Abstract: Shearlet transform serves as a tool extensively applied for random noise attenuation in seismic data. However, it faces critical drawbacks: (1) the over-attenuation tendency of valid signals due to its traditional hard-threshold denoising method, especially the loss of weak signals, and (2) ignoring the spatial correlation of Shearlet coefficients. To address these issues, this paper proposed an adaptive-threshold denoising method for seismic data based on the spatial correlation of shearlet coefficients. Specifically, this method constructed a neighborhood correlation model between directional sub-bands in the scale dimension and established a cross-scale coefficient mapping relationship in the spatial dimension. Based on the correlation constraints, an adaptive threshold matrix with physical significance was derived. Theoretical calculation and experimental validation demonstrate the superiority of the proposed adaptive-threshold algorithm in enhancing the quality of seismic data over traditional hard-threshold methods. Specifically, it retained effective signals to the maximum extent during random noise attenuation and significantly improved the amplitude preservation of weak seismic events.