In this study we compared the measurement results from six different commercially available FRM- and FEM-designated O3 instruments operated in ambient or chamber-generated biomass burning smoke. All instruments were operated according to their FRM or FEM designation. The six instruments differed by measurement principle (chemiluminescence versus UV photometric) and by sample treatment configuration (scrubber material, presence of dryer, etc.). For interference-free O3 measurements, we utilized the newly designated FRM NO-CL method (U.S. EPA, 2015). For the UV photometric methods, we compared both catalytic scrubber and “scrubberless” (gas-phase chemical scrubber) technologies, with the scrubberless monitor using a NO chemical scrubber. Finally, within the catalytic scrubber UV photometric category, we compared instruments with and without Nafion^® tube dryer systems. The operation principle and designations (FRM vs. FEM) for the analyzers under investigation are summarized in Table 1 and described in Sect. 2.1.1–2.1.4. These analyzers were operated immediately downwind of fresh biomass burning plumes over 8 d of prescribed fires in grassland ecosystems in Oregon and Kansas and during laboratory-based studies at the U.S. Forest Service's (USFS) combustion facility at the Fire Sciences Laboratory (FSL) in Missoula, Montana. The grassland fire fuels consisted primarily of mixed native prairie tallgrass of varying moisture content. A total of 7 of the 8 d of prescribed burning were conducted in the tallgrass prairie ecosystem of central Kansas (4 d in March of 2017 and 3 d in November of 2017). The additional burn day was conducted at the Sycan Marsh in central Oregon (October of 2017). Laboratory-based chamber burns at the FSL were conducted during April 2018 and again during April 2019. Fuels for the laboratory based chamber burns consisted of ponderosa pine needles and fine woody debris. Details of the individual studies are provided in Sect. 2.2–2.6.

During the prescribed fire grass burns, all study instrumentation (analyzers, data acquisition systems, and peripheral systems) were mounted in portable instrument racks and installed inside an enclosed EPA 4×4 vehicle (Whitehill et al., 2019). The instruments were connected via perfluoroalkoxy alkane (PFA) Teflon^® tubing (0.64 cm diameter) to PFA Teflon^® filter packs loaded with 47 mm, 5 µm pore size pressure-drop-equivalent Millipore (Burlington, MA, USA) Omnipore^® polytetrafluoroethylene (PTFE) filter membranes which were (i) mounted to a rooftop sampling platform during spring 2017 sampling or (ii) connected to a cross-linked Teflon^®-coated high-flow manifold mounted on the inside roof of the truck compartment during fall 2017 sampling. The truck was positioned downwind of active biomass burning plumes, usually within meters to hundreds of meters of the active fire line, and positioned so that the trailer was downwind of the sample inlets (to avoid interferences from generator exhaust). In addition to the O3 analyzers under investigation, additional monitors were also operated for the determination of carbon monoxide (CO), NO, NO2, total oxides of nitrogen (NOx= NO + NO2), and total hydrocarbons (THCs, to approximate VOC concentrations). The operation principle and designation (FRM vs. FEM) information for the additional analyzers deployed in this study are summarized in Table 2. Data from all instruments were recorded on an Envidas Ultimate data acquisition system.

All instruments were calibrated with multipoint calibrations before and after each sampling day. All pre- and post-calibrations met our quality performance objectives of ±10 % and linearity of r2≥0.99. For the O3 analyzers under investigation, field and laboratory calibrations were performed using a Teledyne API model T700U dynamic dilution calibrator with a NIST traceable O3 photometer and O3 generation system. Zero air for the calibrator was supplied by a Teledyne API model T701H zero-air generator. Calibrations for NO, NO2, NOx, and CO were performed using the same calibrator and zero air generator utilizing a certified EPA protocol tri-blend (CO, NO, SO2) gas cylinder (Airgas). Per the manufacturer-provided operator's manual, calibrations for THC were performed using the T700U calibrator and a certified EPA methane–propane gas cylinder (Airgas). FID response factors for organic compounds can vary significantly based upon factors such as carbon number and compound class (Tong and Karasek, 1984). The carbon numbers for methane and propane vary by a factor of 3 and the FID response factors for those compounds may also vary by a similar amount. In addition, the complex mixture of hydrocarbons found in smoke will have large variations in carbon number and FID response factors. As such, the results obtained with the THC analyzer are an approximation of THC (and VOC) concentrations in smoke. In addition, for THC calibrations, the T701H zero-air generator was replaced with scientific-grade zero-air compressed gas cylinders (Airgas).

