RiverSnap: A citizen science project to monitor and Analyse riverine hydrological parameters from close-range remote sensing images

RiverSnap is a citizen science project as part of the joint project “Zukunftslabor Wasser” that transforms smartphones into measuring instruments for monitoring and analyzing river parameters, responsive to water level changes, natural hazards (e.g., floods), and anthropogenic-induced alterations. A robust stainless-steel smartphone frame is strategically located on or near a bridge for convenient public access to capture river images. This frame facilitates precise image positioning, enabling the capture of river scene images of a predefined and referenced river area that can be uploaded to a centralized database, shared on social media, or sent via email. This collaborative endeavor establishes a community-driven repository documenting river changes over time. Due to water's dynamic nature and structural and sky reflections in close-range images, the RiverSnap project utilizes and develops novel Artificial intelligence (AI) algorithms to extract and predict hydrologic parameters and features.

These advanced algorithms are crucial in detecting water lines, determining positions, and mapping various riverine features with scientific precision. Through this sophisticated technology, RiverSnap transforms community snapshots and additional measurements into a valuable resource for scientifically assessing and understanding alterations in the river environment. As the AI models are data-hungry, RiverSnap is diligently creating benchmark datasets for river water, facilitating the development and training of robust machine learning algorithms. These datasets serve as comprehensive references, allowing the AI models to enhance their understanding of various hydrological patterns, ultimately improving the accuracy and effectiveness of river parameter predictions and feature extractions.

Established in 2023 in Hannover, Germany, the RiverSnap station network has observed significant growth, now covering four monitoring locations around Hannover. Recognizing the pivotal role of detecting the water surface area in approximating riverine parameters, we have developed and implemented various advanced Deep Learning (DL) models for water body segmentation. As part of this initiative, a novel river water dataset named RiverSnap.v1, including 1092 images, has been introduced and is constantly updated. Additionally, various methods have been investigated to geo-reference the analyzed results. In a straightforward approach, artificial or natural markers, such as specific locations of objects around the river or on bridges, were measured with geomatics tools like GNSS receivers and total stations. The DL-extracted water surface was then georeferenced based on these markers to obtain results like the water level. A 3D terrain model derived from LiDAR data or photogrammetric techniques like Structure from Motion (SfM) can be utilized for Geo-referencing parameters and results in more advanced scenarios. This allows for automatically assigning absolute coordinates to each image and subsequent camera pose estimation.

Examples of practical applications of RiverSnap include monitoring high-frequency water level and water line changes and morphological changes in rivers, lakes, wetlands, and urban areas. Additionally, RiverSnap is instrumental in monitoring extended flood areas and observing the time sequence of a flooding event, as demonstrated in data of a German flood of 01/2024.

Funding: This study was performed as part of the joint research project „Zukunftslabor Wasser“ funded by the Lower-Saxon Ministry of Research and Culture (FKZ: 11-76251-1873/2022 (ZN3994))