Title: Ensemble based observations sensitivity applied to storm surge forecasting

Authors: Martin Verlaan (Deltares and TU Delft, The Netherlands)
Julius Sumihar (Deltares)

Adjoint based observation sensitivity is rapidly becoming a standard tool for monitoring data-assimilation in an operational system. For ensemble based data-assimilation the tools are still in an early stage of development. In this work we apply ensemble based observation sensitivity to storm surge forecasting in the North Sea. The aim is to optimize the monitoring network and to monitor the data-assimilation process of the operational system.

In order to apply the approach of Liu and Kalnay, several modifications were needed. First, tide gage observations are frequent in time, which makes the sensitivity to the background dominant if only one observation time is considered. Asynchronous filtering (Sakov2010) of multiple times in one batch was used to avoid this problem. In addition, the forecast accuracy was measured as the misfit between observations and the matching forecasts instead of the more common approach that uses the verifying analysis as a proxy to the truth. The background for this is that in our application the analysis increments may not be beneficial in data sparse areas far from the area of interest. It also avoids the introduction of a norm for the misfit between forecast and analysis, which is not an easy task for some applications. This is also practical since the routines for computing a costfunction based on the observation to model misfits are usually available from the data-assimilation.

The application shows that observations at locations in or near the area of interest provide large improvements for small lead times. Locations further from the area of interest have a smaller impact,but provide longer lead time. There is also a dominant direction to the flow of information caused by the counter clockwise propagation of Kelvin waves for the Northern Hemisphere. Finally, observations from some locations do not improve the forecast. This may be caused by local lack of resolution near the observation location, e.g. when it is situated in an estuary.

GMAO Head: Michele Rienecker Global Modeling and Assimilation Office NASA Goddard Space Flight Center

Curator: Nikki Privé Last Updated: May 27 2011