Notice: Wiley Online Library will be unavailable on Saturday 27th February from 09:00-14:00 GMT / 04:00-09:00 EST / 17:00-22:00 SGT for essential maintenance. Apologies for the inconvenience.
 We have constructed a pulsed cavity ring-down spectrometer (CARDS) for simultaneous measurements of nitrogen dioxide (NO2), the nitrate radical (NO3), and dinitrogen pentoxide (N2O5) in the atmosphere. In this paper, we describe the development of the instrument to measure NO2 via its absorption at 532 nm. The NO2 detection channel was calibrated against a NIST traceable calibration standard as well as a photolysis-chemiluminescence (P-CL) NO2 detector. The absorption cross section of NO2 at 532 nm was determined to be (1.45 ± 0.06) × 10−19 cm2. The NO2 detection limit (1σ) for 1 s data is 40 pptv, and the instrument response is accurate within ±4% (1σ) under laboratory conditions. The linear dynamic range of the instrument has been verified from the detection limit to above 200 ppbv (r2 > 99.99%). For field measurements it is necessary to correct the CARDS NO2 signal for absorption by ozone. Under ambient conditions we report 1 s NO2 CARDS data with total uncertainty ±(4%, 60 pptv + 0.4 × (pptv/ppbv) × O3) (1σ). The instrument was deployed in the field during the New England Air Quality Study–International Transport and Chemical Transformation on board the NOAA research vessel Ronald H. Brown in the summer of 2004 and in Boulder, Colorado, in the winter of 2005. In both campaigns, CARDS and P-CL NO2 measurements were highly correlated (r2 > 98%), indicating the absence of interfering gas phase absorbers at 532 nm other than ozone and the suitability of CARDS to measure NO2 in the troposphere.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 The oxides of nitrogen, or NOx (= NO + NO2), play an important role in the troposphere because they impact ozone (O3) abundance and therefore air quality [Seinfeld and Pandis, 1998]. In the presence of sunlight, there is a photostationary state between NO oxidation by O3 and photolysis of NO2; perturbation of this relationship through NO oxidation by peroxy radicals gives rise to NOx-catalyzed net O3 production [Leighton, 1961]. Tropospheric NOx has both anthropogenic and natural sources; they include combustion processes, soil emission, and lightning. NOx is removed from the troposphere predominately via its conversion to nitric acid (HNO3), either by reaction of NO2 with OH during the day or via reactions of the nitrate radical (NO3) and dinitrogen pentoxide (N2O5) at night [Seinfeld and Pandis, 1998]. Tropospheric NOx mixing ratios may vary by up to four orders of magnitude from the marine background of less than 10 parts per trillion (pptv) to more than 1 part per million (ppmv) in urban areas. Because of this large range, high temporal and spatial variability, and relevance to regional air quality, high-frequency (i.e., 1 s time resolution), sensitive, and accurate measurements of NO and NO2 are of substantial interest.
 In the field, existing P-CL and LIF instruments rely on calibration standards, for example to characterize quenching of NO2 fluorescence by water vapor or variability of the photolytic conversion of NO2 to NO [Kley and McFarland, 1980; Matsumi et al., 2001; Ridley and Grahek, 1990; Ryerson et al., 2000; Thornton et al., 2000; Williams et al., 1998]. In contrast, absorption methods such as DOAS or TDLAS may be considered absolute as long as the analyte's absorption cross section is known, interferences are small or well quantified, and sampling losses are minimal. However, the main drawbacks of absorption methods have been either sensitivity or sample volume. For example, mid- and near-IR TDLAS instruments with optical multipass cells (optical path length > 200 m) [McManus et al., 1995] are highly selective in situ NO2 detectors but less sensitive (e.g., 30 pptv in a 60 s integration) [Li et al., 2004] than either P-CL or LIF. In DOAS, the absorption signals are usually integrated over long, open paths, which yields high sensitivity but limits this technique's ability to monitor NO2 in inhomogeneous and rapidly changing air masses.
