|Variationele golf-data-assimilatie voor operationele golfvoorspellingen = Variational wave data-assimilation for operational wave forecasting|
Vos, S.E. (2002). Variationele golf-data-assimilatie voor operationele golfvoorspellingen = Variational wave data-assimilation for operational wave forecasting. Communications on Hydraulic and Geotechnical Engineering, 02-1. PhD Thesis. TU Delft. Faculty of Civil Engineering and Geosciences: Delft. ISBN 90-9016228-3. 126 pp.
Part of: Communications on Hydraulic and Geotechnical Engineering. Delft University of Technology. Department of Civil Engineering: Delft. ISSN 0169-6548, more
Delft University of Technology (TUDelft), more
In this thesis the possibilities are investigated to use data assimilation for improving the wave forecast under semi operational conditions. Data assimilation is essentially a technique where observations and model results are combined in a hindcast period to improve the wave prediction. Generally such assimilation is used to improve the wave model or driving wind field during the period prior to the forecast. The present study aims at improvements of the wind field as an analysis indicates that this is most beneficial. The intention was to make a data assimilation scheme which is both effective and efficient such that it can be used under operational conditions. Present operational data assimilation schemes are very efficient as these are very fast, but these are not very effective as the effect of the data assimilation is lost quickly in the forecast period. For the present study a method called variational data assimilation is used which in principle can be more effective than the present operational schemes.
An important aspect of the variational method is that the efficiency (i.e. the speed of the method) is partly dependent on the number of control variables, which are the parameters that are varied in the hindcast period. The number of control variables influences the number of needed model runs. In order to keep the number of model runs to a minimum only a limited number of control variables is used.
In order to correct all wind vectors with a limited number of control variables several methods have been investigated. The first two methods use the high correlation of the individual wind vectors in space and time by clustering wind vectors in coherent areas and by applying a limited number of common corrections to the wind vectors in each area. The storm areas technique clusters wind vectors around high and low pressure systems while the zonal technique clusters wind vectors between two parallels of latitude.
The Empirical Orthogonal Function method focuses on splitting the wind field in smallscale and large-scale wind structures. This allows corrections of the small-scale and large-scale wind structures with a limited number of control variables.
The fourth method tries to reduce biases in the wind speed observed by Tolman (1998) in NCEP wind fields, using a simple parameterization with only two control variables. The fifth and last method tries to incorporate the effect of wind gustiness. Wind gustiness (all turbulent motions of the wind with a period less than 30 minutes) has a significant influence on the wave growth but values of wind gustiness are not provided by numerical weather models. The effect of the wind gustiness is therefore assimilated.
The different techniques have been tested in two test series. The first tests consisted of a series of academic tests on the Indian Ocean performed with the wave model DOLPHIN. This is a second generation wave model which was selected for its speed. The test results are positive, indicating that the different methods are capable of applying corrections in the wind field with the help of the data assimilation.
A second series of tests was performed on the Indian Ocean with the third generation wave model (AD)WAM to test the different techniques with real data. The results are not positive as most wave forecasts are not improved at all, but are even deteriorated. Also the average run time of the data assimilation runs is long which limits its use for operational purposes.
The problem is that the present data assimilation method focuses on large scale corrections in the wind field. However, the large-scale wind structures are forecasted reasonably well while small-scale wind structures with a spatial scale up to 500 km are predicted less well. Yet these are important for the details in the wave model results. In order to correct the smallscale wind structures more observations are needed which are presently not available. Next the present wave models are also too diffusive to predict the details correctly. This should also be improved in order for the data assimilation to have any effect.