Fast data assimilation using a nonlinear Kalman filter and a model surrogate: An application to the Columbia River estuary

A fast and adjoint-free nonlinear data assimilation (DA) system was developed to simulate 3D baroclinic circulation in estuaries, leveraging two recently developed technologies: (1) a nonlinear model surrogate that executes forward simulation three orders of magnitude faster than a forward numerical circulation code and (2) a nonlinear extension to the reduced-dimension Kalman filter that estimates the state of the model surrogate. The noise sources in the Kalman filter were calibrated using empirical cross-validation and accounted for errors in model and model forcing. The DA system was applied to assimilate in situ measurements of water levels, salinities, and temperatures in simulations of the Columbia River estuary. To validate the DA results, we used a combination of cross-validation studies, process-oriented studies, and tests of statistical and dynamical consistency. The validation studies showed that DA improved the representation of several important processes in the estuary, including nonlinear tidal propagation, salinity intrusion, estuarine residual circulation, heat balance, and response of the estuary to coastal winds.