Abstract: Traditional seismic reservoir prediction, especially for thin interbedded channel sand bodies, has made significant progress in frequency domain attribute analysis and has been widely used. However, in shallow sedimentary environments such as the Penglaizhen Formation in the Sichuan Basin, facing thin sand bodies with thickness close to the limit of seismic resolution, multi-stage superimposition and frequent lateral changes, the frequency domain method has gradually shown its bottleneck due to its sensitivity to background noise and limited ability to depict the lateral discontinuous boundary. On this basis, this paper explores a new research method, focusing on the phase attribute of seismic signals containing rich geological information, and proposes an instantaneous spectral phase decomposition method based on improved generalized S transform. This method extracts the instantaneous amplitude and phase spectrum through high-precision time-frequency analysis, and reconstructs the seismic signal according to the specific phase interval by using the phase domain decomposition technology, so as to effectively separate the weak response signal related to thin sand body covered by strong background noise. The application in Penglaizhen Formation in Western Sichuan shows that, compared with the "intermittent disappearance" phenomenon of the river channel in the traditional root mean square amplitude attribute map, the phase division amplitude attribute can significantly suppress the mudstone background interference, and clearly depict the narrow channel and multi-phase superimposed channel boundary that are difficult to identify by the conventional method. This method can quickly and conveniently realize the phase decomposition, which provides an effective supplement to the phase domain for the fine prediction of thin sand bodies, and has achieved significant practical application effect.