Environ. Sci. Technol. 2008, 42, 6040–6045
A Practical Alternative to Chemiluminescence-Based Detection of Nitrogen Dioxide: Cavity Attenuated Phase Shift Spectroscopy PAUL L. KEBABIAN, EZRA C. WOOD, SCOTT C. HERNDON, AND ANDREW FREEDMAN* Center for Sensor Systems and Technology, Aerodyne Research, Inc., 45 Manning Road, Billerica, Massachusetts 01821-3976
Received December 20, 2007. Revised manuscript received May 16, 2008. Accepted June 2, 2008.
We present results obtained from a greatly improved version of a previously reported nitrogen dioxide monitor (Anal. Chem. 2005, 77, 724-728) that utilizes cavity attenuated phase shift spectroscopy (CAPS). The sensor, which detects the optical absorption of nitrogen dioxide within a 20 nm bandpass centered at 440 nm, comprises a blue light emitting diode, an enclosed stainless steel measurement cell (26 cm length) incorporating a resonant optical cavity of near-confocal design and a vacuum photodiode detector. An analog heterodyne detection scheme is used to measure the phase shift in the waveform of the modulated light transmitted through the cell induced by the presence of nitrogen dioxide within the cell. The sensor, which operates at atmospheric pressure, fits into a 19 in.-rackmounted instrumentation box, weighs 10 kg, and utilizes 70 W of electrical power with pump included. The sensor response to nitrogen dioxide (calculated as the cotangent of the phase shift) is demonstrated to be linear (r2 > 0.9999) within ( 1 ppb over a range of 0-320 ppb (by volume). The device exhibits a detection limit (3σ precision) of less than 60 parts per trillion (0.060 ppb) with 10 s integration, a value derived from measurements at NO2 concentration levels of both 0 and 20 ppb;thedetectionlimitimprovesastheintegrationtimeisincreased to several hundred seconds. The observed baseline drift is less than ( 0.5 ppb over the course of a month. An intercomparison of measurements of ambient NO2 concentrations over several days using this sensor with a quantum cascade laserbased infrared absorption spectrometer and a standard chemiluminescence-based NOx analyzer is presented. The data from the CAPS sensor are highly correlated (r2 > 0.99) with the other two instruments. The absolute agreement between the CAPS and each of the two other instruments is within the expected statistical noise associated with the infrared laserbased absorption spectrometer ((0.3 ppb with 10 s sampling) and chemiluminescence analyzer ((0.4 ppb with 60 s averaging). The major limitation concerning accuracy is a direct spectral interference with phototchemically produced 1,2dicarbonyl species (e.g., glyoxal, methylglyoxal). However,
* Corresponding author phone: (978)663-9500; e-mail: af@ aerodyne.com. 6040
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 42, NO. 16, 2008
this interference can be readily removed by shifting the detection band to a slightly longer wavelength and ensuring that the lower edge of the detection band is greater than 455 nm.
Introduction We have designed, built, and tested a revised version of a monitor which detects nitrogen dioxide employing cavity attenuated phase shift (CAPS) spectroscopy that achieves near state-of-the-art performance in terms of sensitivity and stability (1–3). Moreover, this level of performance is obtained with an instrument that occupies a relatively small physical package (a 19 in.-wide rack-mounted instrumentation box weighing ∼ 10 kg), utilizes a modest amount of electrical power (∼70 W), and employs less costly components than those used in any other research or commercially available device for the detection of nitrogen dioxide. The CAPS-based monitor operates as an optical absorption spectrometer, utilizing a blue light-emitting diode (LED) as a light source and a sample cell incorporating two high reflectivity mirrors centered at 440 nm; a vacuum photodiode detector replaces a photomultiplier tube reported in an earlier communication (1). The square wave modulated light from the LED passes through the absorption cell and is detected as a distorted waveform which is characterized by a phase shift with respect to the initial modulation. The amount of that phase shift is a function of fixed instrument properties-cell length, mirror reflectivity, and modulation frequency-and of the presence of variable concentrations of nitrogen dioxide. With the improved version (2), we are able to demonstrate not only a detection limit (defined as 3σ, where σ is the standard deviation) of less than 0.060 ppb (by volume) in ten seconds integration time but also a baseline drift that is less than ( 0.5 ppb over several months use. The impetus for the development of this sensor has been the realization that the standard chemiluminescence-based sensor (labeled hereafter as CL) (4–6) used to detect ambient levels of nitrogen dioxide is subject to serious inaccuracies (7, 8). In places as diverse as Mexico City and rural environments in Switzerland, CL instruments which employ heated metal beds to reduce NO2 to NO have been observed to report NO2 levels which at times are twice as high as the true value because they are prone to interferences from compounds such as alkyl and peroxy nitrates. In addition, the presence of ambient humidity, which enhances chemiluminescence quenching and thus lowers readings, can cause errors of several percent (9, 10). A draft report from the U.S. EPA Integrated Science Assessment panel recognizes this deficiency and suggests that the hot metal oxide/chemiluminescence monitor is better suited to report