Abstract: The relationship model between the GPS multipath interferogram and soil moisture is a cri-tical factor for precise soil moisture monitoring. The traditional strategy is based on the linear model by collecting as many normal samples as possible during eliminating the outliers, however, it pays little attention to the factors which change slowly but affect the reflection environment such as slope, vegetation and weather. As the changes of these factors could be ignored in a short term, the time window is introduced. At first the window length is determined with correlation analysis, then the dynamic prediction and interpolation model could be realized by linear regression with samples within the window. The test results with real GPS, soil moisture and weather data show that, the prediction and interpolation error are reduced by 17.4% and 54.6%, and the correlation efficient are increased by 16.2% and 32.9% respectively. The interpolation is more accurate than prediction owing to the future samples, while the prediction model could be applied in real applications. The residual analysis show that the correlation between the epoch of maximal residue and soil moisture fluctuation exists. The maximal prediction residue is slightly weak and prior to the rises of soil moisture.