Biomass burning plumes were sampled over 4 d of prescribed burns (15–17 and 20 March 2017) on the Konza Prairie Long Term Ecological Research (LTER) site outside of Manhattan, Kansas. The fuels for this series of burns consisted of mixed native prairie tallgrass of varying moisture content. Over the 4 d period, a total of 13 burns were conducted and sampled.

A single 10 h day of prescribed grassland burning was measured at the Sycan Marsh Preserve in central Oregon on 11 October 2017. Fuels for the Sycan Marsh burn consisted of mixed native prairie tallgrass of varying moisture content.

Biomass burning plumes were sampled during a single day of prescribed burning (10 November 2017) on the Konza Prairie LTER site outside of Manhattan, Kansas, and on 2 additional days (13 and 15 November 2017) at the Tallgrass Prairie National Preserve outside Strong City, Kansas. Fuels for the November 2017 burns consisted of mixed native prairie tallgrass of varying moisture content. During the 10 November sampling, two separate burns were conducted. Three burns were conducted over the 2 d at Tallgrass Prairie National Preserve.

Laboratory-based studies were performed at the US Forest Service's combustion testing facility at the FSL in Missoula, Montana, by EPA and USFS personnel. These static chamber burns occurred in the spring of 2018 (16–24 April 2018, 33 burns; Landis et al., 2021) and again in the spring of 2019 (15–26 April 2019, 31 burns). The main combustion chamber is a square room with internal dimensions of 12.4 m × 12.4 m × 19.6 m high and a total volume of 3000 m3 and has been described previously (Bertschi et al., 2003; Christian et al., 2004; Yokelson et al., 1996; Landis et al., 2021). During the combustion chamber studies, the facility was fitted with identical instrumentation racks, calibration systems, systems for sampling of test atmosphere, and data acquisition systems as those described in Sect. 2.2. All instrumentation was housed in an observation room immediately adjacent to the combustion chamber with PFA inlet lines extending through the wall into the chamber. All inlet lines contained an identical filter pack and filter assembly described in Sect. 2.2 to protect inlet lines and the analyzers from particulate contamination. Fuel beds consisting of ponderosa pine needles and mixed woody debris were prepared and placed in the middle of the chamber. The amount and moisture content of the fuels were varied to generate different flaming/smoldering conditions during the burns. During the chamber burns the combustion room was sealed and the fuel bed was ignited. Two large circulation fans on the chamber walls and one on the ceiling facilitated mixing and assured homogeneous conditions during the burn periods (Landis et al., 2021). In general, chamber relative humidity (RH) values were below 50 %, facilitating dry burning conditions.

O3 measurement results from the Oregon and Kansas prescribed grassland fire studies are shown as the difference between the FEM and FRM in Fig. 1, and 1 min average time series plots for the studies are presented in Supplement Figs. S1–S3. There were significant differences in the measurement results obtained from the different O3 monitors operated during the prescribed fires. The UV-C instrument (Thermo 49i) consistently showed large increases in O3 concentration readings in fresh biomass burning plumes, with measurements exceeding the FRM measurement by 2–3 ppm. The O3 exceedances were generally correlated in time with CO and THC (biomass burning indicators) and NO2. These correlations will be discussed separately. The UV-C-H instrument (2B 205) also showed increased readings in smoke plumes (also correlated with CO, THC, and NO2), but with absolute measurements roughly an order of magnitude smaller than the UV-C instruments. The NO-CL (T265) instrument results showed the opposite behavior, with reductions in O3 readings inversely correlated with increases in NO2 concentrations, as expected from general O3 titration by NO (NO + O3→ NO2+ O2). For the March 2017 measurements the SL-UV instrument (2B 211) produced readings roughly comparable with the NO-CL monitor, but with substantially more noise on a minute-to-minute timescale. The “in-plume” average O3 concentrations from the four prescribed grassland burning periods are shown in Fig. 2. For the purposes of this comparison, CO measurements were used as an indicator of when sampling occurred in plume. In addition, ambient RH values were generally below 50 % indicating that the spring and fall 2017 prescribed burns were conducted under dry conditions.