total oxides of nitrogen (NOy) rather than individually speciated NO and NO2 (11). In the ambient air quality monitoring community, nitrogen dioxide must be monitored in almost 5000 locations in the United States as a criteria pollutant under the Clean Air Act of 1970. Currently, virtually all sites within the United States easily meet compliance standards (an annual mean of 53 ppb), and thus errors on the order of 10-20% in annual averages are not considered an issue. As compliance standards are lowered, however, the problems associated with the CL technique will pose a serious issue, especially since CL monitors generally provide readings that are erroneously high. We note that the European Union is poised to adopt the annual compliance standard for NO2 recommended by the World Health Organization of 21 ppb, and the California 10.1021/es703204j CCC: $40.75
2008 American Chemical Society
Published on Web 07/09/2008
FIGURE 1. Schematic of CAPS NO2 analyzer. Air Resources Board has approved lowering the standard within that state to an annual average of 30 ppb (12, 13). CAPS is only the first of several highly creative detection technologies, generally referred to as cavity enhanced optical absorption, which have been invented since the advent of high reflectivity (R > 99.9%) mirrors three decades ago (14, 15). In the past few years, a number of sensitive nitrogen dioxide sensors utilizing the most common form of cavity enhanced detection, cavity ringdown (CRD), have been reported (16–18). The only field-tested device is that of Osthoff and co-workers who detected NO2 at 532 nm using a frequencydoubled Nd:YAG laser (19). They report a precision of 0.12 ppb (3σ) in 1 s in the laboratory; in field use, an effective precision on the order of 0.15 parts per billion is attained because of temperature fluctuations and the need to subtract an interference from ambient ozone. Data from this instrument were shown to correlate with results from a CL analyzer that employs a photolytic converter (20) and is thus free from many of the standard CL instrument interferences. Successful laboratory demonstrations of CRD instruments have been reported by Zhang and co-workers (405 nm) and Orr-Ewing and co-workers (operating at 410 nm), both of which are estimated to have precision on the order of 0.15 ppb in reasonable integration times (21, 22). A recent preliminary report on the use of an optical feedback cavityenhanced absorption device by Orr-Ewing and co-workers, which relies on the laser sensitivity to optical feedback, indicates that the technique is also capable of reaching subppb detection limits (23). Kasyutich and co-workers employ the CAPS technique using a CW violet (405 nm) laser instead of a LED (24). With off-axis coupling of the laser beam to the optical cavity, they were able to achieve a detection limit on the order of 1 ppb using phase shift detection. Finally, Langridge and co-workers report a sub-ppb detection limit for NO2 using a spectrally resolved cavity-enhanced broadband absorption spectrometer utilizing a LED (25).
Experimental Section Reagents and Reagent Handling. For the laboratory studies, gas mixtures of known nitrogen dioxide concentration (to within 5% accuracy [2σ]) were created by admixing flows of gas from a calibrated high concentration nitrogen dioxide/ air mixture (19.0 ( 0.4 ppmv, Scott Specialty Gases) and ambient air free of CO2 and H2O (0.9999)-the residuals vary by less than ( 2 ppb from the nitrogen dioxide mixing ratio-from the highest NO2 concentration (∼320 ppb) down to the lowest (2.2 ppb). (See the Supporting Information.) We present results elsewhere demonstrating Rayleigh scattering measurements which are linear down to the equivalent of less than 1 ppb of NO2 (2). The value of the effective absorption coefficient, R (273.15 K, 1 atm), derived from the measured slope is 13.5 cm-1, a value which is approximately equal, within experimental error ((1 cm-1), to that calculated from detailed spectroscopic measurements (28, 29). The 5 orders of magnitude increase in light intensity in this instrument compared to that found in the initial version produces a dramatic increase in the signal-to-noise levels and a concomitant increase in instrument sensitivity. In order to demonstrate this improvement, the sensor was operated continuously for two 50 h segments without adjustment, first with 20 ppb of NO2 in nitrogen and then with pure nitrogen flowing through the sample cell. The signal was corrected for changes in cell pressure and thus nitrogen Rayleigh scattering. An Allan analysis of the data (30, 31), which presents the standard deviation of the data as a function of integration time of the sampling (see Supporting Information), indicates that the short-term noise (10 s integration period) is less than 20 parts per trillion (0.020 ppb), which represents a factor of almost 300 improvement over that measured with the previous version of the apparatus. The long-term drift (i.e., baseline stability) of the sensor has improved as well. Over the course of several months of instrument use in both the laboratory and the Aerodyne Mobile Laboratory, baseline readings stayed within a band of (0.5 ppb. The greatest excursions corresponded to large swings (15 K) in ambient temperature. Thus the need for periodic baseline measurements (using NO2-free air) depends on the variation in ambient temperature and the desired level of precision and accuracy. For instance, for normal ambient monitoring of nitrogen dioxide where an accuracy of 0.5 ppb is sufficient, baseline determination once or twice a day is all that is required. However, if higher accuracy is desired, the Allan analysis indicates that baseline measurements at intervals of ∼1 h or shorter are required. Interferences. Once the issues of linearity and baseline stability have been addressed, the largest uncertainty in the 6042
