Abstract: Accurate slope reliability analysis of unsaturated earth and rock dams requires a combination of spatial variability in soil shear strength parameters and hydraulic parameters.Such multi-parameter problems often result in dimensional catastrophes due to too many variables discrete from the random field,and traditional analysis methods are difficult to apply.For this reason,the multivariate adaptive regression spline (MARS) method based on sliced inverse regression (SIR) is proposed to analyze the slope reliability of earth and rock dams.The method uses SIR to reduce the dimensionality,and the reduced variable matrix is used to construct surrogate models by MARS,and then Monte Carlo simulation (MCS) is used to evaluate the probability of failure of earth-rock dam slopes.The validity of the proposed method is verified by taking an earth and rock dam as an example,and the variation rule of the calculation results with the spatial variability of each parameter is explored,and the relationship between landslide magnitude and frequency is analyzed.The results indicate that the effect of shear strength parameter variability on dam slope failure probability is larger than that of hydraulic parameter variability;the proposed method can produce accurate calculation results under the variation of the coefficients of variation of each parameter and effectively reduce the cost of calculation;the method can provide an effective way to solve the problem of slope reliability of unsaturated earth and rock dams with high dimensionality and small failure probability.