 Instruments based on cavity ring-down spectroscopy and related techniques have the advantage of both long effective path lengths (e.g., several tens of kilometers) and a compact sample volume for in situ sampling. However, for NO2 detection, the full potential of CARDS and related techniques has yet to be realized, since they have yet to achieve the same sensitivity as TDLAS. For example, Mazurenka et al. reported a minimum detectable NO2 concentration of 400 pptv for 15 s integration at 410 nm, while Kasyutich et al. reported 1 s data with a detection limit of 1.2 parts per billion (ppbv) [Kasyutich et al., 2003; Mazurenka et al., 2003]. Using cavity attenuated phase shift spectroscopy, Kebabian and coworkers have recently reported a detection limit of 300 pptv for 10 s integration [Kebabian et al., 2005].
 This paper describes the design and construction of a CARDS NO2 measurement at 532 nm, which is currently embedded within a new three-channel instrument for simultaneous measurements of NO2, NO3 and N2O5. Concurrent NO2 measurements are needed to interpret NO3 and N2O5 observations [Brown et al., 2001, 2002a, 2002b,2003; Simpson, 2003; Wood et al., 2003], because the source of tropospheric NO3 is the reaction of NO2 with O3, and NO2 is in thermal equilibrium with NO3 and N2O5. Furthermore, NO2 varies considerably in space and time. Thus addition of a direct NO2 measurement capability to NO3/N2O5 instruments [Brown et al., 2001, 2002a, 2002b, 2003; Simpson, 2003; Wood et al., 2003] enhances their ability to characterize nocturnal nitrogen oxide chemistry. Furthermore, measurement of NO2 provides a potential method for monitoring the transmission efficiency of the NO3 and N2O5 inlets by conversion of these compounds to NO2, a more stable compound, via reaction with NO [Wood et al., 2003].
 Sample NO2 data from two separate field campaigns are presented. The first was the New England Air Quality Study–International Transport and Chemical Transformation (NEAQS–ITCT) in July and August 2004, during which the instrument was deployed on the NOAA research vessel Ronald H. Brown (R/V Brown) in the Gulf of Maine. The second was a ground based campaign in Boulder, Colorado, in January–February 2005, referred to here as the Boulder winter campaign. The laboratory and field validations of the NO2 instrument demonstrate the utility of a relative simple, single-wavelength, pulsed cavity ring-down spectrometer for in situ measurement of NO2 in the troposphere.
2. Experimental Section
2.1. Pulsed Cavity Ring-Down Spectroscopy
 Since its first description [O'Keefe and Deacon, 1988], cavity ring-down spectroscopy (CARDS) has become increasingly popular in spectroscopic applications because of its high sensitivity. With pulsed laser sources and high finesse optical cavities, it is possible to construct extremely sensitive field instruments with minimal experimental complexity [Atkinson, 2003; Brown, 2003]. For quantitative gas phase absorption measurements in cavity ring-down spectrometers, it is necessary to determine the ring-down time constant of the cavity in the absence of the optical absorber, τ0. The absorber concentration can then be derived from the ring-down time constant in the presence of the absorber, τ, using equation (1).
In equation (1), N is the number density of the absorbing molecule (molecules cm−3), α is the absorption (also referred to as extinction) coefficient (cm−1), σ is the absorption cross section (cm2 molecule−1), RL is the ratio of the total cavity length divided by the length over which the sample is present, and c is the speed of light (cm s−1) [Brown et al., 2002a].
2.2. Detection of NO2 at 532 nm
 The absorption spectrum of NO2 encompasses nearly the entire visible region, with a maximum at about 420 nm (Figure 1, top) [Burrows et al., 1998; Schneider et al., 1987; Vandaele et al., 2002; Voigt et al., 2002]. It is therefore possible to choose a wavelength most convenient from the point of view of instrument development. In the current and past versions of the spectrometer, 662 nm light for detection of NO3 and N2O5 was generated by pumping a tunable dye laser using the 532 nm output of a frequency doubled Nd:YAG laser [Brown et al., 2001, 2002a, 2002b, 2003]. The 532 nm YAG laser light was therefore a convenient choice for NO2 detection in this instrument. Detection of NO2 at 532 nm has several other advantages: highly reflective cavity ring-down mirrors are commercially available for this wavelength region, the absorption cross section of NO2 at 532 nm is relatively large, the only other major gas phase atmospheric absorber is O3 [Burkholder and Talukdar, 1994; Rothman et al., 2005], and the Rayleigh scattering cross section of air, which increases toward shorter wavelengths, is modest [Bodhaine et al., 1999; Bucholtz, 1995]. Since it is difficult to selectively remove ozone from the sample flow without also affecting the NO2 concentration, the absorption signal due to O3 must be subtracted for quantitative NO2 measurements. The total gas phase extinction signal at 532 nm, in the absence of optical extinction due to particles, is thus given by
The NO2 concentration in the cavity ring-down cell is then calculated from
 Quantification of NO2 at 532 nm therefore requires the simultaneous measurement of ambient O3, for example by a commercial UV ozone monitor. The ratio is about 52, so that the contribution of the O3 extinction signal to the NO2 measurements is only detrimental at low NO2 concentrations (see section 3).
