aims to reveal the spatial and temporal variation characteristics of soil moisture in the farmland, and then accurately predict the soil moisture. The soil information fixed-point monitoring was carried out in a typical Wheat Grains field of Taihang mountain front and plain region in Hebei province, China. A total of 150 sampling points were designed from April 19th to May 9th 2017. A wheat field with a width of 100 m and a length of 50 m was meshed by 10 m for the soil sampling in each mesh grid. Soil samples were collected from the depth points at 0-20, >20-40, >40-60, and >60-80 cm six times during the sampling. The temporal stability index and spatial autocorrelation evaluation were then used to determine the spatial and temporal distribution of soil moisture. The CA-Markov model was constructed to predict the spatial and temporal variation of soil moisture in the field. A comparison was made on the prediction with the HYDRUS model. A field test of data detection was finally conducted to verify the accuracy of the simulation. The results showed that the isoline map of soil moisture was changed from dense to sparse with the increase in soil depth. There was the largest variation in the surface soil moisture. The variation coefficient of soil moisture also decreased gradually, as the soil depth increased. Specifically, the proportions of temporal stability index for the soil moisture less than 10% accounted for 52%, 58%, 60%, and 74% in the soil layers of 0-20, >20-40, >40-60, and >60-80 cm, respectively. Correspondingly, there was high temporal stability with the increase in soil depth. Furthermore, irrigation was an important factor with a strong spatial correlation under humid conditions, thus influencing the spatial distribution pattern of soil moisture. A spatial analysis was also conducted using the Moran's I statistic. It was found that the global Moran's I index increased firstly and then decreased with the growth period of wheat. Particularly, the global Moran's I index of relative humidity in the last three times was 0.064-0.142 smaller than that in the first three times, indicating the weak autocorrelation caused by the soil structure. Once the degree of drought reached a certain level, there was a decreasing trend in the spatial autocorrelation of soil moisture and relative humidity. A CA-Markov model was constructed to simulate the change of drought grade, according to the characteristics of soil relative moisture. The average area error was 1.61% for each grade of soil relative moisture, which was 9.25% smaller than that (10.86%) by the HYDRUS model. At the same time, the CA-Markov model was used to simulate the drought grade of soil moisture in late April and early May. The Kappa coefficients of predicted spatial distribution were 89.31% and 91.46%, respectively. Both the Kappa coefficients were higher than 75%, indicating an excellent performance of the improved model on the prediction of soil moisture distribution. The findings can provide a strong reference for crop growth and irrigation water management.