Assessment of an ensemble-based data assimilation system for a shallow estuary

Abstract

Data assimilation (DA) is an essential element for the next generation of operational forecast systems for estuaries, to improve estuarine management. With limited resources and prohibitive cost to collect observations for such system, sensor choice and location is of prime importance in improving hydrodynamic model performance. In this study, we examine an optimal ensemble-based DA platform for improving the hydrodynamic modelling of a shallow estuary. Using an ensemble Kalman filter (EnKF), a set of synthetic (twin) experiments was conducted to test different DA scenarios covering observation types (i.e. water level and velocity) and noise modelling. We also evaluated the impact of the observation location on the DA performance by performing an observing system simulation experiment (OSSE). Results revealed that the assimilation of a single variable can significantly enhance the accuracy of the variable being assimilated, while the level of improvement for another variable is smaller. However, the best model estimates were obtained via a multivariate EnKF (i.e. both observations are assimilated). EnKF was robust to under and overestimation of the model errors, although overestimation led to slightly greater improvements. Our analysis showed that model performance is more sensitive to velocity observation location, rather than water level. These findings suggest that locations with strong velocity gradients are the locations where the hydrodynamic model needs to be enhanced, and accordingly, they are the preferable locations to deploy a velocity sensor.