RivAIrSet: A multitemporal high-resolution UAV imagery dataset for machine learning-based river water segmentation

Understanding and predicting river dynamics requires frequent and systematic observations from imaging sensors deployed in situ or on different platforms. In this context, machine learning methods are increasingly applied to pixel-based classification and image segmentation, enabling innovative approaches for spatiotemporal mapping of dynamic fluvial environments. These methods often rely on large, multitemporal, and well-annotated datasets to ensure robust training and validation. We present RivAIrSet, a dataset of 7630 high-resolution RGB images collected by Unmanned Aerial Vehicle (UAV) along a reach of the Basento River (Southern Italy), with corresponding annotations of river water areas. The images were acquired during multitemporal surveys under varying hydrological and meteorological conditions, capturing different flow regimes. Overall, the dataset provides a valuable resource for fluvial research, supporting advances in machine learning-based river water segmentation, enabling the calibration and validation of hydrological and hydraulic models, and fostering the development of intelligent systems for monitoring fluvial environments. RivAIrSet is available in the UAVRiverMonitoring community, an open repository we created to promote data sharing, enable comparative studies, and drive new research on UAV-based river monitoring.