O3 measurement results from the 2018 and 2019 USFS chamber burn studies are shown in Fig. 3. Time series plots of the chamber study data are included in Supplement Figs. S4 and S5. Figure 4 gives a more detailed view of UV-C and NO-CL O3 results (2 d from 2018 and 1 d from 2019) during the chamber burns. In contrast to the prescribed grassland burns, the Missoula chamber burns employed differing fuel types (ponderosa pine needles and fine woody debris) that are more typical of fuels consumed during western US forest fires. In addition, the concentrations of pollutants generated and observed during the chamber studies were approximately an order of magnitude smaller than those observed during the prescribed grassland fires. For reference, maximum PM2.5 concentrations observed during the prescribed fires were in excess of 50 mg m-3 while maximum chamber PM2.5 concentrations were less than 2 mg m-3. Regardless of these differences, there were still significant (order of magnitude or more) differences in the measurement results between the different FEM O3 instruments operated during both the 2018 and 2019 chamber studies. The NO-CL method showed identical trends to those observed during the grassland burns in that its measured O3 concentrations dropped to near zero during the active burning periods as indicated in Fig. 4 (active burning periods shaded in grey). The only periods when significant O3 concentrations were measured by the NO-CL method were when outside air was brought in to flush the chamber in between burns. The post-burn calibration checks on 23 April 2018 revealed a +8 % bias in the NO-CL method and a -2 % bias in the UV-C-H method. These biases were evident during the chamber flush periods on that day. Each analyzer was re-zeroed and spanned, resulting in the elimination of the bias between the two methods as observed in the results from the subsequent day (24 April 2018). No other calibration corrections were made during the 2018 and 2019 chamber studies. As in the grassland fire plumes, the UV-C method showed increased O3 concentration (positive analytical artifact) readings that were correlated in time with CO and NO2; see Supplement Figs. S9 and S10. Similarly, the UV-C-H instrument also showed increased positive analytical artifacts during the chamber burns but with absolute measurement values about an order of magnitude smaller than the UV-C instruments. The SL-UV method gave similar results to the NO-CL method during both the 2018 and 2019 chamber burns. Newly added during the 2019 burns, the UV-G method (2B 211-G) gave mixed results: at times it provided similar results compared to the NO-CL and SL-UV methods, and at others it provided results in line with those provided by the UV-C method. See Supplement Fig. S5 for the 2019 chamber burn time series plot. The burn average O3 concentrations from the 2018 and 2019 chamber burns are presented in Fig. 2.

During the 2018 chamber burns the UV-C results were biased high by 15–20 ppb even during non-burn (i.e., overnight) periods as evident in Figs. 4 (top panel) and S4. The initial hypothesis was that the bias was associated with high chamber backgrounds of interfering species due to years of heavy burning in the chamber. However, it was later discovered during a subsequent summer–fall 2018 ambient air study in North Carolina in the absence of smoke that sampling heavy smoke plumes during the fall 2017 prescribed grassland burns followed by subsequent storage of the UV-C analyzer irreversibly damaged the MnO2 scrubber in the UV-C instrument. It is hypothesized that the damage resulted in the scrubber removing some of the interfering species in addition to ozone, preventing them from being subtracted off as background in the reference measurement and subsequent detection as ozone (positive bias) during the measurement cycle. The effect of the bias was observed mainly when sampling ambient and chamber air and not readily observed during routine calibration checks (zeroes and spans) except for an increase in the time required to obtain stable zero and span values. The bias was not observed during any of the 2017 prescribed grassland burns. During the summer–fall 2018 North Carolina study and prior to the start of the 2019 chamber burns, a new MnO2 scrubber was installed and resulted in a significant and immediate reduction of the observed high bias, shown in Figs. 4 (bottom panel) and S5.

As discussed in Sect. 3.1 and 3.2, there are large (order-of-magnitude level) differences in O3 concentration measurement results obtained from the FRM (NO-CL) and the FEM UV photometric with catalytic scrubber (UV-C) O3 methods. The extremely low O3 concentrations measured by the NO-CL instrument are consistent with O3 depletion in the presence of high NOx concentrations (up to parts-per-million levels) observed in the grass burning plumes and during chamber burns. The reaction between NO and O3 is rapid and occurs on the timescales of seconds to minutes. As a result, high NO in the fresh biomass combustion plumes will efficiently titrate out O3, leading to near-field depletion within the plumes relative to background concentrations. There was no sign of a positive interference in the NO-CL monitors, and it remains the most robust and accurate routine method for O3 measurement in fresh and downwind biomass burning plumes.

