Performing data assimilation experiments with hydrodynamic models

Abstract

Floods are a big problem for Jakarta; the capital of Indonesia is subsiding below mean sea level and floods will become more frequent. Flood protection and measures against subsidence are, therefore, of high importance to Jakarta’s government. One of the options they are studying is a great sea wall to regulate the water levels in Jakarta Bay. These projects have to be tested and should be build to withhold extreme water levels. In order to simulate these water levels one can use a hydrodynamic model or look at observations. In theory these could be combined by using data assimilation which should improve the hydrodynamic model in such a way that it predicts the results for both observed as unobserved locations quite good. One of the most used data assimilation methods is the Ensemble Kalman Filter, which will be applied to an hydrodynamic model on the Java Sea. The model is simulated using Delft3D-FM and performs quite well as its results are in good correspondence with the observations. On this model data assimilation will than be applied, by making use of OpenDA, in order to further improve the model. With this set-up various twin-experiments will be performed to test the data assimilation method; some experiments with real observations will also be performed to test if it the set-up applicable in real projects. The experiments with real observations perform on the same level as the twin-experiments; the observations thus seem to be suitable for using in real projects. Unfortunately, the twin-experiment does not perform well; the assimilation worsens the results of the hydrodynamic model in both observed, as unobserved locations. In the assimilation process only one component, the localization, seems to performing quite well. This performance can not be studied fully, because the twin-experiments do not give correct results. It will be shown that future work is needed to research the various problems encountered, and that at the time of writing the assimilation with OpenDA does not improve the Delft3D-FM model on the Java Sea.