Ground roll is a type of common regular noise and features low frequency,low velocity,strong energy,and dispersion.It overlaps with effective waves in terms of frequency bands,propagation direction,and velocity,making the transform-domain filtering methods fail to yield desired denoising results.At present,prediction plus matched subtraction is the mainstream ground roll suppression technique in the industry.However,direct predictive modeling of the ground roll is often difficult and of limited accuracy due to the complex wavefield propagation process near the surface.This study proposed a new ground roll suppression technique based on invertible superposition transformation.The main approach is as follows:first,the reflected waves can be modeled using a superposition operator in the CMP domain to construct an inverse problem for reflected wave estimation.Then the inverse problem is solved using an optimization method.The whole process is equivalent to the projection of seismic data to the reflection space.Since ground roll and reflection waves reside in different spaces,clean reflected wave data can be obtained through projection.To reduce the influence of ground roll on inversion results,this study used a data preconditioner to reduce the leakage of ground roll,achieving effective noise suppression effects.