Reconfigurable intelligence surface (RIS) consists of a large number of low-cost passive reflecting elements has attracted significant attention from communication communities, which is mainly due to its ability to substantially improve the communication performance. In light of this, we develop a three-dimensional (3D) geometry-based multiple-input multiple-output (MIMO) propagation model in this article, which is mainly suitable for describing the high-speed railway communications assisted by RIS in macrocell environments, such as train-to-train scenarios. In such propagation model, the waves emitted from the mobile transmitter experience non line of sight (NLoS) propagation links before reaching the mobile receiver, as well as the RIS propagation links before reaching the receiver. We compute the expressions of the RIS-assisted high-speed railway channel propagation properties, for example, spatial-time (ST) cross-correlation functions (CCFs), between different propagation links. Then, we study the temporal auto-correlation functions (ACFs) of the propagation links in the provided propagation model for distinct movement time instants. Analysis shows that the communication performance can be enhanced by rising the number and the size of RIS reflecting elements, and it is validated by numerical simulation results.