1. Results
    5.1   Introduction
    NO2 diffusion tube measurements have been compared with simultaneous measurements taken with the automatic chemiluminescent analysers, co-located at each AUN site. These instruments record 15 minute average concentration data which have been averaged for comparison with 2 and 4-week diffusion tube measurement data. For the purposes of the analyses in this report, it has been assumed that no error exists within the automatic chemiluminescent NO2 concentration data, and that all variation is due to uncertainty within the diffusion tube measurements. It should be noted, however, that in practise the accuracy associated with NO2 measurements taken using the automatic chemiluminescent equipment installed at AUN sites is approximately     ±10-11%6, and the precision of measurements is estimated to be 3.5 ppb. This gives an overall uncertainty of about 24% (equivalent to ±6 ppb) at a concentration of 25 ppb.

    5.2   Data Pre-Processing
    Prior to comparison of diffusion tube measurements with automatic data, outlying and anomalous data were identified and removed from the diffusion tube datasets. This process was performed systematically, by analysing the distribution of differences between diffusion tube and corresponding automatic measurement data.

    The deviation between diffusion tube measurements and the automatic measurements were calculated for both individual tubes, and for the average results at each site (by exposure type). Where an individual diffusion tube difference from the corresponding automatic measurement exceeded the average difference by more than 2 standard deviations, these results were considered as outliers and removed. On average, this procedure resulted in approximately 5 diffusion tube measurements being subsequently removed from the each of the diffusion tube exposure type datasets.

    Scatter plots showing the relationship between the bulked data from all 16 sites, grouped by diffusion tube exposure type, are presented in Figures 1 to 6 and in Figure 1a to 6a. Similar plots showing the relationships between measured data at individual sites are shown in Appendix 1, although no formal analyses of these data has been attempted, due to the relatively small number of data.

    5.3   Comparison of Diffusion Tube vs Chemiluminescent Measurements
    Figures 1-6 and Figures 1a to 6a show scatter plots of diffusion tube measurement data against automatic chemiluminescent analyser measurement data, grouped by exposure type. Outlying data have been removed (as discussed above), and linear least squares regression lines demonstrating the best fit relationship applied. In all cases a forced (through zero) regression model was used. The justification for this approach is discussed below.

    Owing to the location of sampling sites in major urban areas in this study, measurement data below 10 ppb are not well represented. The application of unforced regression models to these data produced significant intercepts. However, previous studies7,8 have shown that the relationship between diffusion tube and automatic measurements at low concentrations (<10 ppb) does not produce a significant intercept. This observation is confirmed by an analysis of the 4-week diffusion tube data co-exposed with a chemiluminescent analyser, at the DETR's remote rural monitoring station at Strath Vaich dam, Scotland, between 1991-1997. The relationship between these data is shown in Figure 7, and no significant intercept is observed when an unforced regression model is applied. Therefore, in this study, forced regression models have been used to represent the correlation between diffusion tubes and chemiluminescent monitors over the dynamic range of concentrations measured, and also at concentrations < 10 ppb.

    Regression statistics for the regression lines shown in Figures 1-6 are presented in Table 1. Overall correlations, indicated by the correlation coefficient (r) were excellent, ranging from 0.97-0.99. Highly significant correlations were found between diffusion tube measurements and automatic chemiluminescent measurements for all diffusion tube exposure types at the 99.9% confidence level.

    Table 1:        Least Squares Regression Statistics
    Exposure Type Regression Equation
    uncertainty @ P=0.05 in parentheses    		 Correlation Coefficient (r)
    2 week normal y = 1.09 (+0.020)x 0.99
    4 week normal y = 1.08 (+0.062)x 0.97
    2 week blacked out y = 1.06 (+0.021)x 0.99
    4 week blacked out y = 1.02 (+0.034)x 0.98
    2 week sheltered y = 0.91 (+0.016)x 0.99
    4 week sheltered y = 0.90 (+0.020)x 0.97
    where y = diffusion tube result and x = chemiluminescent monitor result

    Best fit relationships between measured data are described by the equation for the least squares regression line given in Table 1. Uncertainties associated with the gradient, at the 95% confidence level, are given in parentheses. An estimate of the magnitude of systematic difference between diffusion tube and chemiluminescent measurements is provided by the gradient of the least squares regression line. All diffusion tube measurements are shown to be within 10% of the chemiluminescent value, with unsheltered tubes being slightly higher than the chemiluminescent monitor and sheltered tubes lower. Further investigation of the uncertainties associated with gradients at the 95% confidence level, however, showed that the gradients of all regression equations, bar that of the 4-week blacked out exposure, were significantly different from 1.0.

    5.4   Comparison of Sheltered and Unsheltered Diffusion Tube Exposures

    Sheltered and unsheltered diffusion tubes were routinely exposed throughout the study to assess the interfering effect of wind on diffusion tubes. Comparison of the regression equations for each diffusion tube exposure type shows differences in the performance of the exposure types. From this data it can be seen that unsheltered tubes, (normal and blacked out tubes), are shown to overestimate by 8-9% and 2-6% respectively compared to automatic measurements, whilst sheltered tubes underestimate by 9-10%. Observed differences in the performance of sheltered and unsheltered tubes may, tentatively, be linked to the interference of wind on unsheltered tubes resulting in increased uptake rates for these diffusion tube exposures3.

    Study undertaken jointly by Stanger Science and Environment and the National Environmental Technology Centre.
    Site prepared by the National Environmental Technology Centre, part of AEA Technology, on behalf of the UK Department of the Environment, Transport and the Regions