This paper proposes a multi-level planning method for energy storage stations considering the spatial-temporal correlation of compound natural disasters. This is to improve the system resilience of a distribution network in severe weather. First, a joint disaster intensity probability distribution model considering the correlation between rainstorm intensity and wind speed in severe weather is established. Through the multi-dimensional spatio-temporal correlation analysis between the distribution characteristics of composite disaster intensity and the operational risk of distribution network components, a distribution network fault risk model considering the spatio-temporal correlation of composite natural disasters is constructed. Second, a multi-level planning strategy of energy storage power station to improve the operational flexibility of a distribution network under complex disaster weather is proposed. A planning model framework of energy storage power station considering the spatial and temporal distribution characteristics of disasters, including energy storage configuration, risk modeling and distribution network scheduling, is constructed. Based on Benders decomposition, the framework is transformed into a three-level mixed integer nonlinear programming problem for efficient solution. Through a simulation test and comparative analysis of typical distribution networks, it can be seen that the proposed method can effectively improve the operational flexibility and recovery ability of distribution networks to deal with fault risks in severe weather, and ensure the reliable power supply of important loads of distribution networks when there are disasters.