An application of a dual tunable diode laser absorption spectrometer (TDLAS) to ambient measurements of formaldehyde (HCHO), nitrogen dioxide (NO2) and sulfur dioxide (SO2) is described. During the PM2.5 Technology Assessment and Characterization Study-New York (PMTACS-NY) 2002 intensive field campaign at Whiteface Mountain, New York, a dual TDLAS was deployed during a field instrument comparison study along with five other state-of-the-art trace gas measurement techniques. Prior to the field study, a thorough laboratory characterization of the instrument was undertaken to establish the optimum operational conditions including the selection of absorption features and the implementation of continuous laser frequency locking and rapid background subtraction. The Allan variance method was used to determine the optimal background subtraction cycle and to evaluate instrument performance. During the field campaign, measurements of HCHO, NO2 and SO2 were made by the TDLAS with the measurement precisions (1-min time interval, 1σ) of 80 ppt, 30 ppt and 40 ppt, respectively. The HCHO time series indicated a high variability of HCHO concentrations ranging from below the detection limit to 5 ppbv. The diurnal cycle of HCHO measurements showed higher concentrations during the late afternoon and early morning hours. The measured NO2 concentrations varied from less than the detection limit to the maximum of 25 ppbv. Broad NO2 plumes from long-distance transport of air masses and as well as high spikes from local pollution sources were observed during the campaign. The measured SO2 results also show high variability with the average concentration of 0.75 ± 0.95 ppbv (1σ).
 Since the early atmospheric measurements by the tunable diode laser absorption spectrometer (TDLAS) were reported using a long-path retro-reflector for urban CO measurements [Hinkley and Kelley, 1971; Ku et al., 1975], the method has been extensively used to measure a wide variety of gaseous species on the ground from both stationary and mobile platforms, as well from aircraft and ships at sea. Besides being applied to trace gas concentration (see reviews by Brassington [1995, and references therein], Kormann et al. , and Fried et al. ) and flux measurements in the field [Anderson and Zahniser, 1992; Wienhold et al., 1994; Edwards et al., 1994; Zenker et al., 1998; Horii et al., 1999], TDLAS is often employed as a reference method against which other methods are compared, because of its high specificity, freedom from interferences and fast time response. For example, Heikes et al.  carried out a formaldehyde measurement methods comparison in the remote lower troposphere with a TDLAS deployed along with four other techniques during the Mauna Loa photochemistry campaign. Gilpin et al.  performed a formaldehyde measurement comparison study with five other measurement techniques and a TDLAS system employed as a mission reference to confirm the absolute accuracy of the standard gases. Holloway et al.  conducted aircraft-based carbon monoxide measurements with a TDLAS system employed on the NOAA WP-3 aircraft to evaluate two vacuum UV florescence instruments.
 From 10 July to 7 August 2002, the PM2.5 Technology Assessment and Characterization Study-New York (PMTACS-NY) 2002 intensive measurement campaign was performed at the Whiteface Mountain base station (44°23.6′N, 73°51.5′W) of the Atmospheric Sciences Research Center, University at Albany, State University of New York. One of the primary PMTACS-NY objectives was to assess new measurement technologies and establish their potential for routine monitoring. In this campaign, a dual diode laser system developed by Aerodyne Research, Inc. was employed. Accurate measurements of atmospherically important gases, HCHO, NO2 and SO2 were performed by the dual TDLAS system, as well as five other detection techniques including an Alpha-Omega Methanalyzer, a silica gel DNPH cartridge technique, a photolytic NO-NOx analyzer, a TEI 42C NO-NOx analyzer and a TEI 43C pulsed fluorescence SO2 analyzer.
 In this paper, we will briefly describe the dual TDLAS system including the optical layout of the instrument, the electronics, and the data processing and spectral analysis techniques. Special emphasis is given to the discussion of the absorption features used for measuring HCHO, NO2 and SO2 concentrations in the field campaign, the determination of experimental parameters and the instrument performance both in laboratory experiments and field measurements. The time series of HCHO, NO2 and SO2 are presented, followed by the brief discussion of experimental results. More detailed experimental results and the corresponding instrument comparisons are given in a separate paper.
2. Dual TDLAS Instrument
 The dual TDLAS system employed at Whiteface Mountain campaign is a new generation of instruments developed by Aerodyne Research, Inc. It has been designed for real-time monitoring of atmospheric trace gases. Between two and six trace species can be monitored simultaneously. The major components of the TDLAS system including optical layout, data processing and analysis techniques have been described previously [Zahniser et al., 1995; Nelson et al., 1996; Horii et al., 1999]. Only a brief overview will be given here, and some new features incorporated into this dual TDLAS system will be emphasized.
