Widely adopted models for estimating channel geometry attributes rely on simplistic power-law (hydraulic geometry) equations. This study presents a new generation of channel geometry models based on a hybrid approach combining traditional statistical methods (Multi-Linear Regression (MLR)) and advanced tree-based Machine Learning (ML) algorithms (Random Forest Regression (RFR) and eXtreme Gradient Boosting Regression (XGBR)), utilizing novel datasets. To achieve this, a new preprocessing method was applied to refine an extensive observational dataset, namely the HYDRoacoustic dataset supporting Surface Water Oceanographic Topography (HYDRoSWOT). This process improved data quality and identified observations representing bankfull and mean-flow conditions. A compiled dataset, combining the preprocessed dataset with datasets containing additional catchment attributes like the National Hydrography Dataset Plus (NHDplusv2.1), was then used to train a suite of models to predict channel width and depth under bankfull and mean-flow conditions. The analysis shows that tree-based ML algorithms outperform traditional statistical methods in accuracy and handling the data but face limitations in prediction capabilities for streams with characteristics outside the training range. Consequently, a hybrid method was selected, combining XGBR for streams within the dataset range and MLR for those outside it. Two tiers of models were developed for each attribute using discharges derived from distinct sources (HYDRoSWOT and NHDPlusV2.1, respectively), where the second tier of models offers applicability across approximately 2.6 million streams within NHDplusv2.1. Comprehensive independent evaluations are conducted to assess the capability of the developed models in providing stream/reach-averaged (rather than at-a-station) predictions for locations outside the training and testing datasets.

Heavy rains and tropical storms often result in floods, which are expected to increase in frequency and intensity. Flood prediction models and inundation mapping tools provide decision-makers and emergency responders with crucial information to better prepare for these events. However, the performance of models relies on the accuracy and timeliness of data received from in-situ gaging stations and remote sensing; each of these data sources has its limitations, especially when it comes to real-time monitoring of floods. This study presents a vision-based framework for measuring water levels and detecting floods using Computer Vision and Deep Learning (DL) techniques. The DL models use time-lapse images captured by surveillance cameras during storm events for the semantic segmentation of water extent in images. Three different DL-based approaches, namely PSPNet, TransUNet, and SegFormer, were applied and evaluated for semantic segmentation. The predicted masks are transformed into water level values by intersecting the extracted water edges, with the 2D representation of a point cloud generated by an Apple iPhone 13 Pro LiDAR sensor. The estimated water levels were compared to reference data collected by an ultrasonic sensor. The results showed that SegFormer outperformed other DL-based approaches by achieving 99.55% and 99.81% for Intersection over Union (IoU) and accuracy, respectively. Moreover, the highest correlations between reference data and the vision-based approach reached above 0.98 for both the coefficient of determination (R2) and Nash-Sutcliffe Efficiency. This study demonstrates the potential of using surveillance cameras and Artificial Intelligence for hydrologic monitoring and their integration with existing surveillance infrastructure.