Abstract: Groundwater is one of the key factor affecting the stability of residue slope, and the prediction of groundwater depth in residue slope plays a significance role in analyzing the slope stability. A prediction model of groundwater depth, namely complementary ensemble empirical mode decomposition-random forest (CEEMD-RF) model based on phase space reconstruction, is proposed considering the highly non-stationary and nonlinear characteristics of groundwater level in residue slope. The CEEMD-RF model is applied to the groundwater depth prediction of SW2 hydrological observation hole of a residue slope in Guangzhou and the prediction results are compared and analyzed with that calculated by phase space reconstruction-RF model. The result shows that the fitting goodness of the groundwater depth predicted by CEEMD-RF model is 0.997, the root mean square error (RMSE) is 0.03 m, which is better than those using phase space reconstruction-RF model. The predicted groundwater depth series based on the CEEMD-RF model reflects the sharp point very well.