Influence of environmental humidity on measurements of benzene in ambient air by transportable GC-PID

Calibration of in situ analysers of air pollutants is usually done with dry standards. In this paper, the influence of environmental humidity on benzene measurements by gas chromatography coupled with a photo ionisation detector (GC-PID) 10 is studied. Two identical chromatographs were calibrated for benzene with dry gases and subjected to measure reference standards with humidity (20% and 80% at 20 oC). When measuring a concentration of 0.5 μg/m benzene in air, the levels of humidity tested did not produce any significant interference in measurements taken with any of the analysers. However, when measuring a concentration of 40 μg/m, biases in measurements of 18% and 21% for each analyser, respectively, were obtained when the relative humidity of the sample was 80% at 20 oC. Further tests were carried out to study the nature of this 15 interference. Results show that humidity interference depends on both the amount fractions of water vapour and benzene. If benzene concentrations in an area are close to its annual limit value (5 μg/m), biases of 2.2% can be expected when the absolute humidity is 8.6 g/cm –corresponding to a relative humidity of 50% at 20 oC-. This can be accounted for in the uncertainty budget of measurements with no need for corrections. If benzene concentrations are above the annual limit value, biases become higher. Thus, in these cases, actions should be taken to reduce the humidity interference, as an underestimation 20 of benzene concentrations may cause a mismanagement of air quality in these situations.

Purified compressed ambient air was used as zero gas. Humidity was added to a portion of the zero air by means of an inhouse designed humidifier (Figure 1a). The humidifier consists of a glass sphere with two lateral inlets (1 and 2) for the zero air to enter and exit the humidifier, respectively. Water is pumped through a glass tube (3) inserted in a third inlet located at the bottom of the sphere (4). The water impacts the top of the sphere and falls down creating a wet film on the walls which favours the mass transfer. The water is collected at the bottom of the sphere (5) and taken to a container provided with thermal 5 insulation where it is stored. When the system is working, the water from the container is pumped to the humidifier through a thermostatic bath where its temperature is readjusted. The whole system is leak-tested and relative humidities up to 99% are attainable depending on the temperatures of the zero air and the humidifying water, and the ratio of zero air flowrate through the humidifier to the flowrate of dry zero air. In Figure 1b a schematic of the humidifying system integrated in one of the lines of dry zero air is shown. 10 Sample relative humidity and temperature were measured with a Testo 645 thermo hygrometer (precision ± 1%). From these values, the humidity mixing ratio, W, -the ratio of the actual mass of water vapour present in the sample to the mass of the dry air-was derived. The water vapour flow rate, qH2O (l/min), added to the flow of zero air, qz (l/min), was calculated using Eq.
(1). It is important to know this flow so that the final concentration of benzene in the reference mixture is calculated accounting for it. Thus, any decrease in benzene measurements when measuring wet samples cannot be attributed to the dilution effect of 15 water vapour. 2 = · · 28.8

18
(1) A high concentration mixture of benzene in nitrogen (1000 µg/m 3 nominal concentration) from a gas cylinder was mixed with the humidified zero air to attain the experimental concentrations required. The flow of gas in each branch was controlled and measured with Bronkhorst mass flow controllers (0-0.4 l/min range for the benzene in nitrogen mixture, and 0-12 l/min for 20 the zero air). The whole piping system was set up inside a thermally controlled chamber, and sample inlet pressure was set up to be equal to normal atmospheric one (101.3 ± 0.2 kPa). Control of ambient conditions is pivotal to ensure that changes on measurements are due to the effect of humidity and not to other environmental conditions. Nevertheless, in order to achieve higher absolute humidities the sample temperature was increased in some of the tests to 35 ºC. Therefore, influence of sample temperature was 25 firstly evaluated and results showed that this variable had no influence on measurements.

Calibration
The analysers used in this work have three different calibration options, namely, a linear calibration using a least squares regression; a calibration line forced through the origin; and finally, a non-linear regression. All three calibration options were 30 tested with eight different mixtures of benzene in air with concentrations ranging from 0.0 to 47.2 µg/m 3 (0.0, 0. 65, 2.60, 5.20, 10.4, 15.6, 26.3, 36.7 and 47.2 µg/m 3 ). Thus, three calibration curves were obtained and the squared sum of residuals of the Atmos. Meas. Tech. Discuss., doi:10.5194/amt-2017-95, 2017 Manuscript under review for journal Atmos. Meas. Tech. Discussion started: 23 May 2017 c Author(s) 2017. CC-BY 3.0 License. concentration tested was obtained for each calibration. The lowest sum of squares (1.16) was obtained with the non-linear (quadratic) calibration, followed by the least squares regression (1.66) and the linear regression forced through the origin (1.78). Therefore, the quadratic option was chosen every time the analysers were calibrated. Calibration was performed at 20 ºC and using dry gases.

