Recently, reconfigurable intelligent surface (RIS) has received much attention to improve spectral efficiency and coverage by deliberately adjusting the phase shifts. In reality, due to the mobility of users, the stationary RIS might not provide a persistent LoS link. In such cases, RIS on the unmanned aerial vehicle (UAV-RIS) can provide a reliable LoS link since it is operated at a high altitude. In this paper, we propose a two-stage channel estimation technique for the UAV-RIS communication systems. Key idea of the proposed scheme is to decompose the UAV-RIS reflected channel into two components: slow-varying angles and fast-varying path gains, and then estimate these components in different time scales. Further, by configuring phase shift vectors for the UAV-RIS reflected channel estimation, we can improve the channel estimation accuracy. In doing so, the proposed scheme can accurately estimate the channel while minimizing the pilot overhead. From the simulation results, we demonstrate that the TUCE-PS significantly improves the NMSE over the conventional channel estimation techniques.