Autonomous hazard detection and avoidance is a key technology for future landing missions in unknown surface conditions. Current state-of-the-art stochastic algorithms assume simple Gaussian measurement noise on dense, high-fidelity digital elevation maps, limiting the algorithm's applicability. This paper introduces a new stochastic hazard detection algorithm capable of more general topographic uncertainty by leveraging the Gaussian random field regression. The proposed approach enables the safety assessment with imperfect and sparse sensor measurements, which allows hazard detection operations under more diverse conditions. We demonstrate the performance of the proposed approach on the existing Mars digital terrain models.