2.3. Instrument Layout
 The optical layout of the spectrometer is similar to the previously described NO3/N2O5 instrument [Brown et al., 2001, 2002a, 2002b, 2003], with the addition of a third detection channel as shown in Figure 2 and described below. The output of a pulsed (20–100 Hz, up to 25 mJ), frequency-doubled Nd:YAG laser is passed through a beam splitter. Most of the laser beam energy (95%) is used to pump a tunable dye laser. The remaining 5% (∼1 mJ) of the 532 nm beam is passed through an optical isolator and collimated using a 50 cm focal length lens onto the back of a 2.54 cm diameter cavity ring-down mirror seated in a custom-built bellows and mirror mount. Each mirror is kept isolated from the sampling region by a purge flow of 50 standard cubic centimeters per minute (sccm) of ultrapure (“zero”) air. The mirrors have 100 cm radii of curvature and are located 107 cm apart. The active sampling region is 93 cm long and enclosed by 1.27 cm O.D. PFA Teflon tubing. Sample air enters and exits the sampling region through PFA Teflon fittings. The sample flow from the common inlet is diverted into separate flows for each detection channel. The sampling flow rates for the NO3, N2O5, and NO2 channels are typically set to 14, 6, and 5 standard liters per minute (slpm), respectively. Oil-free diaphragm pumps are used to pull sample air through the cavity ring-down cells. The sample pressure is measured using a pressure transducer located between the sample cell and the diaphragm pump. The sample temperature is measured with a thermocouple inserted in the sample flow.
 The optical cavities are mechanically stabilized using three 2.54 cm diameter carbon rods, which are mounted on a 120 cm by 60 cm optical bench using four custom-built holders. Light passing through the far mirror of the NO2 detection channel is collected using a fiber-collimating lens and passed through a multimode fiber-optic cable. The light is imaged onto a photomultiplier tube using a second fiber-collimating lens and a 532 nm band pass filter. The photomultiplier outputs of the three detection channels are digitized simultaneously at 1 MHz by 16 bit A/D converters, which are triggered from the Q switch pulse of the Nd:YAG laser.
 The operation of the instrument is automated and controlled by software developed in house. Usually, the signals from multiple consecutive laser pulses are coadded to produce 1 s data. Ring-down time constants are determined from a linear fit to the logarithm of the acquired trace immediately after acquisition, and the fitting parameters are stored to disk. The fit errors are small enough (e.g., 0.07 μs for 1 s data and time constant of 190 μs) to ensure that any potential artifacts arising from fitting single-exponential functions to multiexponential or nonexponential ring-down signals [Hodges et al., 1996; Naus et al., 2001; Yalin and Zare, 2002] are negligible and do not affect the accuracy of the measurement.
 Because aerosols are typically the largest contribution to atmospheric optical extinction in the visible region, sensitive gas phase absorption measurements require removing the aerosol particles from the sample stream. The air sample for the NO2 measurement was filtered through a 47 mm diameter, 25 μm thick Teflon membrane filter with a specified pore size of 1 μm mounted in a commercial polycarbonate housing. The filter traps particles of considerably smaller diameter than the specified pore size, such that it effectively removes particles from the air sample that scatter or absorb visible radiation. Loss of gas phase NO2 was not observed on these filters in laboratory tests. In the field, the filter was installed directly above the optical cavity and replaced every 48 hours.
