This presentation explores image-based stage estimation as a scalable alternative to traditional sensor networks, evaluating multiple approaches across a network of USGS monitoring sites with fixed-mount cameras. First, commercially available optical gaging software, Tenevia and Noema, are assessed for operational viability and accuracy against sensor-derived stage data. Second, a deep learning regression model (EfficientNet-B0) is trained to predict gage height directly from full scene imagery. Third, a computer vision segmentation pipeline targeting USGS staff plates extracts the air-water interface to derive stage through physical image measurements.
