Abstract: The random noise in the desert zone has low-frequency characteristics in addition to nonlinear and
non-Gaussian characteristics． It exists in both analog records and actual desert seismic records． The effective
signal is largely submerged in the noise posing great difficulties to the processing of subsequent data． Based on
these problems，this paper combines the CEEMDAN (Complete Ensemble Empirical Mode Decomposition with
Adaptive Noise) with the ＲOSL(Ｒobust Orthonormal Subspace Learning)． The CEEMDAN algorithm is used to
decompose the desert seismic data． All the modalities decomposed are combined into a new record，and the new
record is subjected to low rank decomposition． Then superimpose all the modes of each channel in the obtained
sparse part to obtain the denoising result． The combination of the two solves the problem that the single low rank
processing has no obvious effect on the desert seismic data，and avoids the problem of choosing the modal of the
CEEMDAN algorithm decomposition． The simulation experiment and the actual data processing prove that the
proposed algorithm has obvious advantages in suppressing the low frequency random noise． The amplitude of the
effective signal can be guaranteed to be more than 85%． The suppression of the surface wave in the actual data
is relatively thorough．

Key words:  , complete ensemble empirical mode decomposition with adaptive noise ( CEEMDAN), robust orthonormal subspace learning (ＲOSL), random noise reduction, desert seismic signal