 Because the instrument has typically been housed several meters away from the sampling point during field campaigns, a fast flow system was constructed to bring the air sample rapidly to the instrument inlet. The fast flow inlet line was mounted horizontally through a window of the container housing the CARDS instrument. A 1.27 cm O.D. elbow and an additional 20 cm section of 1.27 cm O.D. Teflon PFA tubing were connected to the outside end of the Teflon tubing. The inlet, facing downward, was secured in position and covered with an inverted plastic funnel to prevent sampling of precipitation and condensation. Fast flow (∼150 slpm, measured with a heated wire anemometer) was achieved with a large volume, high capacity blower. The fast flow line was brought as close to the cavity ring-down cells as possible and slower flows were tapped in for NO3/N2O5 and NO2 sampling. When the instrument was operated in the field, the entire PFA Teflon sample line was exchanged every 4 days to minimize losses of NO3 and N2O5 by contamination of the inlet.
2.4. Zero Measurement
 The ring-down time constant in the absence of the absorber, τ0, must be determined at regular intervals to account for potential drift due to changes in the cleanliness of the mirrors, sample pressure, sample temperature, and alignment of the optical cavity. Several methods to chemically remove NO2 from air have been reported in the literature. These usually involve reduction of NO2, for example by heated Mo wire [Clemitshaw, 2004]. A Mo scrubber showed promise in the laboratory, but proved to be problematic in the field due to variability of the conversion efficiency, which was less than 90%. Therefore zeroes were acquired by periodically filling the sampling region with zero air through an external, three-way solenoid valve located above the filter holder. The stability of successive τ0 measurements in the field (∼190 μs) was ∼0.1% for 15 min intervals. Values of τ0 between zero measurements were determined by linear interpolation.
 The zeroing method introduces a small difference in pressure between the sample and zero, which in turn leads to a measurable difference in extinction due to Rayleigh scattering (<8 × 10−10 cm−1 in extinction or <200 pptv equivalent NO2). We account for this difference in field measurements by subtracting the calculated Rayleigh scattering signal arising from the pressure difference. The uncertainty introduced by this correction is negligible.
 Zero determination in zero air neglects the contributions of ozone, water vapor, CO2, Ar, and other trace gases to the optical extinction at 532 nm. Refractive index changes due to CO2 (375 ppmv) or Ar (0.9%) are calculated to be below the detection limit. Laboratory experiments showed that the instrument is insensitive to the refractive index change or any absorption due to water vapor. Since we correct for O3 absorption (equation (3)), use of zero air does not, to our knowledge, introduce measurement artifacts outside of the uncertainties given in the next section.
3. Results and Discussion
 The remaining sections of this paper are structured as follows: Determination of reflectivity of the cavity ring-down mirrors, determination of the relevant absorption cross sections and their temperature and pressure dependence, uncertainties due to O3 subtraction and chemical changes in the inlet, and determination of the limit of detection and overall accuracy. Finally, we show results from laboratory and field intercomparisons with P-CL instruments.
3.1. Mirror Reflectivity
 The maximum achievable ring-down time constant in the absence of optical absorbers, and thus the maximum CARDS instrument sensitivity, is limited by Rayleigh scattering and the reflectivity of the mirrors. To determine the mirror reflectivity, we investigated the pressure dependence of the cavity ring-down time constant. At 70 torr, we observed ring-down time constants around 410 μs, compared to 190 μs at ambient pressure. The pressure dependence of the observed ring-down time constants is consistent with recent Rayleigh cross-section calculations [Bodhaine et al., 1999; Bucholtz, 1995] and a vacuum ring-down time constant τvac of 488 μs. From
and using L = 107.7 cm as the distance between two mirrors, the mirror reflectivity R is calculated to be 99.9993% at 532 nm.
3.2. Determination of Absorption Cross Sections
 To perform quantitative measurements, the absorption cross sections of NO2 and O3 at the wavelength and line width of the frequency-doubled Nd:YAG laser must be determined. The absorption cross sections of O3 in the Chappuis band are well known and does not exhibit fine-scale structure [Burkholder and Talukdar, 1994]. In contrast, the NO2 absorption spectrum [Burrows et al., 1998; Schneider et al., 1987; Vandaele et al., 2002; Voigt et al., 2002] exhibits considerable structure at 532 nm (Figure 1, middle), so that the effective absorption cross section is wavelength and line width dependent.
