4.1.  Ozone generation

The same source of purified air is used for all the ozone photometers being compared. This air is used to provide reference air as well as the ozone–air mixture to each ozone photometer. Ambient air is used as the source for reference air. The air is compressed with an oil-free compressor. It is dried and scrubbed using a commercial purification system so that the mole fraction of ozone and nitrogen oxides remaining in the air is below detectable limits. The relative humidity of the reference air is monitored and the mole fraction of water in air is typically less than $3\mathrm{\mu mol}/\mathrm{mol}$. The mole fraction of volatile organic hydrocarbons in the reference air was measured (November 2002), with no mole fraction of any detected component exceeding $1\mathrm{nmol}/\mathrm{mol}$.

A common dual external manifold in Pyrex is used to supply the necessary flows of reference air and ozone–air mixtures to the ozone photometers. The two columns of this manifold are vented to atmospheric pressure.

4.2.  Comparison procedure

All the instruments were switched on and allowed to stabilize for at least 8 hours before the comparison commenced. The pressure and temperature measurement systems of the instruments were checked at this time. If any adjustments were required, these were noted. For this comparison, no adjustments were necessary.

One comparison run normally includes ten different amount-of-substance fractions distributed to cover the range, together with the measurement of reference air at the beginning and end of each run. The nominal amount-of-substance fractions were measured in a sequence imposed by the BIPM.QM-K1 protocol, plus three additional values to extend the range (0, 220, 80, 420, 120, 320, 30, 370, 170, 500, 270, 0, 570, 840, 700 and 0) $\mathrm{nmol}/\mathrm{mol}$. Each of these points is an average of ten single measurements.

For each nominal value of the ozone mole fraction $x_{\text{nom}}$ supplied by the ozone generator, the standard deviation $s_{\text{SRP27}}$ on the set of ten consecutive measurements $x_{\text{SRP27},i}$ recorded by BIPM-SRP27 was calculated. The measurement results were considered as valid if $s_{\text{SRP27}}$ was less than $1\mathrm{nmol}/\mathrm{mol}$, which ensures that the photometers were measuring a stable ozone concentration. If not, another series of ten consecutive measurements was performed.

4.3.  Comparison repeatability

The comparison procedure was repeated continuously for BIPM.QM-K1 to evaluate its repeatability. The participant and the BIPM jointly decided when both instruments were stable enough to start recording a set of measurement results to be considered as the official comparison results. The measurements reported here were recorded in an additional run that was performed immediately afterwards.

4.4.  SRP27 stability check

A second ozone reference standard, BIPM-SRP28, was included in the comparison to verify its agreement with BIPM-SRP27 and thus follow its stability over the period of the on-going key comparison.

5.1.  Measurement equation of a NIST SRP

The measurement of ozone mole fraction by an SRP is based on the absorption of radiation at $253.7\mathrm{nm}$ by ozonized air in the gas cells of the instrument. One unique characteristic of the instrument design is the use of two gas cells to overcome the instability of the light source. The measurement equation is derived from the Beer–Lambert and ideal gas laws. The number concentration ($C$) of ozone is calculated from:

Using the ideal gas law, Equation (1) can be reformulated in order to express the measurement results as a mole fraction ($x$) of ozone in air:

The formulation implemented in the SRP software is:

5.2.  Absorption cross–section for ozone

The linear absorption coefficient at standard conditions ${\alpha }_{\text{x}}$ used within the SRP software algorithm is $308.32{\mathrm{cm}}^{-1}$. This corresponds to a value for the absorption cross–section $\sigma$ of $\mathrm{1.147 6}\times {10}^{-17}{\mathrm{cm}}^2\text{/molecule}$, rather than the more often quoted $1.147\times {10}^{-17}{\mathrm{cm}}^2\text{/molecule}$. In the comparison of two SRP instruments, the absorption cross section can be considered to have a conventional value and its uncertainty can be set to zero. However, in the comparison of different methods or when considering the complete uncertainty budget of the method the uncertainty of the absorption cross–section should be taken into account. A consensus value of 2.12 % at a 95 % level of confidence for the uncertainty of the absorption cross–section has been proposed by the BIPM and the NIST in a recent publication [3].

5.3.  Condition of the BIPM SRPs

Compared to the original design described in [1], SRP27 and SRP28 have been modified to deal with two biases revealed by a study conducted by the BIPM and the NIST [3]. In 2009, an “SRP upgrade kit” was installed in the instruments, as described in [4].

5.4.  Uncertainty budget of the common reference BIPM-SRP27

The uncertainty budget for the ozone mole fraction in dry air ($x$) measured by the instruments BIPM-SRP27 and BIPM-SRP28 in the nominal range $0\mathrm{nmol}/\mathrm{mol}$ to $900\mathrm{nmol}/\mathrm{mol}$ is given in Table 1.

