[Objective] Empirical rainfall thresholds have geographical limitations, it is of great significance to carry out a rainfall threshold modeling study for landslide hazard on the Chinese Loess Plateau. [Methods] Taking the Yanhe River Basin on the Loess Plateau as the research area, the landslide identification of the Yanhe River Basin based on InSAR was carried out by integrating Sentinel-1A, DEM, GPM rainfall and other multi-source data. Then the correlations between rainfall factors and landslide shape variables were calculated to determine the rainfall days and effective rainfall attenuation coefficient in the early period. Finally, the effective rainfall (E), rainfall intensity (I) and rainfall duration (D) were calculated, and the E-D, I-D and E-I rainfall threshold models were established by combining the power function. [Results] (1) A total of 107 landslides from 2017 to 2021 were identified, including the specific location and accelerated deformation time of landslides. (2) With the increase of rainfall days in the early period, the correlations between cumulative rainfall and landslide shape variables showed a trend of first increasing and then decreasing. When the attenuation coefficient of effective rainfall in the current period (K) was equal to 1, the correlation between the effective cumulative rainfall in nine days (the rainfall on the day of landslide occurrence and the rainfall in the previous eight days) and the landslide shape variable was the strongest. (3) The expressions corresponding to the E-D, I-D and E-I rainfall threshold models are E=12.19D^0.661,I=10.66D^－0.255 and E=3.84I^1.141, respectively. [Conclusion] The rainfall threshold model can provide a scientific basis and reference for meteorological early warning of landslide hazards in the Yanhe Basin of the Chinese Loess Plateau.