Abstract: In the southern part of the Ordos Basin, the Loess Tableland features a particularly thick loess layer.Seismic data usually exhibit a strong small-offset strong energy noise, as well as a shallow broadband strong refraction noise.The coherent noise suppression and the frequency-division abnormal amplitude attenuation are commonly utilized noise suppression methods in these cases.However, in practical applications, these methods do not perform well in terms of signal-noise separation, and they exhibit boundary effects and spatial spurious frequencies.The Loess Tableland is characterized by flat layers, small reflection dip angles, and limited energy variations in the effective signal during a recording time interval.Therefore, a data reconstruction method based on random function is utilized to transform the small-offset strong energy noise and the shallow broadband strong refraction noise into a random strong energy noise, which can be suppressed using the frequency-division abnormal amplitude attenuation.The application to experimental data showed that the method can solve the problem of noise suppression and improve the signal-to-noise ratio of the seismic data in areas of the Loess Tableland with thick loess layers.