 The emission frequency of the Nd:YAG laser, as measured with a commercial wave meter, was 532.15 nm in air (532.30 nm in vacuum) and stable to within ±0.01 nm (±0.35 cm−1) in the laboratory. The manufacturer stated laser line width is 1.4 cm−1 FWHM. We calculated the effective 532 nm absorption cross section from the overlap between a 1.4 cm−1 Gaussian laser line and two high-resolution literature NO2 spectra [Vandaele et al., 2002; Voigt et al., 2002] with stated cross-section uncertainty of less than ±4% (Figure 1, bottom). This calculation is not sensitive to structure that may be superimposed on the nominally Gaussian line shape of the pulsed Nd:YAG as a result of longitudinal modes in the laser cavity. At a vacuum wavelength of 532.30 nm, the calculated effective cross sections are 1.48 × 10−19 cm2 and 1.45 × 10−19 cm2, respectively. In the observed range of laser frequencies, 532.29 to 532.31 nm, the calculated cross sections vary by ±0.7%. This uncertainty could be larger (e.g., up to 10% for a 0.05 nm drift) if the laser frequency is less stable in a field environment than in the laboratory.
 Variation in the NO2 cross section within the line width of the Nd:YAG laser can, in principle, give rise to multiexponential ring-down traces and thus concentration-dependent effective absorption cross sections [Hodges et al., 1996; Naus et al., 2001; Yalin and Zare, 2002; Zalicki and Zare, 1995]. We numerically modeled the ring-down traces as a function of NO2 concentration, and found that this effect is negligible (e.g.,<0.1%) for concentrations up to 10 ppmv, well in excess of those encountered in the troposphere. In addition, the observed calibration plots were linear from the detection limit to above 200 ppbv, the largest concentration tested (see below).
 For quantitative NO2 measurements, the ratio of the total cavity length to the length over which the sample is present (RL) must be accurately known. The effective sampling volume is reduced because of purge volumes that separate and protect the cavity ring-down mirrors from the sample gas, as seen in Figure 2. We determined RL in the laboratory by measuring the optical extinction due to O3, whose concentration was monitored simultaneously with two commercial UV absorption spectrometers to ±2% accuracy. Ozone was generated using a commercial ozone generator and diluted with zero air. The effective cross section of O3 in our spectrometer was σO3,eff = σO3/RL = (2.52 ± 0.05) × 10−21 cm2 at 50 sccm purge flow per mirror. Using the O3 cross section at 532 nm of σO3 = (2.79 ± 0.03) × 10−21 cm2 measured by [Burkholder and Talukdar, 1994], we obtain RL = 1.11 ± 0.03 (All quoted uncertainties in this paper are at the 1σ level unless specifically stated otherwise). The RL value was also measured by comparing the gas phase optical extinction of a constant flow of ozone at fixed concentration in the presence and absence of purge flow. In the latter case, the sample fills the entire optical cavity, so that the ratio of the two extinction measurements give an RL value independent of the O3 cross section and concentration. In these experiments, we obtained RL = 1.10 ± 0.03. The observed RL value is smaller than the estimation of 1.15 calculated from geometric measurements and indicates that sample air partially mixes into the purge volumes. This diffusion region encompasses about 1.3 cm3, and the purge flow of 50 sccm would flush this volume in about 1.5 s. The time resolution of the NO2 measurement is therefore not significantly affected.
 To measure σNO2 at the wavelength of the frequency doubled Nd:YAG laser directly in the spectrometer, we prepared a series of NO2 samples by flow dilution using a NIST traceable standard cylinder of 5.2 ± 0.2 ppmv NO2 in zero air, and recorded the optical extinction as a function of NO2 mixing ratio calculated from the dilution ratios. Plots of optical extinction against NO2 number density (e.g., Figure 3) ranging from the detection limit to 5 × 1012 molecules cm−3 (∼200 ppbv) are highly linear (r2 ≥ 99.98%), which indicates that the sample is not subject to line saturation and that the ring-down experiment is not significantly affected by variability of the absorption cross section over the laser line width. The plots have a slope equal to the effective NO2 absorption cross section at 532 nm divided by RL. The largest uncertainty in this determination was the accuracy of the concentration of NO2 delivered from the cylinder [Fried et al., 1988]. In what were deemed the best set of experiments, in which the sample lines and regulator were fully passivated, we obtained σNO2/RL = (1.30 ± 0.05) × 10−19 cm2 and σNO2 = (1.43 ± 0.07) × 10−19 cm2.
