Abstract: Soil strength parameters govern tunnel face stability, and in natural soils they exhibit spatial variability and depth-dependent trends. This study models soil parameters using a non-stationary random field and investigates the effect of spatial variability of the internal friction angle in sandy soil on tunnel face collapse using the smoothed particle hydrodynamics (SPH) method. Both linear and non-linear depth-dependent variations are considered to better represent soil heterogeneity. The SPH method and random field generation are validated through benchmark comparisons. Collapse distance, mass of inflowing soil, and maximum surface settlement are adopted as evaluation metrics, and Monte Carlo simulations are performed to assess the effects of vertical correlation length and coefficient of variation. Results show that increasing vertical correlation length and coefficient of variation significantly enhances the uncertainty of tunnel face collapse. Non-linear depth-dependent variation leads to larger collapse scales than linear variation, and neglecting this feature results in underestimation of collapse extent.