Beyond the Algorithm Challenge

Flood detection using satellite imagery faces challenges due to the large-scale, heterogeneous nature of the data and the computational intensity required. We propose QUAFFLE (Quantum U-Net Assisted Federated Flood Learning and Estimation), a hybrid framework that integrates quantum-enhanced U-Net architectures with federated learning to enable distributed and accurate flood mapping. QUAFFLE combines SAR and optical satellite data in a federated setting, reducing the need to transmit raw images while maintaining performance. A variational quantum layer at the U-Net bottleneck reduces parameter count and improves feature extraction. We implemented QUAFFLE using PennyLane for gate-based and ORCA-SDK photonic quantum simulations, Flower for federated learning, and PyTorch for model development. Evaluations on the IEEE DFC24 SAR and Optical datasets show improved AUC and accuracy with fewer parameters than classical counterparts. The framework is compatible with real and simulated quantum hardware, offering a resource-efficient approach to enhance flood mapping accuracy under practical constraints.