<p>Subaerial landslide-generated (SLG) waves involve complex processes with multi-factors, including landslide geometry and motion as well as valley landform. The uncertainty from prior empirical wave height models with the best fitting increases and therefore fails to provide reliable predictions when waves generated by landslides containing hard-to-obtain governing parameters. We feed into a Bayesian method to optimize the empirical SLG maximum wave height model with trade-off goodness of fit and complexity based on the formation-mechanism of the SLG wave from laboratory experiments. Furthermore, the effectiveness and practicality of the proposed method are respectively demonstrated by the past events in the literature and the Ridi landslide. The performance function of the former is record-wave-based developed and that of the latter is determined based on hazard-bearing bodies. Meanwhile, the uncertainty of water level and slide volume is, for the first time, quantified using dense samples. Eventually, we merge the optimized model with wave propagation to forecast the probability of wave heights exceeding hazard-bearing bodies in the Ridi landslide. Outputs of the optimized model have a high robustness indicated by six diverse reservoir landslides and the 95% confidence interval of prediction can also accommodate the impact of record deviation. Exceedance probability maps considering multi-uncertainly provide a reliable prediction for the risk assessment framework of the landslide hazard chain. Without loss of generality, the proposed method can serve for rapid and holistic preliminary assessments of the SLG wave hazards.</p>