HydroVision is a deep learning-based scene classification framework that estimates optically active water quality parameters, such as chlorophyll-alpha, chlorophyll, pigmented dissolved organic matter (CDOM), phycocyanin, suspended sediment, and turbidity, using RGB images of surface water. Trained using over 500,000 seasonal images collected from 2022 to 2024 from the U.S. Geological Survey's Hydrological Image Visualization and Information System, HydroVision demonstrates its potential for real-world water quality monitoring under diverse conditions. It leverages RGB images as a scalable and cost-effective alternative to traditional multispectral and hyperspectral remote sensing. In an evaluation of VGG-16, ResNet50, MobileNetV2, DenseNet121, and the Vision Transformer network, DenseNet121 achieved the best performance with an R2 score of 0.89 for CDOM prediction. While the current model is optimized for well-lit images, future work is planned to improve its robustness in low-light and obstructed situations to expand its operational utility.
