This paper investigates the anti-jamming three-dimensional (3D) trajectory design to safeguard the legitimate communications of unmanned aerial vehicle (UAV)-enabled wireless sensor networks (WSNs). Specifically, a UAV is dispatched to collect data over multiple ground sensors (GSs) in the presence of a malicious ground jammer. Under the practical probabilistic line-of-sight (LoS) channel model, we aim to maximize the minimum (average) expected rate among GSs over a finite flight period, by jointly optimizing the GS transmission scheduling, the UAV horizontal and vertical trajectories. To make the formulated non-convex problem tractable, we provide an achievable lower bound for the expected rate, based on which an efficient iterative algorithm is proposed to solve it suboptimally by exploiting the block coordinate descent and successive convex approximation techniques. Numerical results show the significant rate improvement of the proposed joint design and provide new insights into the anti-jamming 3D trajectory design under the probabilistic LoS model, as compared to the conventional two-dimensional (2D) UAV trajectory or under LoS channel models.