The varying pattern of soil moisture in crop field environment was measured, analyzed and compared. The correlation coefficients between soil moisture and precipitation amount as well as between soil moisture and temperature were closely studied. The measured soil moisture data were nonstationary time series, which were quasiperiodic. The periodicity characteristic was varied with growth stages of spring maize. Precipitation amount was one of the most important stochastic environmental factors which interfered soil moisture. Oriented to this randomness feature, soil moisture was measured in different soil depths (10cm and 30cm, represented two typical types of soil moisture’s fluctuation) at four different maize growth stages. The soil moisture data measured in 10cm soil depth represented a volatile fluctuation, while that in 30cm soil depth was relatively stable. A model group for shortterm soil moisture prediction was introduced based on the GM(1,1) method. The model group was tested by using soil moisture data acquired from field. Results showed that compared with single prediction model, the prediction precision was evidently improved by corresponding submodel of the model group. The overall mean relative error was less than 2% for the model group. Comparing results of different combinations in soil moisture prediction using the model group, the result at maize seedling stage in 10cm soil depth was the best. Moreover, considering the influence of precipitation, parameter u in model group was optimized and adjusted, which further improved the performance of the moded group for soil moisture prediction.