In contrast with the NO-CL FRM instrument results, the UV-C FEM results showed substantial increases in reported O3 concentrations in the fresh biomass burning plumes. There is no known pathway for direct O3 emission from biomass burning, and the proximity (meters to hundreds of meters) and timescales (travel time of seconds to minutes from the combustion source to measurement) involved are too short for the usual NOx–VOC photochemistry to produce secondary O3. Further, since the FSL chamber interior is not exposed to sunlight, photochemistry would not have been active in the Missoula laboratory burns. For the purposes of this work, the positive analytical artifact in the UV-C method, termed ΔO3(UV-C), is estimated using Eq. (6) as the difference between UV-C and the NO-CL O3 concentration measurement results for the same time period: 6ΔO3(UV-C)=UV-C – NO-CL. Figure 5 shows in-plume regressions between ΔO3(UV-C) and the FRM measurement and CO for the three measured prescribed grassland burns in 2017 (Supplement Fig. S6 shows the time series of ΔO3(UV-C) and CO). Figures 5 and S6 show good correlations within the smoke plumes. The average and maximum values of ΔO3(UV-C) are summarized in Table 3. It is hypothesized that the large “O3” measurement observed in the UV-C method results from a positive interference or artifact, likely linked to VOC emissions in the grassland burn plumes. VOCs are emitted in higher concentrations from the smoldering phase of combustion, which is also characterized by large CO emissions (Yokelson et al., 1996, 1997), so a correlation between CO and O3 artifacts would support the hypothesis of a VOC-linked interference for the UV-C instruments. This is also consistent with observed VOC interferences in previous studies (Grosjean and Harrison, 1985; Kleindienst et al., 1993; Spicer et al., 2010) and observed following fireworks (Fiedrich et al., 2017; Xu et al., 2018).

The presence of a Nafion^®-based humidity conditioning system (Nafion^® tube dryer) significantly reduced the magnitude of the observed artifact as evident by comparing the UV-C and UV-C-H results shown in Figs. 1–3 and S1–S5. As with the UV-C method, the artifact in the UV-C-H method, ΔO3(UV-C-H), is calculated using Eq. (7) as the difference between UV-C-H and the NO-CL O3 concentration measurement results for the same time period: 7ΔO3(UV-C-H)=UV-C-H – NO-CL. The addition of the Nafion^®-based humidity conditioning system reduces the magnitude of the ΔO3(UV-C-H) artifact by approximately an order of magnitude compared with the UV-C method. This is further illustrated in the 2018 chamber burns, where prior to beginning the final burn day on 24 April 2018, a Nafion^® tube dryer (PermaPure, MD Monotube Dryer Series) was installed in the UV-C method (Thermo 49i), in effect converting it to a UV-C-H method. As shown in Figs. 4 and S4, the addition of the Nafion^® tube dryer significantly reduced the ΔO3(UV-C) artifact to a point comparable with that observed in the UV-C-H method (2B 205). A possible explanation for this effect is presented and discussed in Sect. 3.5. In addition, the previously described bias related to the damaged MnO2 scrubber was also reduced upon addition of the Nafion^® dryer to the UV-C method.

For the March 2017 Konza Prairie study (Fig. 1) and the 2018 and 2019 USFS chamber studies (Fig. 3) the SL-UV instrument concentration results were comparable to, although noisier and slightly higher than, the NO-CL reference instrument. On numerous occasions during the prescribed and chamber burns, the SL-UV instrument shows short (i.e., 1 min data point) positive or negative excursions that are not also observed in the NO-CL results. In addition, these excursions are not correlated with changes in CO concentrations. Because the SL-UV is a dual-cell instrument that measures O3 by comparing the absorbance of two cells, it is critical in highly dynamic environments (such as during this study) that both cells be measuring the same air at the same time. A slight difference in flow rates or residence times between the two pathways (or a delay in one pathway relative to the other) will cause short-term variability in the difference between the two cells. Although this does not pose an issue for longer time averaging (i.e., hourly data) under stable conditions, the dynamic nature of biomass burning plumes (i.e., changing on the order of seconds) and short time averages (i.e., minute) can create issues (noise) for the SL-UV method.

