Abstract
Existing systems using field PIV or high frequency radar to measure surface velocity over a certain area shows apromising possibility to construct real-time monitoring systems of depth-scale flows based on surface velocity. In thisscenario, we propose Proper Orthogonal Decomposition-based Unscented Kalman Filtering, namely POD-UKF, asthe first step toward constructing such systems. Radial Basis Function is used as state-space models of POD-UKF.Measurement models are POD-based regression models using Kernel Ridge Regression. Data used in this work is1600 velocity snapshots of simulated flow over a Backward-Facing Step in an open channel. Estimation results areshown for tracking flow information at some subsurface regions based on "virtual sensors" of surface velocity.
