Short-term precipitation forecasting is a critical task in the fields of meteorology and hydrology. To overcome the limitations of traditional forecasting methods in handling complex meteorological phenomena and the issue of cumulative errors in image sequence prediction by recurrent neural networks, a short-term precipitation forecasting method called multi-scale attention encoding-dynamic decoding network (MAEDDN) has been developed. It predicts future precipitation by learning the spatiotemporal features of the input data. Within the encoding process, convolutional blocks with spatial and channel attention are utilized for encoding, and a multi-scale fusion module is employed to address the challenge of capturing both small-scale and large-scale information in precipitation distribution simultaneously. In the short-term precipitation processes to address the generation and dissipation. In the decoding process, a dynamic decoding network is proposed to flexibly select the decoding process based on the learned intensity distribution and change trends from the past input data Experiments are conducted by using the precipitation data from the open-source SEVIR dataset, and comparisons are made with the best methods reported so far. The experimental results reveal that: (1) MAEDDN enhances the forecasting capability in areas with high-intensity precipitation, and (2) MAEDDN outperforms other models in terms of the resolution of predicted image sequences. The constructed multi-scale attention encoding captures the complex relationships in meteorological data more effectively, while the dynamic decoding adapts the decoding process based on different scenarios, resulting in more accurate prediction outcomes.