Flood detection is critical for effective disaster management, allowing for early warning systems and focused relief activities. Traditional flood detection approaches rely on manual interpretation of satellite data and ground-based sensors, are time-consuming, and are subject to errors. Neural network models have emerged as powerful tools for automated image analysis and recognition applications, including flood detection, particularly with the advent of deep learning techniques. This paper offers an automated flood identification strategy based on the Visual Geometry Group 16 (VGG16) model. This deep learning architecture improves the speed and accuracy of flood detection in satellite images. Using transfer learning approaches, the model is optimized to detect slight variations between flooded and non-flooded areas. The proposed approach combines dropout and global average pooling layers to improve generalization and performance. Experimental results show that the VGG16-based model obtains a detection accuracy of 97%, beating traditional Convolutional Neural Network (CNN) models that achieve 91% accuracy. This strategy can potentially improve disaster response and reduce mortality and infrastructure damage by detecting floods more precisely and quickly.