<p>Rapid and accurate landslide mapping following major rainfall or earthquake events is essential for emergency response, hazard mitigation, and post-event management. Existing deep learning-based landslide mapping models often achieve high accuracy at the cost of increasingly complex architectures and substantial computational demands, which limit their scalability. Rapid-response mapping therefore requires models that maintain strong accuracy while enabling more computationally efficient and scalable deployment without reliance on large models or resource-intensive hardware. To address this challenge, the Landslide Mapping Network (LDM-Net) is introduced as a computationally efficient Mamba-based network enhanced with a Global Vision State Space module. Designed for very-high-resolution (VHR) optical imagery, LDM-Net captures multi-scale contextual information while maintaining favourable efficiency for large-area deployment. The model was evaluated using the newly developed Unmanned Aerial Vehicle Landslide Mapping Dataset (ULMD), derived from 0.2 m UAV imagery, together with several public benchmarks. Compared with Transformer-based baselines, LDM-Net required only ~ 25–50% of the GPU memory and nearly an order of magnitude fewer floating-point operations (FLOPs), while achieving modest but consistent gains in segmentation accuracy (approximately 2%). Additional experiments on independent rainfall- and earthquake-triggered landslide cases further demonstrated its cross-event applicability under fixed post-event deployment conditions. Overall, the results indicate that LDM-Net enables accurate post-event landslide mapping with favourable computational efficiency and scalability, supporting timely inventory generation for emergency response and subsequent hazard analysis.</p>