Concentration measurements from the Joint Urban 2003 (JU2003) full-scale (FS) and wind-tunnel (WT) experiments are analyzed and compared. At first, concentration statistics are computed for the six FS 30-min continuous releases conducted during the two intensive observation periods IOP 3 and 6. The chosen concentration time series were measured with fast-response analyzers at nine different sites. The measurement set-up is shortly described and the meteorological conditions prevailing during the chosen tracer releases are summarized. The variation of mean concentrations, fluctuation intensity, peak-to-mean ratio, 98 percentiles and intermittency are discussed. High intermittency is observed at sites close to the edge of the urban plume. Even though the meteorological conditions during all six releases were quite similar, all concentration parameters generally show great variability. This highlights the difficulty of obtaining representative statistics from FS experiments, were typically only relatively short time series can be recorded and diurnal variations of weather patterns are typical. In addition to the concentration statistics, the measured concentration probability density (PDF) and cumulative probability (CDF) functions are compared with normal, clipped-normal and gamma-distribution PDFs and CDFs. It can be shown that the gamma distribution probabilities agreed best with the observations particularly in the upper tail of the distributions. The gamma distribution fits thus provided also the best estimates for the concentration percentiles. It is further advantageous that the parameters of the gamma distribution can easily be determined using a method of moments and that realistic concentration percentiles are predicted with the gamma distribution even for highly intermittent cases. For such cases on the other hand, the clipped-normal distribution which is often discussed in the literature and applied in air quality models for computing peak concentrations, resulted in unrealistic, much too-low estimates of the concentrations percentiles and can thus not be recommended.

To address the issues of limited statistical representativeness of the FS data, concentration time series measured in the wind tunnel data were also analyzed. The measurements were conducted with a realistic model of the Oklahoma City central business district at the scale of 1:300. Concentration statistics, as well as PDFs and CDFs were computed for time series measured at three different sites for two wind directions. Taking into account the different length scales and wind speeds in the WT and FS experiments a comparative analysis must be based on dimensionless concentrations and a dimensionless time scale. Computing such dimensionless time, both for the FS and WT records, the WT time series cover a 15-20 times longer time span than the FS time series. In addition to analyzing the concentration statistics for the complete WT records, the WT time series are thus also divided into 15 blocks of equal length and the statistical significance of parameters based on relatively short records is assessed by studying the variability of the concentration statistics and probability functions for the different blocks. In particular at sites closer to the plume edge, the results for the individual blocks are very different and data from short records are statistically thus not very representative. Finally, the FS and WT concentration parameters measured at two sites are directly compared. While the location of the sampling sites in the WT closely matched the sites during the FS experiments, the prevailing wind directions during the FS releases were not exactly matched in the WT. The comparison should thus be interpreted in a more qualitative rather than quantitative sense. In the WT, the concentration time series were less intermittent than in full-scale, which is not really surprising given the more controlled and less variable approach flow. Accordingly, the FS mean concentrations had a tendency to be lower than in the WT, while the fluctuation intensity was typically higher for the FS conditions. Overall, the data agreed however fairly well, and the FS 98 percentiles for almost all releases analyzed are within the bandwidth of the percentiles observed in the block analysis of the WT times series.