Due to its unique macropores and weak cementation structure, loess shows significant water sensitivity and dynamic vulnerability compared with other soil types. In China, the distribution of loess is very extensive, especially in the Loess Plateau, where the crustal activity is frequent. Active tectonics and frequent strong earthquakes can easily lead to landslides and other disasters, resulting in significant casualties and economic losses. Studies have shown that ground motions with different spectra have a significant impact on slope stability. Identifying the main frequency bands that cause slope instability is crucial for assessing the stability of loess slopes and helping to effectively prevent disasters such as landslides.

Based on the field survey data of loess topography and geological conditions in Xiji-Haiyuan-Guyuan area, this paper constructs a typical generalized numerical model of slope site. The model is divided into three layers: basement, mudstone and loess. Each layer of soil is regarded as isotropic and simulated by Mohr-Coulomb material model. Through a large number of geotechnical experimental data, the parameters of different soil layers in the model are accurately assigned. On this basis, the appropriate dynamic artificial boundary is selected, and the actual record of Kobe seismic wave is used as the original seismic data. By screening different frequency bands （0−2, 2−4, 4−6, 6−8, 8−10 Hz）, the filtered acceleration time history amplitude is adjusted to make the five acceleration peaks consistent as input waves. At the same time, according to the monitoring requirements, multiple sets of monitoring points are set up, and the influence of ground motion spectrum characteristics on the seismic response of loess slope is explored by observing the maximum velocity, maximum acceleration and maximum displacement time history data curves of each monitoring point.

The results show that: ① The slope acceleration is significantly affected by the input frequency of ground motion, and the peak acceleration amplification coefficients of different monitoring points under different frequency inputs of ground motion show different trends. In addition, the velocity and displacement are also significantly affected by the input frequency of ground motion, and the peak velocity and peak displacement of each monitoring point have the same trend with the input frequency. ② Low-frequency ground motion is the key frequency band to induce slope instability, and the slope is not easy to fail under high-frequency vibration. The slope has an amplification effect on low-frequency seismic waves, while it shows a certain ‘filtering’ effect on high-frequency seismic waves. ③ With the increase of elevation, the displacement of the slope increases, and the lower the input frequency of the ground motion, the more obvious the effect of this increase. These findings can provide a scientific basis for the stability assessment and disaster prevention of loess slopes.