Abstract: To address the insufficient characterization of non-stationary properties in ship-to-shore channels, this paper presents a comprehensive analysis and statistical modeling study based on broadband measurements at 720 MHz in maritime environments. First, the channel non-stationarity is quantitatively analyzed using a time-domain correlation coefficient method based on the Local Scattering Function (LSF). The measured average stationary time of approximately 1.35 s serves as the basis for determining the statistical window for parameter extraction. Subsequently, a joint non-stationary modeling method is proposed. Gaussian Mixture Models (GMM) cluster the Rician K-factor of the strongest path and logarithmic Root Mean Square (RMS) delay spread to classify the channel into dynamic scattering and stable Line-of-Sight (LoS) states. A two-state Markov chain model then characterizes the dynamic transitions, revealing that the stable LoS state dominates in near-shore regions while an alternating pattern emerges offshore. Finally, Tapped Delay Line (TDL) models are established for each state to capture their specific multipath fading characteristics. Simulation results demonstrate that the proposed model accurately reproduces the time-varying non-stationary features of ship-to-shore channels, providing theoretical support for the design and evaluation of maritime communication systems.