BlessemFlood21: Advancing Flood Analysis with a High-Resolution Georeferenced Dataset for Humanitarian Aid Support

TL;DR

The paper presents BlessemFlood21, a high-resolution, georeferenced RGB-NIR dataset of non-coastal river floods captured after the 2021 Erftstadt-Blessem event, with a human-in-the-loop annotation workflow that leverages NIR information for water masks. It establishes a robust baseline for semantic water segmentation using three state-of-the-art models and 4623 512×512 tiles, split into 80/10/10 across train/val/test, demonstrating high accuracy (IoU up to 90.6% and Dice up to 95.1%) and highlighting domain-specific advantages over existing datasets like FloodNet. The study also shows the critical value of high spatial resolution for precise boundary delineation and the limitations of cross-domain transfer, emphasizing the dataset's practical relevance for humanitarian response. Overall, BlessemFlood21 fills a gap in high-resolution, non-coastal flood imagery and provides a foundation for developing fast, reliable water-detection tools for disaster relief operations.

Abstract

Floods are an increasingly common global threat, causing emergencies and severe damage to infrastructure. During crises, organisations such as the World Food Programme use remotely sensed imagery, typically obtained through drones, for rapid situational analysis to plan life-saving actions. Computer Vision tools are needed to support task force experts on-site in the evaluation of the imagery to improve their efficiency and to allocate resources strategically. We introduce the BlessemFlood21 dataset to stimulate research on efficient flood detection tools. The imagery was acquired during the 2021 Erftstadt-Blessem flooding event and consists of high-resolution and georeferenced RGB-NIR images. In the resulting RGB dataset, the images are supplemented with detailed water masks, obtained via a semi-supervised human-in-the-loop technique, where in particular the NIR information is leveraged to classify pixels as either water or non-water. We evaluate our dataset by training and testing established Deep Learning models for semantic segmentation. With BlessemFlood21 we provide labeled high-resolution RGB data and a baseline for further development of algorithmic solutions tailored to flood detection in RGB imagery.

Figures (3)

• Figure 1: Human-in-the-loop framework used for generation of ground truth masks.
• Figure 2: Three representative sample pairs of preprocessed imagery (left) and generated ground truth water masks (right). White pixels in the masks indicate water pixels.
• Figure 3: Qualitative comparison of ground truth and predictions of water masks obtained after fine-tuning with 100 epochs.