Underwater optical wireless communication (UOWC) systems have been widely researched to achieve high-speed and secure wireless communications. The non-line-of-sight (NLOS) UOWC system that uses the water surface to reflect signal light is widely studied to overcome the line-of-sight (LOS) channel limitation, particularly the channel blockage issue by marine biology or complex underwater topography. However, most previous NLOS UOWC studies have assumed a flat water surface or a general sine or cosine surface wave model for simplicity, leading to inaccurate performance estimations. In this paper, we build a theoretical NLOS UOWC framework with the Pierson wave model which considers both spatial correlation and time relativity information of wave incorporating wind speed, and investigate the signal-noise-ratio (SNR) and bit-error-rate (BER) performance. Results show that compared with the previous flat surface, the wavy surface can reduce the probability of achieving a satisfying signal level by up to 70%, affecting the performance of NLOS UOWC systems. Furthermore, we investigate the multiple-input-multiple-output (MIMO)-based NLOS UOWC under wavy surfaces. Results show that the MIMO principle can reduce the impact of the wavy surface, where the probability of achieving a satisfying signal level can be increased by up to 50% using the 2 × 4 MIMO configuration. However, results also show that further increasing the number of receivers may not further improve the system performance. The proposed model enables more accurate design and analysis of NLOS UOWC systems by accounting for the overlooked impact of wavy surfaces.