The physical layer security (PLS) of a space-ground communication system is examined. To improve the security performance, a pair of reconfigurable intelligent surfaces (RISs) is integrated into the system and benchmarked against a scheme, where there is only a single RIS close to the ground station. As for the double-RIS scenario, we formulate a secrecy rate maximization problem, and then propose an alternating optimization (AO) algorithm for jointly optimizing three vectors, namely the beamformer of the ground station and the reflecting vectors of two different RISs. Similarly, as for the single-RIS case, we also propose another AO algorithm for optimizing a pair of vectors, namely the beamformer of the ground station and the reflecting vector of the single RIS. Both the double-RIS and the single-RIS AO algorithms are developed on the basis of the first-order Taylor expansion and the Dinkelbach's method, which allow us to approximate non-convex optimization problems by convex ones. Our results demonstrate that the proposed double-RIS scheme outperforms the single-RIS benchmark scheme in terms of its security.