First set of experiments
As a first approach to the subject, the tests described in Standard EN 14662-3:2005 were carried out after calibrating the analysers according to Section 2.2.1. These tests consist of measuring a reference mixture of 0.5 µg/m 3 nominal concentration benzene in air with a relative humidity of 20% and 80% at 20 ºC and comparing the results.
Standard EN 14662-3:2005 defines the influence of the relative humidity by means of coefficient brh, calculated as: 10 Where ̅ ℎ, and ̅ ℎ, are the average of 6 consecutive readings when measuring the reference gas mixture (0.5 µg/m 3 nominal concentration benzene in air) with an 80% and 20% relative humidity, respectively, at 20 ºC. Standard EN 14662-3:2005 establishes that brh has to be lower than 4%.
The tests were repeated with a reference mixture of 40 µg/m 3 nominal concentration benzene in air with the same relative 15 humidities and temperature. A significant difference in readings was noticed when working with the high concentration reference mixture with both analysers. Further tests with analyser I were then performed to study in depth this phenomenon.

Second set of experiments
An in-depth study of the influence of humidity on measurements was carried out by measuring several reference mixtures of benzene in air (5 µg/m 3 nominal concentration) with different absolute humidity (AH) ranging from 0 and 32 g/m 3 . These tests 20 were repeated with a reference mixture of 40 µg/m 3 nominal concentration of benzene in air.
Humidities in the range 0-17 g/m 3 were obtained at 20 °C and relative humidities ranging from 0 to 99%. Higher absolute humidities were attained increasing the working temperature to 35 °C. Given that the temperature of the test was not constant, a preliminary evaluation of the influence of sample temperature on measurements was performed. For this, two reference gas mixtures (40 µg/m 3 and 5 µg/m 3 nominal concentration) were measured with analyser I at different temperatures. Two 25 temperature cycles were performed. First cycle was performed at 20 ºC, then changed to 5 ºC and back to 20 ºC (temperature control precision ±2 ºC). The second one was performed at 20 ºC, then changed to 35 ºC and back to 20 ºC again. Once the sample temperature was stabilised 4 measurements were taken at each concentration level. Results showed that sample Atmos. Meas. Tech. Discuss., doi:10.5194/amt-2017-95, 2017 Manuscript under review for journal Atmos. Meas. Tech. Discussion started: 23 May 2017 c Author(s) 2017. CC-BY 3.0 License. temperature inside the tested range did not influence benzene measurements. Thus, this parameter can be changed in order to achieve a high absolute humidity in the samples. Table 1 summarises the results obtained when carrying out the tests described in section 2.2.2.1. Whereas humidity does not 5 have a significant influence on readings at 0.5 µg/m 3 level, it does at 40 µg/m 3 (calculated brh coefficients of 18% and 21% for analyser I and II, respectively). This influence has a negative sign, that is, readings are lower than expected when the relative humidity increases for a constant temperature. Moreover, coefficient brh is higher than Standard EN 14662-3:2005 allows when working with the highest amount fraction of benzene in air. In order to study deeper this phenomenon, the tests described in section 2.2.2.2 were carried out and the results are shown in section 3.3. 10

Influence of sample temperature on benzene measurements
As mentioned in Section 2.2.2.2, a preliminary test to evaluate the influence of sample temperature was carried out. The rationale for this was to know if this parameter affects the readings. If it is not the case, temperature can be changed during the tests and, therefore, the maximum absolute humidity tested is not limited by the saturation humidity of the sample at 20 ºC. Table 2 shows the results of the tests when analyser I measured two reference gas mixtures (40 µg/m 3 and 5 µg/m 3 nominal 15 concentration benzene in air) subjected to two temperature cycles (20 ºC / 5 ºC / 20 ºC and 20 ºC / 35 ºC / 20 ºC). As it can be seen, the change in sample temperature did not produce any significant change in readings and, thus, temperature was increased to 35 ºC in some of the tests in order to work with a higher absolute humidity in our reference gas mixtures. The nondependence of measurements on sample temperature can be explained by the fact that the sample is heated in the oven up to 70 ºC, being the initial temperature of the sample irrelevant in the whole process. 20 Table 3 summarises the humidity conditions, the reference concentration of benzene generated (after considering the dilution effect of water vapour), the average reading of analyser I and the calculated relative difference from the reference concentration of each test. These differences were plotted against the absolute humidity of the test, Figure 2.

