<p>Landslide-tsunamis are common geohazards in reservoir areas, presenting significant risks to nearby residents and infrastructure. Parameters like initial wave height and maximum wave amplitude are direct indicators of a landslide-tsunami's attack, making them crucial for reservoir area risk assessment. Regression models for the initial wave height and maximum wave amplitude are proposed based on a database using a modified multivariate power regression method. The importance of the factors in the models are evaluated using an adjusted <i>R</i>², and irrelevant variables are removed to refine the models. The influence mechanisms are explained based on the lift-up effect, push-ahead effect, and drag-along effect. The models proposed in this paper consider more cases and parameters, particularly landslide composition and morphology, resulting in better accuracy and applicability compared to existing models. The modified multivariate power regression method can further improve the accuracy of regression models compared to commonly used multivariate logarithmic regression. The lift-up effect helps to explain the influence of landslide volume, composition, morphology, and still water depth on initial wave height and maximum wave amplitude. The push-ahead effect explains the influence of landslide impact velocity, density, and morphology. The drag-along effect is mainly influenced by the landslide's length, width, and thickness. Notably, the length and thickness of the landslide significantly impact the maximum wave amplitude but have little effect on initial wave height. This is because the initial wave height is generated by the interaction of the base of the landslide's leading edge with the water. The proposed models were applied to the 1958 Lituya Bay and the 2013 Huangping landslide-tsunami events, demonstrating high accuracy and good applicability. This study aims to provide references for managing landslide-tsunamis in reservoir areas.</p>