<p>Rock engineering problems are mixed uncertainty problems with the simultaneous presence of stochastic (or, random variables), and non-stochastic (intervals/fuzzy/p-boxes) inputs due to their varying information levels. This study presents a hybrid reliability methodology capable of considering the spatial variations of stochastic/non-stochastic inputs. A multi-dimensional convex model-based method is initially presented to model the spatial variations of bounded variables via bounded fields. These bounded fields are shown to be the intersection of multiple convex ellipsoids constructed via available data. The method is further generalized to consider the spatial variations of fuzzy/p-box inputs by discretizing them into bounds. Finally, this method is coupled with the random field theory to consider spatial variations of both stochastic and non-stochastic inputs. The methodology is implemented via FLAC-2D, by generating random and bounded fields of stochastic and non-stochastic inputs using Expansion Optimal Linear Estimator (EOLE) and Non-Probabilistic Series Expansion (NPSE), respectively. The methodology is demonstrated for a rock slope by modelling inputs via different uncertainty models based on their information levels. The effect of incremental increase in the information levels of inputs on the FOS (Factor of Safety) estimates of the slope was also investigated via additional analyses. Finally, a systematic parametric study was performed to investigate the effect of input imprecision on the FOS. The method could truthfully propagate the input uncertainties to estimate the intervals/fuzzy/p-box of FOS. The information levels of input rock properties and the FOS were directly correlated. The reduction in the impreciseness (interval lengths) of inputs, especially UCS and GSI, reduced the imprecision of FOS.</p>