Influence of humidity on benzene measurements
There is a clear linear relationship between analyser readings and absolute humidity. For a given benzene amount fraction, the 25 higher the absolute humidity in the sample the lower the chromatograph readings. This result was previously obtained by Barksy et al. (Barksy et al., 1985) but using concentrations of volatile organic compounds at ppm levels.
The data in Figure 2 were fitted by linear least-squares regression, which gave the following equations: E= -1.066·AH+4.783 (r 2 =0.91) and E= -1.557·AH-3.341 (r 2 =0.94) for nominal reference benzene concentrations of 5 µg/m 3 and 40 µg/m 3 , respectively. E is the relative difference between the reference concentration generated in the test chamber and the analyser 30 Atmos. Meas. Tech. Discuss., doi:10.5194/amt-2017-95, 2017 Manuscript under review for journal Atmos. Meas. Tech. Discussion started: 23 May 2017 c Author(s) 2017. CC-BY 3.0 License. reading. Differences between the slopes were studied to find out whether they were significantly different; for this, we used a t-value calculated from t =(m1-m2)/SE(m1-m2), where m1 and m2 are the slopes of the two straight lines compared and SE(m1-m2) is the standard error of the difference, calculated as the square root of the quadratic sum of the standard error of each slope.
A t-value of 2.272 was obtained. This value was lower than the critical one for p=0.05 and 14 degrees of freedom (df=(n1-2)+(n2-2)), which meant that the difference in the slopes was significant and could not be attributed to random measurement 5 error. This is interesting as it shows that the variation of readings by effect of ambient humidity is more pronounced at higher ambient ratios of benzene. Moreover, higher concentrations of benzene are more affected by ambient water vapour as for the same absolute humidity, relative differences are higher in the tests at 40 µg/m 3 than at 5 µg/m 3 .
For 35 °C and 80% relative humidity (31 g/m 3 absolute humidity approx.) the bias in readings was 33% and 47% for a reference concentration of benzene in air of 5 µg/m 3 and 40 µg/m 3 , respectively. These conditions, although a bit extreme, can easily 10 occur in many locations (e.g. Mediterranean areas in summer). Less extreme conditions can also have an important bias in readings (for instance, at 20 °C and 50% relative humidity there is a 2.2% bias in the concentration readings at 5 µg/m 3 level and 13% at 40 µg/m 3 ). Considering a location where mean annual benzene concentrations are close to the annual limit value (5 µg/m 3 ), a bias in measurements of approximately 2% can be easily expected due to a water vapour mixing ratio close to 8.6 g/cm 3 . This bias can be acceptable, taking into consideration that benzene data quality objective in current legislation for fixed 15 measurements is 25%. Thus, it should be incorporated to the uncertainty budget of the measurements with no need for further corrections. Moreover, if ambient concentrations are below the annual limit value, the interference of environmental humidity although not negligible will not change the air quality situation of that area. However, if benzene ambient ratios are above, measurements will be systematically underestimated by effect of ambient humidity, precisely in those areas where a stricter control of concentrations is required. It could be the case of a location that apparently meets the air quality limits because 20 concentrations are underestimated but, in reality, its environmental situation is not acceptable. Thus, it is in these cases where humidity interference on measurements should be addressed. Areas with concentrations of benzene above the annual limit value are widely reported in the literature (Anttila et al., 2016;Bruinen De Bruin et al., 2008;Licen et al., 2016;Al Madhoun et al., 2011).
The effect of humidity on PID performance has been proved to be double. Despite the ionising potential of water vapour being 25 higher than the energy of the PID, it can produce a small background signal at high non-condensing relative humidity, overestimating VOC concentrations. The second effect is the quenching of part of the UV light. When the analysers have been calibrated with dry gases and they measure a sample with humidity, the water vapour molecules in the sample absorb part of the UV radiation emitted. For a given concentration of benzene, the higher the absolute humidity in the sample the higher the absorption of UV radiation and the less energy available to ionise the molecules of benzene. This bias depends not only on the 30 water vapour concentration but also on the benzene one, as we have checked in our tests. For a given concentration of humidity, if the concentration of benzene is very low (e.g. 0.5 µg/m 3 ) the residual UV radiation, that is, the radiation not absorbed by the water vapour, is enough to ionise and, therefore, quantify all the molecules of benzene in the sample. This seems to be the case of the tests conducted in section 3.1 at 0.5 µg/m 3 , as no effect was observed when changing the amount fraction of water vapour in the reference mixture. However, as the amount fraction of benzene increases, the residual UV radiation may not be able to ionise all the molecules of benzene, as it is apparently happening with the samples with 5 µg/m 3 and 40 µg/m 3 benzene in air. From these two effects -background signal and radiation quenching-, the latter seems to be the most influencing as there is a decrease in readings with humidity and not an increase as it would be expected from a background signal effect.
A third phenomenon may be occurring as well. The benzene radical cation formed after ionisation of benzene can react with 5 water to give hydroxycyclohexadienyl radical, which in turn can dissociate to benzene and OH radicals (Eberhardt, 1981).
This effect is in line with the quenching effect of water vapour as both of them reduce the amount of ionised benzene reaching the electrodes.
The influence of humidity on many air quality monitoring techniques has always been a major problem. PID detectors are not the only ones affected. FID were proved to be affected as well (LeBouf et al., 2013); however, these tests were performed at 10 ppm levels. Among the reference measurement techniques to measure air pollutants, chemiluminescence with ozone to measure NO and NO2 is also humidity-dependent (Gerboles et al., 2003;Hayden, 2003;Miñarro and Ferradás, 2012;Steinbacher et al., 2007); and also UV photometry to measure ozone (Wilson and Birks, 2006). Recently, Bluhme et al. (Bluhme et al., 2016) have shown that measurements of SH2 by UV fluorescence are also affected. The interference mechanism is different in each technique but the result is always an underestimation of measurements. Some manufacturers have opted 15 for adding filters or driers to their equipment in order to keep humidity in the sample to a minimum. These implementations have been proved to reduce biases in some cases (Bluhme et al., 2016;Steinbacher et al., 2007;Wilson and Birks, 2006). An alternative to scrubbers, which have the drawback of potentially adsorbing the target molecule, is calibration with wet gases.
Ideally, calibration procedures should be done at the same ambient conditions as sampling. Calibration with wet gases may reduce measurement uncertainty due to environmental humidity in many cases. However, a thorough work regarding short and 20 long-term stability of wet calibration gases in gas cylinders should be first tackled by metrology institutes. Using wet calibration gases obtained by dynamic dilution could bridge the gap and help reduce the uncertainty of benzene measurements and other pollutants in ambient air.
The behaviour observed in this work is likely to be shown by GC-PID instruments by other manufacturers, although to a different extent, which means that benzene concentrations -and, presumably, ethylbenzene, toluene and xylenes concentrations 25 as well-may be systematically underestimated. In areas where ambient concentrations of benzene are usually above the annual limit value, the humidity interference on measurements should be urgently addressed. A joint effort from manufacturers, metrology institutes and users is advisable to reduce the bias due to ambient humidity on BTEX measurements obtained by GC-PID -but also on measurements of other atmospheric pollutants-, as relievable data is the starting point for a correct environmental management. 30 Atmos. Meas. Tech. Discuss., doi:10.5194/amt-2017-95, 2017 Manuscript under review for journal Atmos. Meas. Tech. Discussion started: 23 May 2017 c Author(s) 2017. CC-BY 3.0 License.