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 42, NO. 16, 2008
accuracy of the monitor is caused by the potential for chemical interferences. There are two classes of interferences which affect the accuracy of this sensor. The first comprises chemical species which demonstrate a spectroscopic overlap with the instrumentsi.e., they absorb at wavelengths within the measurement band of 440 ( 10 nm. This band was chosen as it was believed that it was essentially interference-free. As noted in our first correspondence (1), various halogen compounds (bromine, iodine, etc.) absorb in this spectral band; they are not expected to be present in normal ambient air monitoring situations. However, subsequent to the taking of the data presented herein, it was discovered that one class of direct spectral interference consists of 1,2-dicarbonyl compounds (the product of photolytic decomposition of isoprene and similar molecules) such as glyoxal (CHOCHO) and methyl glyoxal (CH3COCHO), which are yellow in color and absorb in the same spectral regions as nitrogen dioxide with approximately 20% of the absorption strength (32–34). Although concentrations of glyoxal and methyl glyoxal in rural areas are below 100 pptv, concentrations in polluted urban environments as high as 2 ppb have been reported, although the uncertainty in the measurements is often sizable (35–40). This interference is readily removed by simply shifting the lower edge of the spectral measurement band to slightly longer wavelengths (λmin > 455 nm). Using a detection band of 460-480 nm, for instance, there would be a slight loss of detection sensitivity for NO2 (∼25%). The only known interference with a species present in the ambient atmosphere at this wavelength is with ozone. A relatively high concentration of 100 ppb ozone would produce a response of ∼0.12 ppb NO2, assuming that no ozone was destroyed in the stainless steel components of the monitor (41). The report of an anomalously large spectral interference from water vapor at 405 nm-∼2.8 [ppb NO2]/[% H2O]-in measurements of nitrogen dioxide using a conventional cavity ringdown apparatus, suggests that water vapor could also provide an interference here (21). However, experimental data (42) indicate that the level of interference posed by the presence of several isolated absorption lines at 440 nm is small-∼0.08 [ppb NO2]/[% H2O] and is effectively counterbalanced by the decrease in the Rayleigh scattering coefficient caused by the presence of water vapor (-0.06 [ppb NO2]/[% H2O]). The Nafion dryer used in the monitor ensures that even under high humidity conditions (3% water vapor or a dew point of 25 °C), the humidity in the cell is always kept below 1%. Another salutary effect of the dryer is that it has a buffering effect with respect to humidity. Thus, in the laboratory, when the sensor is subjected to an abrupt increase in humidity (0% to 1% to 2%), no change (0.99), and the obtained slope
FIGURE 3. Point-by-point correlation plot of the data shown in Figure 2 shown on a log scale. Error bars (2σ) reflecting the expected statistical deviation are also shown.
FIGURE 4. NO2 measurements presented as 10 s averages taken using TILDAS and CAPS-based analyzers. is equal to 1 within the expected error in determining the span values for both instruments. The average point-bypoint deviation is consistent with the noise associated with the CL instrument ((0.4 ppb). (See the Supporting Information.) Although the intercept is virtually equal to zero, there is a noticeable systematic deviation from linear at concentrations below 2 ppb which is attributed to the response of the CL instrument. (See the discussion below.) The cause for this is unknown. Data for both the TILDAS and CAPS instruments obtained over a 2 day period are presented as 10 s averages in Figure 4. (See the Supporting Information for comparison over shorter time periods.) The two instruments were connected to colocated inlets to accommodate the fact that the TILDAS requires a mass flow that is a factor of 5 greater than that of the CAPS and also needs far more frequent baseline measurements (every ten minutes) for accurate measureVOL. 42, NO. 16, 2008 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
6043
FIGURE 5. Point-by-point correlation plot of the data shown in Figure 4 shown on a log scale. Error bars (2σ) reflecting the expected statistical deviation are also shown. ment. Because the QCL instrument has a time response of less than 1 s, the data from this instrument, taken at 1 s intervals, were convolved with a broadening function to approximate the time response of the CAPS instrument and inlet system (∼17 s) and then averaged to produce 10 s sampling periods. Figure 5 provides a point-by-point correlation plot between the TILDAS and CAPS instruments. The excellent agreement between the two instruments (r2 ) 0.998) is similar to that obtained above; on the average, the point-by-point discrepancy between the CAPS and TILDAS monitors is within the expected statistical error ((0.4 ppb with 10 s integration [2σ]). (See the Supporting Information.) The only major deviations (>1 ppb) occur when the TILDAS responds to plumes which are of short duration (