Table 1.  Uncertainty budget for the SRPs maintained by the BIPM

As explained in the comparison protocol, following this budget the standard uncertainty associated with the ozone mole fraction measurement with the BIPM SRPs can be expressed as a numerical equation (numerical values expressed as $\mathrm{nmol}/\mathrm{mol}$):

5.5.  Covariance terms for the common reference BIPM-SRP27

Correlations in-between the results of two measurements, performed at two different ozone amount-of-substance fractions with the BIPM–SRP27, were taken into account in the OzonE software. More details on the covariance expression can be found in the protocol. The following expression was applied:

where:

The value of $u_{\text{b}}$ is given by the expression of the measurement uncertainty: $u_{\text{b}}=2.92\times {10}^{-3}$.

5.6.  Condition of the CHMI SRP17

Compared to the original design, the CHMI SRP17 has been modified to deal with the two biases revealed in [3]. In August 2007, an “SRP upgrade kit” was installed by NIST, as already described in the previous comparison report [5].

5.7.  Uncertainty budget of the CHMI SRP17

The uncertainty budget for the ozone mole fraction in dry air $x$ measured by the CHMI standard SRP17 in the nominal range $0\mathrm{nmol}/\mathrm{mol}$ to $900\mathrm{nmol}/\mathrm{mol}$ is given in Table 2.

Following this budget, as explained in the protocol of the comparison, the standard uncertainty associated with the ozone mole fraction measurement with the CHMI SRP17 can be expressed as a numerical equation (numerical values expressed as $\mathrm{nmol}/\mathrm{mol}$):

No covariance term for the CHMI standard SRP17 was included in the calculations.

Table 2.  SRP17 uncertainty budget

7.1.  Measurement results

The measurement results of the last of the twelve recorded runs are shown in Table 3. For each nominal ozone mole fraction, the standard deviation is reported on the ten successive measurements that were recorded. The values reported here show that both instruments were in a stable regime.

An interesting way to look at these results is to display the difference between the ozone amount-of-substance fractions measured by SRP17 and SRP27 ($x_{17}–x_{27}$) versus the ozone mole fraction measured by SRP27. Figure 1, shows that measurement results from both instruments differ little from each other over the entire concentration range of the comparison. Considering a coverage factor of $k=2$ (95 % confidence interval), the difference between both standards is much lower than their combined measurement uncertainties.

Table 3.  measurement results of the comparison between the CHMI standard SRP17 and the BIPM reference standard SRP27

Figure 1 — difference between the ozone amount-of-substance fractions measured by SRP17 and SRP27 versus the ozone mole fraction measured by SRP27.

The relationship between SRP17 and SRP27 is given by the result of the generalized least-square regression performed following the method described in Chapter 6:

The standard uncertainties on the parameters of the regression are $u\left(a_1\right)=\mathrm{0.003 2}$ for the slope and $u\left(a_0\right)=0.21\mathrm{nmol}/\mathrm{mol}$ for the intercept. The covariance between the two parameters is $\text{cov}\left(a_0,a_1\right)=–1.66\times {10}^{-4}$.

The least-squares regression results confirm that a linear fit is appropriate, with a sum of the the squared deviations (SSD) of 0.30 and a goodness of fit (GoF) equal to 0.17.

To assess the agreement of the standards using Equation (10), the difference between the calculated slope value and unity, and the intercept value and zero, together with their measurement uncertainties, need to be considered. In this comparison, the value of the intercept is consistent with an intercept of zero, considering the uncertainty in the value of this parameter; i.e $\left|a_0\right|<2u\left(a_0\right)$, and the value of the slope is consistent with a slope of 1; i.e. $\left|1–a_1\right|<2u\left(a_1\right)$.

7.2.  Repeatability

The results of the twelve comparison runs performed successively between SRP17 and SRP27 are reported in Table 4. With a standard deviation of $0.04\mathrm{nmol}/\mathrm{mol}$ on the intercept and a relative standard deviation of 0.03 % on the slope, these data show a good repeatability.

Table 4.  results of the ten comparison runs repeated successively

7.3.  History of comparisons between BIPM SRP27, SRP28 and CHMI SRP17

Results of the previous comparison performed with CHMI, on the same measurement range and since the first one in December 2002, are shown in Figure 2 together with the results of this comparison. The slopes $a_1$ of the linear relation $x_{\text{SRP}n}=a_0+a_1{x}_{\text{SRP27}}$ are represented together with their associated uncertainties calculated at the time of each comparison. Results of previous comparisons have been corrected to take into account the changes in the reference BIPM-SRP27 described in [4], which explains the larger uncertainties associated with the corresponding slopes.

Figure 2 shows that all standards included in these comparisons remained in close agreement.

Figure 2 — Results of previous comparisons between SRP27, SRP28 and CHMI-SRP17 realised at the BIPM. Uncertainties are calculated at $k=2$, with the uncertainty budget in use at the time of each comparison.