 Because of the difficulty in delivering a reproducible NO2 concentration from the calibration cylinder, we repeated the calibration with an existing NO-O3 titration system used for routine calibration of a P-CL instrument. The P-CL instruments are calibrated by titrating O3 with excess NO to produce a known quantity of NO2, which is highly reproducible [Ryerson et al., 2000] and accurate to ±3%. Plotting observed optical extinction at 532 nm as a function of amount of NO2 delivered and verified by P-CL measurements (Figure 3), we obtained σNO2/RL = (1.32 ± 0.04) × 10−19 cm2 and σNO2 = (1.45 ± 0.06) × 10−19 cm2. This result is consistent with the values from literature spectra calculated earlier (Figure 1, bottom).
3.2.1. Temperature and Pressure Dependence of σNO2
 The NO2 absorption cross section at 532 nm is slightly pressure and temperature-dependent [Vandaele et al., 2002; Voigt et al., 2002]. The flow rate through the NO2 cavity is slow enough for the gas sample to come to thermal equilibrium with the cell walls, which greatly reduces potential artifacts from ambient temperature fluctuations in field measurements. We have estimated the cross-section changes for typical temperature (15°C to 35°C) and pressure variability (±30 mbar or <3% of total pressure in the troposphere) in the sample cell and concluded that they are negligibly small. However, the pressure dependence of the absorption cross section could be an important consideration for deployment of a similar instrument in an aircraft or at high elevation under conditions of ambient pressure sampling.
3.3. Uncertainty due to O3 Subtraction
 The correction of the total gas phase absorption signal at 532 nm due to ozone is (in NO2 equivalent units)
 A typical ozone measurement by a commercial UV absorbance instrument is accurate to within ±(slope = 2%, offset = 1 ppbv). Thus the uncertainty associated with the O3 correction is ±(19 pptv + 0.4 × (pptv/ppbv) × O3). For example, at ambient O3 mixing ratio of 50 ppbv, the O3 subtraction increases the NO2 measurement uncertainty by ±39 pptv. Thus subtraction of O3 does not present a significant (i.e., >10%) uncertainty unless the ratio of O3 to NO2 concentrations is greater than about 250. Also, this uncertainty may be significantly larger if O3 and NO2 are not sampled from the same inlet and if the air mass is inhomogeneously mixed, for example due to local NO sources.
3.4. Chemical Changes in Inlet
 During the day, NO2 is in a photostationary state with NO, O3, and sunlight [Leighton, 1961].
 Sampling NO2 with a finite residence time in the dark inlets can alter this steady state and affect the accuracy of the NO2 measurement. Assuming the Leighton photostationary state [Leighton, 1961], a general expression for the relative change is
where tres is the sample residence time. For an inlet residence time of 1.4 s, this correction will vary between 0% (at night) and 2% (at the largest j value encountered in the troposphere) of the NO2 signal; the uncertainty introduced by this correction is negligible. Other chemical changes of the air sample in the inlet, such as NO2 production from thermal decomposition of N2O5 and NO2 loss on the aerosol inlet filter [Cantrell et al., 1993; Finlayson-Pitts et al., 2003], are too small to be significant sources of error in field measurements.
 In the above equation, δτ0 is the fractional uncertainty in τ0, given by σ(τ0)/τ0, where σ(τ0) is the standard deviation of repeated τ0 measurements. Using the 100 Hz laser and averaging for 1 s, we have observed average values for τ0 and σ(τ0) of 190 μs and 0.12 μs, respectively, under optimal laboratory conditions. Then, equation (9) yields a minimum detectable absorption coefficient αmin at the 1σ level of 1.7 × 10−10 cm−1. This corresponds to 1.1 × 109 molecules cm−3 or about 40 pptv NO2 in zero air at 760 torr and 298 K for 1 s averaging in the laboratory. This sensitivity could be improved by longer temporal averaging.