Conclusions
In this work, the influence of ambient humidity on benzene measurements obtained with an automated in situ GC-PID is studied.
The chromatograph was calibrated with dry gases, which is nowadays a current practice, and, subsequently, different amount fractions of humidity were added to the reference mixture. The absolute humidity tested ranged from 0 to 31 g/cm 3 . The 5 dilution effect of adding water vapour was taken into account in the reference concentration calculation.
When measuring 5 µg/m 3 of benzene in air, biases in readings ranged from 1 to 32% depending on the absolute humidity in the gas mixture. For an absolute humidity close to 8.6 g/cm 3 -corresponding to a relative humidity of 50% at 20 ºC-the bias in measurements is about 2.2%. Tests were repeated with a 40 µg/m 3 benzene in air mixture. In this case, biases of up to 47% were obtained when the absolute humidity in the sample was 30 g/cm 3 . A less extreme absolute humidity in the sample (8 10 g/cm 3 ) produced a bias of approximately 13%. Results show that water vapour interference depends on both the water and benzene amount fractions in the sample.
If the concentrations of benzene in a certain location are far below the annual limit value (5 µg/m 3 ), the bias due to water interference can be acceptable, taking into consideration that benzene data quality objective in current legislation for fixed measurements is 25%. Thus, it should be incorporated to the uncertainty budget of the measurements with no need for further 15 corrections. Moreover, if ambient concentrations are below the annual limit value, the interference of environmental humidity although not negligible will not change the air quality situation of that area. However, if benzene ambient ratios are above, measurements will be systematically underestimated by effect of ambient humidity, precisely in those areas where a stricter control of concentrations is required. Thus, it is in these cases where humidity interference on measurements should be addressed. Using appropriate scrubbers or wet calibration gases could help reduce measurement uncertainty of benzene and 20 many other air pollutants monitored with analytical techniques also affected by water vapour.