Abstract: To balance computational efficiency and accuracy in calculating excavation volumes in reservoir perimeter expansion during the site selection phase of pumped storage power stations, this paper proposes a geometric-probabilistic hybrid model. First, a sector-weighted geometric excavation model is established. This model employs a unified reference excavation width to account for topographic constraints in various azimuthal directions, iteratively solving for the effective excavation volume required to achieve the target reservoir capacity. Second, a one-dimensional lognormal distribution is introduced to characterize the elevation-dependent variability of weathered layer thickness. Locally weighted linear regression is adopted to estimate the distribution parameters, and analytical-numerical expressions for the expected cross-sectional areas of distinct weathered layers are derived, enabling probabilistic estimation of stratified excavation volumes. A case study of a specific pumped storage power station is conducted, comparing the proposed method with the design-recommended scheme and the conventional average thickness method. Results indicate that the overall error in total excavation volume is constrained within 20%. Moreover, in estimating the excavation volumes for individual weathered layers, the proposed method significantly outperforms the average thickness method. This model has been integrated into an software for optimizing scale parameters of pumped storage power stations, providing effective technical support for scheme comparison during the site selection stage.