HydroFuseNet: Event-Generalizable Multimodal Deep Learning for Flood Segmentation with Uncertainty-Aware Mapping

Abstract

This release adds the HydroFuseNet reproducibility implementation accompanying the research data and experiment metadata. It includes scripts for multimodal preprocessing, event-disjoint dataset partitioning, HydroFuseNet model construction, training and checkpoint selection, independent-test evaluation, predictive uncertainty analysis, independently trained modality-ablation experiments, and figure generation. The repository also includes experiment configuration and reconstruction documentation identifying which settings are directly supported by the finalized study materials and which implementation details were reconstructed where the original Colab configuration was unavailable. Associated manuscript: "HydroFuseNet: Event-Generalizable Multimodal Deep Learning for Flood Segmentation with Uncertainty-Aware Mapping."