Enhanced Flood Management Using a Climate Disaster Image Dataset

Floods are among the most destructive climate-related disasters, necessitating the development of effective tools for precise forecasting and prompt action. This study proposes a hybrid flood detection framework that integrates temporal rainfall trend analysis with spatial image classification. The system uses a specially created dataset that includes 650 annotated images with flood and non-flood labels, along with the associated meteorological variables, temperature, humidity, precipitation, and symbolic weather conditions. When used for image classification, MobileNetV2, which was chosen for its effectiveness in resource constrained environments, achieved a 94.36% detection accuracy and a 32% decrease in misclassification compared to conventional models. An 80:20 train-test split with cross validation was used to train and assess the model. The time-series component of the system looks for patterns in seasonal flood risk by analyzing historical rainfall data. The integration of time-series and image-based analyses into a single predictive platform, which permits spatiotemporal flood detection, is one of the main contributions of this study. To aid decision making, a visualization dashboard shows rainfall trends. These findings imply that the system can assist with disaster preparedness and response planning and is appropriate for real-time deployment in flood-prone areas. To improve the predictive power of the system, future research should focus on expanding the dataset and incorporating sophisticated forecasting models.