<p>Precipitation nowcasting is crucial for disaster prevention and mitigation. This paper proposes a novel encoder–decoder model—Spatial attention and dual-path gated network (SA-DGNet), which better identifies precipitation regions and estimates intensity. First, we introduce the Local window self-attention block (LWSA), which implements self-attention with relative positional biases within non-overlapping local windows, alleviates the limitations of modeling local dynamic spatial correlations under the fixed receptive fields of convolutions. Its linear computational complexity enables rapid processing of complex meteorological time series in large-scale, high-performance computing environments. Next, the Bidirectional path residual block (BPRB) integrates standard convolutions with dilated convolutions to facilitate multiscale feature extraction and cross-domain fusion, alleviating the common shortcomings in most models concerning the representation of convection generation, dissipation, and deformation; Furthermore, the Gated skip attention block (GSA) uses gated skip connections to alleviate the issues of low-level spatial detail loss, blurred precipitation prediction boundaries, and morphological distortion caused by traditional skip connections in U-Net. We evaluated SA-DGNet on radar precipitation maps from the Netherlands; it achieved a 6.6% improvement in critical success index over UNet on the NL-50 dataset, reaching 0.7427.</p>