Significant analytical artifacts were observed for FEM UV photometric O3 instruments with (UV-C-H) and without (UV-C) Nafion^®-based humidity conditioning systems, where it appears that the dual effect of ambient humidity fluctuations and VOC interferences caused large positive over-measurement of in-smoke O3 concentrations. Chemiluminescence monitors are highly specific to O3 and have long been known to be free of VOC interferences (Long et al., 2014; U.S. EPA, 2015). However, studies have shown that the chemiluminescence method can be impacted by changes in relative humidity (Kleindienst et al., 1993). As such, upon promulgation in 2015, the new NO-CL FRM regulatory text requires a humidity correction–dryer system to eliminate the potential water vapor interference. As configured from the manufacturer, the NO-CL-based Teledyne-API model T265 instrument operated during this comparative study employs Nafion^® drying technologies to reduce or eliminate the water vapor interferences. The use of a chemical (NO) scrubber for UV photometric instruments (such as the 2B Technologies model 211) is very specific to O3 and shows a much better response than the catalytic scrubber instruments, performing almost as well as the NO-CL FRM, and has significant potential as a low-interference O3 method. Of the catalytic scrubber photometric instruments, those with Nafion^®-based humidity equilibration (2B Technologies model 205) perform significantly better than those without (Thermo 49 series).

In areas highly impacted by smoke or for studies focusing on biomass burning plumes, the use of a NO-CL FRM instrument is highly recommended as it was found to be essentially interference-free. These instruments are anchored to absolute O3 concentrations through the use of certified O3 calibration sources, many of which are based on UV photometry. The newest generation of commercially available NO-CL FRM instruments, including that used here (the Teledyne T265), have a built-in drying system to correct for the humidity artifacts that affected earlier-generation chemiluminescence instruments (Kleindienst et al., 1993), making remaining interferences negligible compared to other technologies.

The gas-phase chemical scrubber UV instrument (2B 211) did not perform as well as the FRM under the prescribed grassland burns or chamber experimental conditions tested here, with the high-time-resolution (1 min) data showing a much higher degree of variability than the NO-CL FRM instrument. We hypothesize that the main factor driving this divergence between this method and the NO-CL FRM is the dual-cell differential configuration of the instrument, which is not conducive to rapidly changing concentrations in O3 or other absorbing gases, such as VOCs.

In smoke-impacted monitoring situations where the use of a UV photometric instrument is still preferred or required, the choice of a monitor with humidity equilibration provides a significant analytical improvement over those monitors without the humidity corrections. In the absence of an instrument with a Nafion^® tube dryer and in non-regulatory applications, a dryer can be installed before the inlet or measurement cells to reduce the interference, as was demonstrated on the final day of the 2018 Missoula chamber burns. This will have the added benefit of reducing positive biases from humidity and reducing equilibration time for calibrations (especially when switching from high-humidity ambient air to dry calibration gases).

