Abstract: In traditional seismic reservoir prediction, especially for thin interbedded channel sand bodies, frequency-domain attribute analysis methods have achieved significant progress and extensive application. However, in shallow sedimentary environments such as the Penglaizhen Formation in the Sichuan Basin, where thin sand bodies approach the seismic resolution limit and feature multi-stage superposition and frequent lateral migration, frequency-domain attribute analysis methods are sensitive to background noise and limited in characterizing laterally discontinuous boundaries, and their thin-bed identification capability is gradually reaching a bottleneck. To address this issue, focusing on phase attributes that contain abundant geological information in seismic signals, we propose an instantaneous spectral phase decomposition method based on the improved generalized S-transform. Through high-precision time–frequency analysis, this method extracts instantaneous amplitude and phase spectra, and uses phase-domain decomposition to reconstruct seismic signals within specific phase intervals, effectively separating weak thin-sand responses masked by strong background noise. Applications in the Penglaizhen Formation of the western Sichuan Basin show that the instantaneous spectral phase decomposition method, by suppressing mudstone background interference, significantly improves the identification of narrow channels and multi-stage superimposed channel boundaries, overcoming the insufficient thin-bed prediction accuracy of conventional frequency-division RGB attribute fusion methods. The phase-divided amplitude attribute extracted by this method provides a phase-domain supplement for fine thin-sand prediction, demonstrating significant practical value.