 During field measurements, the actual baseline noise was equivalent to an NO2 mixing ratio of ±60 pptv during NEAQS-ITCT 2004 and ±160 pptv during the Boulder winter campaign. Factors affecting the Boulder winter measurements were a lower pulse repetition rate of 20 Hz, turbulent flow noise due to temperature gradients in the sample line, poorer beam quality of the Nd:YAG laser and therefore poorer coupling of the laser beam to the optical cavity. Using the data observed during NEAQS-ITCT 2004, the instrumental limit of detection (1σ) in the field is
 For example, at ambient O3 of 50 ppbv the NO2 detection limit is 100 pptv. This detection limit is sufficient for tropospheric NO2 measurements in polluted environments.
 When deriving atmospheric concentrations of NO2 from CARDS, the accuracies of the following quantities must all be taken into consideration: τ, τ0, σNO2/RL, laser wavelength stability, sample pressure, temperature, dilution by purge flows, interference of ozone, and chemical changes in the inlet. The factors contributing most to the uncertainty of the CARDS measurement are σNO2/RL and laser stability, which have relative standard errors of ±3.0% and ±0.7%, respectively, and a combined uncertainty of ±3.7%. The other factors are estimated to be accurate within ±0.1% (τ and τ0), ±0.5% (pressure), ±0.2% (temperature), and ±0.2% (flow dilution). Uncertainties due to the zero measurement (i.e., background interpolation and Rayleigh scattering correction) are smaller than the statistical fluctuations of τ0 and therefore neglected. Addition of the uncertainties mentioned above (in quadrature) give a slope uncertainty of ±4% and a total uncertainty of ±(4%, 60 pptv + 0.4 × (pptv/ppbv) × O3) for 1 s data in field measurements.
3.7. Laboratory Comparison of CARDS With P-CL Instruments
 Subsequent to the initial calibration, we verified the CARDS instrument response and stability in a three-way blind laboratory comparison with two photolysis-chemiluminescence instruments. One of the P-CL instruments has been described in detail elsewhere [Ryerson et al., 2000]. In this paper, we will refer to this instrument as “aircraft P-CL” because it is usually deployed on board the NOAA WP-3 research aircraft. The other P-CL instrument is based on the same design as the aircraft P-CL, but with some minor modifications. In this paper, we will refer to this instrument as “ship P-CL” because it has been used aboard the NOAA research vessel Ronald H. Brown.
 Samples of NO2 in zero air were prepared by flow dilution of two standard NO2 cylinders and analyzed simultaneously by the three instruments. Stated slope and baseline uncertainties of the aircraft and ship P-CL 1 s data for this laboratory exercise were ±(5%, 50 pptv) and ±(15%, 100 pptv), respectively. The larger uncertainty of the ship P-CL measurements was mainly due to a sensitivity difference between the NO and NOx CL detection channels, which was not observed before (i.e., during NEAQS-ITCT 2004) or after the laboratory comparison.
 Data from the comparison are shown in Figure 4. Comparing 1 s data, we obtained a correlation of the CARDS NO2 measurements of (0.9722 ± 0.0001) [aircraft P-CL NO2] + (0.36 ± 0.01) ppbv (r2 = 99.996%) and (0.8959 ± 0.003) [ship P-CL NO2] + (0.24 ± 0.02) ppbv (r2 = 99.98%) over a period of two days. The three measurements were therefore in quantitative agreement within their stated uncertainties from the CARDS detection limit to 200 ppbv of NO2, although the comparison between CARDS and the aircraft P-CL instrument was clearly better.
3.8. Field Comparison of CARDS With P-CL NO2 Measurements
 Sample time series of CARDS and P-CL NO2 data from the two field campaigns are shown in Figure 5. In the field, the P-CL instruments are, in principle, sensitive to photolytic interferences [Ryerson et al., 2000]. The concentrations and conversion efficiencies of BrNO3, ClNO3, and HONO, which can be photolyzed in the P-CL inlet to NO2, were likely too small to have influenced either data set. Furthermore, scatterplots of both the difference and the ratio of the CARDS to P-CL field data versus the concentrations of, for example, NO, O3, peroxyacetyl nitrate (PAN), NO3, N2O5, NOy, water vapor, aerosol surface, and temperature, showed no correlation.
