In order to confirm the effectiveness and application value of high-frequency information outside of seismic wave pass band for sand body prediction, the convolution model was used to verify that seismic wave high-frequency is the basis of fine delineation of geological body spatial structure for sand body. De-noising processing is the key of retaining high-frequency information in the course of seismic data processing flow. The authors put forward the methods, which on the minimal impacting effective signal conditions, utilizing Curvelet transform that is characteristic by minimal overlap in Curvelet transform domain for effective signal and noise, attenuate noise and retain effective high-frequency components of seismic wave, and using the adjustable flexible and aggregation time-frequency properties of self-adapting S transform (SAST) to extract high-frequency information, obtaining high-quality single-frequency data for sand body prediction. The processing result on actual data in TH area showed along horizon lateral iso-frequency slices of high-frequency. Meanwhile, outside the pass band, the spatial distribution and boundary of the sand body was clearly depicted.