Satellite images are crucial for applications like environmental monitoring, disaster management, and urban planning [1, 2]. It is still difficult for non-experts to interpret such complex images. To solve this, we propose a Visual Question Answering (VQA) system specifically for satellite images, allowing users to pose natural language queries and get image-based responses. Our method combines CNNs for visual feature extraction, BERT-based models for understanding questions, and a multimodal fusion mechanism. Trained on RSVQA and Sentinel-2 datasets, the system attained more than 92% accuracy on tasks such as flood identification, land cover classification, and urban analysis. Through this research, access to geospatial intelligence for planning, monitoring, and emergency response is improved.