The ultimate goal is to demonstrate the degree of correspondence between the results of three-dimensional advection-dispersion models and the in-situ observations of those processes and to delineate the limits of that correspondence. Such a goal can not be reached without a scientifically founded methodology aiming at the evaluation, intercomparison and classification of three-dimensional hydrodynamic and advection-dispersion models. To our knowledge, such a methodology is still lacking. Comparisons between model results and in-situ observations are still more qualitative than quantitative. They clearly do not answer the main questions of the policy makers: "What degree of confidence do you have in your model predictions?", "What are the error bars on your model predictions?".
The objectives are seen as contributing to the description of the variability expected in nature and within (and between) models.
These objectives will:
The statistical tools (1) will provide the measure of variation. The sensitivity studies (2) will provide a measure of expected model variability. Objectives (3) and (4) will provide measures of the actual variability, and (5) will provide a detailed assessment of the variability.
Obviously, there are limitations on what can be achieved in measuring, quantifying and assessing natural and model variability.
Within the proposed study, the limitations applied will be:
Such a choice is fully justified by the following reasons:
1) The model and/or field experiments (both at national and international levels) already performed in this area are numerous (e.g., JONSDAP '73 and '76, the Belgian North Sea project 1971-1976, the UK North Sea Project 1988-1991, FLUXMANCHE, XTRANS, SKAGEX, ...) and it is believed that the information needed for the purposes of the project can be found in the data sets generated by those programmes, complemented by the remote sensing data collected over similar period of times, and existing data archives such as BODC, ICES, NOWESP;
2) The wide variety of regimes (strong tidal and small thermohaline influences, strong tidal and strong thermohaline influences, small tidal and strong thermohaline influences) existing in that area offer a unique opportunity to test the world-wide validity of the methodology to be developed.
Experiences from the MAST-II Concerted Action NOMADS, designed to intercompare the results of existing advection-dispersion models currently used within European seas, showed that it was not possible to clearly identify the causes for deviation between different models. In particular it was not possible to determine whether differences were due to differences in the model setup or due to the employment of different forcing fields.
Moreover, the project demonstrated that field data are required to enable a real intercomparison between models. Without any data the model-model intercomparison of the project was only able to give a qualitative measure of agreement between different models based on different parameters and different model experiments. However, since different models had similar outputs for different experiments, a clustering of the models was very hard to do. True differences between models, differences due to different inputs and errors in implementation could not be separated. In a complex system (without any data on the real world) there is also the problem that the outliers might be the most correct ones. It is believed that all models within the MAST-II project were mature enough to deal with real data.
It therefore seems appropriate that studies should be conducted to identify the kind of variability that it is reasonable to expect from model simulations when compared with the real world and to try and place some expectation on the agreement between models in model-model intercomparison studies. The MAST-II Concerted Action experience shows that this must be done within a relatively simple controlled environment where variation in model results can be identified with specific components of the model (be it inputs, algorithms or numerical implementation) set against a background of quantifiable variability in the real world.
Two components can be identified which will contribute to an insight into the variability seen in model results:
 Next: Management of Task
|
 |
Previous: Abstract
|