The emergence of increasingly powerful computers enables the development and use of CFD even for complex urban type applications, whereas the focus of flow and dispersion modeling is shifting from mean flow to transient flow and dispersion phenomena predictions such as peak wind loads and probability maps of pollutant concentrations. The use of such models and the possible differences of model results for one and the same problem lead to an increasing awareness of the need for rigorous evaluation. LES based models provide a far better physical potential for predicting dispersion from accidental releases in urban areas (puff dispersion) or simulating the gustiness of winds in complex urban geometries but, these models need to be validated considering their intended use.

However, a number of difficult issues are associated with effective model evaluation. There is no rigorous and commonly accepted standard for specific types of micro-scale flow and dispersion models to be evaluated. Another problem is a general lack of reference data qualified for model validation. The field data from flow and dispersion measurements in complex urban structures cannot provide the repeatability or the quantity of measurements required for a statistically verifiable model evaluation procedure. Providing validation data would require several hundred releases to be simulated at full scale for fixed boundary conditions in order to increase the probabilistic representativeness of the ensemble of results to a level acceptable for model validation purposes. To properly characterize the inflow conditions of a model domain and to provide the required reference data for LES-based model validation, representative mean values are no longer sufficient.

The temporal and spatial distribution of flow and dispersion patterns needs to be characterized and provided for comparison. At this point only the combination of field and laboratory data and complementary experimental work can increase the quality of data substantially. Using contemporary instrumentation and approaches for modeling atmospheric boundary layer flows in a capable wind-tunnel laboratory data can at least partially bridge the existing gap between numerical model results and full-scale measurements.

To illustrate what type, extent and quality of validation data that can be achieved from systematic state-of-the-art wind tunnel modeling; results from a joint research project, sponsored by DTRA, between the Naval Research Laboratory (NRL) and the Environmental Wind Tunnel Laboratory at the University of Hamburg (EWTL) are considered. The main focus of the wind tunnel tests was to provide statistically reliable test data qualified for validating puff dispersion modeling with the LES-based models, for example, the FAST3D-CT contaminant transport and dispersion model developed at NRL. This contribution will introduce the extensive model- and application-specific validation database generated for LES-based puff dispersion modeling in a complex urban structure. Furthermore, the data collected in the wind tunnel is needed to mediate between the complex full-scale conditions measured during Joint Urban 2003 (JU2003) and the complex model results produced by FAST3D-CT, facilitating a sound evaluation of the simulation results and a proper interpretation of the existing field data.

A comprehensive campaign, utilizing the full Oklahoma City geometry, was undertaken at EWTL to develop and build application specific data sets. Careful attention was given to frequency/probability distributions of the flow to give reliable results that could be used for model and field experiment evaluations. Specific issues of model validation (for example due to geometric differences between field, wind tunnel model and numerical model) and the quality of the data set compiled from systematic wind tunnel testing will be presented and documented by a comparison with corresponding results from the JU2003 field trials and the FAST3D-CT simulations.

As an example of the use of field data for model evaluation purposes, the puff data from the Intense Operating Period Eight (IOP8) from JU2003 was selected for comparison with the numerical simulations. Sixty computationally independent passive tracers, enough to draw a statistically significant conclusion, were instantaneously released separated by ten minutes in time (enough to insure de-correlation) to measure the variability of the concentration at the sampler locations. We will present statistical analysis of the computational results and discuss the quality of the field data at the measurement locations in IOP8 with regard to the computational data. Though, the results show reasonable agreement, the small number of field puff releases (only four) limits the degree of certainty we can assign to this comparison. This is further emphasized by the following result.

The comparison of computational results from FAST3D-CT to wind tunnel data sets described above was undertaken. Using the Oklahoma City geometry, results from fifty puff releases from numerical simulations were compared with data taken from up to 200 controlled puff releases in the wind tunnel. The results of this comparison show that, with the careful application of proper statistical measures, the certainty of the conclusions that can be drawn are greatly improved. Based on these comparisons, an example for an application specific evaluation procedure for instantaneous puff dispersion modeling will be given.

A more sophisticated validation approach should be based on statistically representative ensembles of results. By comparing frequency/probability distributions of flow and dispersion results with qualified reference data, the reliability of complex model results can be evaluated. The mediating role of wind tunnel data with respect to a proper interpretation of field results and with respect to a comparison of LES results with corresponding field data is important. Based on the experiences gained from this joint experimental and numerical project, possible strategies for the development further model- and application specific test data sets are expected to play an important role in the model evaluation and validations studies.