It is difficult to estimate an absolute magnitude or correct for the observed O3 analytical artifact since primary emissions from biomass combustion are highly variable and depend upon the fuel loading, fuel type and condition, phase of the fire, and the burn conditions (Yokelson et al., 1996, 1997). However, assuming the interference is driven primarily by VOCs, the artifact should be correlated with the excess CO (ΔCO = COplume–CObackground). Because CObackground during the prescribed grassland burns was below 200 ppb (relative to typical conditions of >2 ppm in the plume), ΔCO is estimated as the total measured CO concentration. A simplified view of biomass combustion assumes an approximate linear combination of two dominant emission phases, flaming combustion (characterized by emission of highly oxidized compounds, such as CO2, NOx, and SO2) and smoldering combustion (characterized by emission of reduced or mixed oxidation state compounds, such as CO, CH4, NH3, H2S, and most VOCs) (Yokelson et al., 1996, 1997). Because the majority of VOCs are in a reduced or mixed oxidation state, they tend to co-emit with CO during smoldering combustion, and the VOC concentrations tend to be highly correlated with CO in fresh biomass burning plumes (Yokelson et al., 1996). Scatter plots comparing the FEM instrument artifacts (ΔO3(UV-C)) and CO for the three prescribed grassland burning periods are shown in Fig. 5. Regression statistics of the comparison of ΔO3(UV-C) and ΔO3(UV-C-H) with CO and THC for grassland burns are given in Table 4. The magnitude of the artifact (estimated by the slope of the regression line of the CO vs. ΔO3 comparison), in parts per billion of apparent O3 per part per million of CO, ranges between 16–24 ppb ppm-1 for the UV-C instrument and 1.5–3 ppb ppm-1 for the instrument with humidity correction (UV-C-H). It is important to point out that CO, in and of itself, is not considered to be an interfering species in the UV photometric determination of O3 in that CO absorbs in the infrared (IR). The slight differences in the magnitude of the artifacts (fitted regression slopes) along with the low uncertainty (standard errors) values indicate that the magnitude of the artifact may be influenced by local conditions that make each burn unique. Such conditions might include meteorological conditions, fuel composition, fuel moisture content, and time spent in combustion phase (flaming vs. smoldering). Similar to CO, THCs and NO2 are indicative of combustion processes and are correlated with ΔO3 as given in Table 4 and Figs. S7 and S8. In terms of THC, the magnitude of the artifact, in parts per billion of apparent O3 per part per million THC, is significantly higher at ∼88 ppb ppm-1 for the UV-C instrument and ∼13 ppb ppm-1 for the UV-C-H instrument. Both the prescribed grassland and Missoula chamber burns resulted in what would be considered high PM concentrations (2–50 mg m-3). These high PM concentrations, however, are not considered to be interfering due to the presence of the inline particle filter assemblies described in Sect. 2.2 and 2.6.

Since the CO concentrations (from upwind fires) observed at most stationary sites from fire plumes are usually on the order of 1 ppm to greater than 10 ppm (Landis et al., 2018), it is reasonable to assume that O3 artifacts in the range of 15 ppb to greater than 250 ppb can be observed when employing a UV-C method. Similarly, O3 artifacts in the range of 1.5 to above 30 ppb might be observed at smoke-impacted sites monitoring with UV-C-H methods. As such, Nafion^®-based humidity conditioning systems are highly recommended for use if employing UV photometric methodology for monitoring O3 in areas impacted by wildfires or prescribed burns. As stated previously and as seen in Fig. 3 and Table 3, O3 artifacts were observed during the Missoula chamber 2018 and 2019 burns in both the UV-C and UV-C-H methods, although reduced compared to the prescribed grassland burns. The presence and magnitude of the O3 artifact strongly suggest that smoke generated from fuels typical of forests in the western United States also result in a measurement interference in UV photometric methods. Since downwind O3 production in biomass burning plumes is a significant issue in fire-impacted regions, having reliable, interference-free methods is critical for assessing the contribution of wildland fires to ambient O3 levels.

Figure 6 gives a detailed time series view of ΔO3(UV-C) and CO from 2 burn days from 2018 and a single day during 2019. As indicated, ΔO3(UV-C) and CO appear to be correlated in time, but when performing linear regression comparisons of ΔO3(UV-C) and CO during each year's chamber burns as a whole, correlations tend to be poor. We suspect the positive O3 bias is driven by one or more VOCs (likely oxygenated VOCs). In fresh smoke the excess concentrations of individual VOCs (ΔX) and VOC sums (ΔVOC) tend to be highly correlated with ΔCO (Yokelson et al., 1999; Gilman et al. 2015). The emission ratios of individual VOCs to CO (ΔX/ΔCO) can vary considerably with combustion conditions such as fuel type and condition (e.g., moisture content and decay state); fuel bed properties, such as bulk density; and the relative mix of flaming and smoldering combustion (Gilman et al., 2015; Koss et al., 2018). Additionally, the response of ΔX/ΔCO to burn conditions varies among VOCs. When each burn is considered individually or in groups with similar conditions, the correlations between ΔO3, CO, and THC are enhanced. An example of this behavior is shown in Fig. S10. For the chamber burns the magnitude of the ozone artifacts in parts per billion of apparent O3 per part per million of CO, ranges between 6–210 ppb ppm-1 for the individual burns. R2 and standard error values were consistent with those observed during the prescribed burns (see Table 4). The lack of a consistent relationship between the O3 artifact and ΔCO across all FSL chamber burns, while observing a good correlation for individual burns, likely reflects the variable response of artifact-producing emission(s) to the different combustion conditions of the burns.