 In Boulder, the inlets of the two instruments were separated by less than 1 m, and the agreement was excellent, given that for the Boulder winter data the aircraft P-CL instrument's measurement uncertainty was ±(8%, 30 pptv), and the CARDS uncertainty was ±(15%, 180 pptv + 0.4 × (pptv/ppbv) × O3). However, on the first three days of the campaign, CARDS NO2 measurements were tightly correlated (r2 > 99%) but 20% larger than those of the P-CL instrument. In contrast, for the final five days the linear least squares fit parameters of the intercomparison of CARDS to P-CL data (averaged to one minute) were CARDS NO2 = (0.23 ± 0.01) ppbv + (1.049 ± 0.001) P-CL NO2, r2 = 99.1%. The correlation of the data was excellent (Figure 6), indicating the absence of an atmospheric interference. However, there was no obvious reason for the shift in the slope of the comparison, and this shift was not reproducible in the laboratory.
 For the R/V Brown data set, direct comparison of CARDS with the P-CL data was complicated by a 3.6 m vertical displacement of the instrument inlets. As a result, the two NO2 instruments were on occasion sampling different air masses arising from inhomogeneous vertical mixing, such as ship plumes. After filtering the R/V Brown data set for plume signals (identified by positive spikes in NO, CO, and/or CO2 levels and negative O3 spikes due to titration), the following least squares parameters were obtained: CARDS NO2 = (−0.009 ± 0.004) ppbv + (0.855 ± 0.001) P-CL NO2, r2 = 98.8%. The ratio of the two measurements tended to be slightly lower at night (∼0.81) than during the day (∼0.89), although the reason for the shift is unclear. The ship P-CL NO2 instrument's stated accuracy during NEAQS-ITCT 2004 was ±(11%, 93 pptv) at [NO]/[NO2] = 1, while the CARDS uncertainty was ±(8%, 160 pptv + 0.8 × (pptv/ppbv) × O3) at the 95% confidence level.
 Overall, the CARDS measurements in the field were bracketed by those of the two P-CL instruments, but the observed range of measurement ratios was large (0.81 to 1.20). Even though calibration differences of similar magnitude have been observed in previous intercomparison studies involving P-CL, LIF, and DOAS [e.g., Thornton et al., 2003], the observed slope differences are somewhat unsatisfactory and difficult to interpret after the fact. At the present time, we do not fully understand the reasons for the observed calibration differences in the field. Since the P-CL instruments' stability was substantiated by frequent calibrations, it will be imperative to monitor the CARDS instrument response with a calibration standard in future field campaigns to verify the long-term stability of the instrument.
4. Conclusions and Future Work
 A cavity ring-down spectrometer with adequate selectivity and sensitivity for in situ measurements of NO2 in the troposphere has been developed and combined with our previously described NO3/N2O5 spectrometer. The CARDS NO2 instrument was compared in the laboratory and in the field over a total of eight weeks with P-CL instruments under a variety of sampling conditions. The CARDS and P-CL NO2 measurements in both the R/V Brown and Boulder winter site data sets were highly correlated over wide concentration ranges of NO2 and were within the combined stated measurement uncertainties. The calibration drifts relative to the P-CL instruments in field measurements highlight the need for validation of the long-term stability of the CARDS instrument, for example with an automated calibration scheme. The degree of correlation between the two data sets corroborates the assumption that the only significant tropospheric absorbers at 532 nm are NO2 and O3 and demonstrates the capability of a single, fixed wavelength CARDS instrument to measure NO2 in the field. Future work will include implementation of a similar NO2 measurement capability in the aircraft version of our NO3/N2O5 CARDS instrument [Dubé et al., 2006] and construction of a NO3/N2O5 calibration scheme based on conversion of NO3 to NO2 by reaction with NO [Wood et al., 2003].
 The authors thank R. McLaughlin and M. Paris for designing and machining various CARDS spectrometer components, the crew and scientists on board R/V Brown during NEAQS-ITCT 2004, and J. Burkholder, J. Burrows, and A.C. Vandaele for allowing the use of reference spectra in this manuscript. Funding for this work was provided in part by NOAA's Air Quality program.