Remote Sensing Scene Classification (RSSC) encounters challenges from diverse land-cover types, intra-class variability and inter-class similarity. To tackle these issues, we introduce the Global Local Dynamic Attention Network (GLDAN), a novel framework designed to boost classification accuracy by dynamically integrating global and multi-scale local features. The core component, the Global Local Dynamic Attention Module (GLDAM), consists of the Global Attention Module (GAM) to capture global context, the Multi-scale Local Attention Module (MLAM) to extract fine-grained local details, and the Spatial-Channel Attention Fusion Module (SCAFM), which dynamically fuses global and local features rather than relying on simple concatenation or multiplication. Unlike traditional methods that apply attention across all bottlenecks, GLDAM is integrated solely into the final bottleneck of ResNet50 stage 1, enhancing accuracy while minimizing parameters. Experiments on the RSSCN7 and SIRI-WHU datasets demonstrate that GLDAN achieves state-of-the-art performance, with overall accuracies of 98.75% and 98.54%, respectively.