One interesting observation from the data obtained from both the prescribed grassland and chamber burns is the order of magnitude difference in the average and maximum O3 artifact between the UV-C and the UV-C-H instruments as shown in Table 3. Considering that the prescribed grassland and chamber burns were conducted under dry (RH <50 %) conditions, the size of the difference (as large as hundreds of ppb) cannot be explained purely by the previously observed relative humidity effects on measurements (Leston et al., 2005; Wilson et al., 2006), suggesting that the Nafion^® dryer is directly impacting the concentrations of other interferents in the sample stream.

Nafion^® is a sulfonated tetrafluoroethylene polymer that is highly permeable to water but shows little permeability to many other organic and inorganic species (Mauritz and Moore, 2004). As a result, Nafion^®-based drying systems are often used as part of sample preparation or conditioning systems in analytical chemistry to remove water vapor from sample streams prior to sample analysis. Nafion^® membranes were introduced to some O3 monitors as a method to address humidity effects observed in UV-C O3 monitors, particularly when there are rapid changes in relative humidity level (Wilson and Birks, 2006). Humidity can affect the transmission of the UV light through the detection cell, and catalytic O3 scrubbers can modulate the water vapor in the scrubbed channel by acting as a temporary reservoir, resulting in significant positive or negative O3 interferences during rapid swings in relative humidity (Wilson et al., 2006). Adding a Nafion^®-based equilibration dryer immediately prior to the measurement cells reduces this water vapor interference without affecting O3 concentrations and thus significantly reduces the humidity artifacts in UV photometric O3 instruments.

Despite the high selectivity of Nafion^® to water vapor, it does demonstrate partial to complete permeability to various VOC or semivolatile organic compounds. Nafion^® membranes are highly permeable to alcohols, amines, ketones, and some water-soluble ethers (Baker, 1974), as well as some biogenic oxygenated compounds (Burns et al., 1983). In addition, Nafion^® membranes have been shown to catalyze the decomposition and rearrangement of monoterpene compounds (Burns et al., 1983). Systematic study of Nafion^® permeability and reactivity for polar and oxygenated compounds has been limited, with most users of Nafion^® membranes basing their use on operational testing and confirmation for specific applications.

The significant (order of magnitude) reduction in the O3 artifact with the addition of a Nafion^®-based dryer to the UV-C suggests that the Nafion^® dryer is directly impacting the major interfering species, which was hypothesized to be VOCs emitted during combustion processes. The species that are responsible for most of the O3 artifact in UV-C O3 instruments would have to be permeable through Nafion^® membranes or reactive with Nafion^® membranes; would have to be scrubbed by solid-phase, catalytic O3 scrubbers (such as MnO2 or hopcalite); and would have a significant absorption cross section around 254 nm. The absorption cross section of O3 around 254 nm is on the order of 10-17 cm2 molec.-1 (Molina and Molina, 1986), which means species with absorptions around 10-17 cm2 molec.-1 at 254 nm would be potential interfering species. As a class, aromatic VOCs and specifically oxygenated aromatic species (and other polar-derivatized species) absorb strongly in this region of the UV spectrum, and their potential permeability through Nafion^® membranes results in them being likely compounds to interfere in UV-C instruments. As an example, aromatic aldehydes such as o-tolualdehyde and p-tolualdehyde absorb around 5×10-18 cm2 molec.-1 and 4×10-18 cm2 molec.-1, respectively (Etzkorn et al., 1999). Both 2,4-dimethylbenzaldehyde and 2,6-dimethylbenzaldehyde have absorption cross sections above 10-17 cm2 molec.-1 at 254 nm (El Dib et al., 2008). Baker (1974) found 75 % of benzaldehyde was removed by a Nafion^® membrane, meaning that the Nafion^® permeability of tolualdehydes and dimethylbenzaldehydes is also likely to be high. In addition, benzaldehyde was almost quantitatively removed by several commercial catalytic O3 scrubbers, including the Thermo 49i MnO2 catalytic scrubber (Kleindienst et al., 1993), so similar aldehydes are likely to behave in a similar manner. Therefore, substituted aromatic aldehyde species are one class of compounds that fit the necessary criteria for causing the interference on the UV-C while having a reduced interference on the UV-C-H instrument. Future work examining the potential interferences from different species (or classes of species) on a species- or class-specific basis are required to confirm this potential mechanism and suggest others.