Figure 1 schematically shows the layout of the dual-channel infrared spectrometer employed in this field campaign. The instrument mainly consists of two modules: optical and electronic. The optical module is built on a 2 by 4 foot aluminum honeycomb table which contains one liquid-nitrogen-cooled Dewar for temperature control of the tunable diode lasers and detectors, optics for laser beam collection and transport and a reduced pressure multipass absorption cell. The electronics module contains a fast computer with two data acquisition boards (National Instruments 6111E), and a dual laser control unit (Laser Components, Inc. Model 5830). A thermally conductive aluminum cover encloses the optical table. The whole TDLAS system is then surrounded by a foam-insulated weatherproof polyethylene case. The inside temperature can be closely controlled by the thin Minco heaters attached to the aluminum shell and optical table. This temperature control system is used to minimize thermal gradient on the optical table.
2.1. Optical Module
 The purpose of the optical module is to collect the light from two infrared laser diodes into a pair of beams, that are directed into a multiple pass absorption sampling cell and then direct the light leaving the cell back to a pair of detectors. A key element in the optical module is a low volume long path length astigmatic Herriott multipass absorption cell [McManus et al., 1995]. In addition to the main optical path, a portion of each laser beam is sent through a 5 cm long reference cell containing a high concentration of the gas of interest (in this experiment, HCHO, NO2 and SO2), in order to give a high-contrast spectral signal for line position locking.
 As indicated in Figure 1, the laser diodes are mounted on two separate gold-coated plates (up to four infrared laser diodes can be mounted on each plate) in a liquid-nitrogen-cooled Dewar, in which four infrared detectors (Kolmar Technologies, Inc.) are also installed. Light from each of the laser diodes is collected by two reflective microscope objectives (15×) and focused onto two 200 um pinholes, which define the input aperture. These pinholes are used mainly during the alignment, and removed after the optical alignment optimization is achieved. The microscope objectives are mounted on X-Y-Z translators which allow the light to be positioned and focused onto the fixed apertures. After the input aperture, each infrared beam is divided into two beams, signal beam and reference beam using a BaF2 window. The signal beam is focused into the astigmatic Herriott multipass absorption cell. The exit beams from the cell are steered back to the Dewar and focused on two signal detectors. The reference beams are directed through the reference cells and onto the reference detectors. Light passing through the reference cell has strong absorption features, which are used to characterize laser diode mode and identify the spectral line position for implementing the laser frequency position locking.
 To facilitate the optical alignment, a parallel visible optical system is installed. For each infrared laser channel, a red diode “trace” laser beam passes through a dichroic beamsplitter and is coaligned with the infrared beam. Coalignment is guaranteed by focusing the beam through the input aperture. In most cases, the alignment operation for the optical system is accomplished by observing the visible trace beams. The visible trace beam is also used to accurately calibrate the monochromator, via higher-order diffraction. The monochromator is very useful for characterizing the mode properties for a new laser diode and quickly identifying the laser wavelength during the instrument setup.
 The on-table astigmatic Herriott multipass absorption cell, developed at Aerodyne Research, Inc., provides a long path length of 153.50 m in a volume of 5 L. The small volume of the cell insures a fast time response (up to 0.5 s) depending on the pumping speed. The sample gas is pumped into the absorption cell from atmospheric pressure to 25 Torr (a typical sampling pressure in the field measurements) across the inlet orifice in order to reduce the pressure broadening of the absorption lines. The astigmatic Herriott multipass absorption cell consists of two spherical mirrors separated by nearly their radius of curvature. An optical beam is injected through a hole in one mirror in an off-axis direction, and it recirculates a number of times (174 times for this dual TDLAS system) before exiting through the coupling hole. The beams from the two laser diodes are angularly multiplexed such that the beams from side “A” of the Dewar where NO2 and SO2 are mounted have a horizontal opening angle, while the lasers from side “B” where HCHO diode is installed have a vertical opening angle.
2.2. Electronic Module
 The electronics module controls the laser diode frequency and processes the detected absorption signals to return trace gas mixing ratios. Both of these functions are controlled via a Windows 95/98 based data acquisition program “TDLWintel” developed by Aerodyne Research Inc. The computer sends commands to the laser controller, which in turn adjusts the laser diode temperature and average current as well as providing a fast ramp that sweeps the laser frequency across the trace gas absorption feature. The laser light is detected and converted to electrical signals digitalized by a fast data acquisition board. The analysis program calculates the change in absorption as the laser frequency is swept across the spectral feature of interest. The absorbance is fit to a calculated line shape, based on tabulated spectral parameters, and the measured sample temperature and pressure, to yield an absolute trace gas concentration.
2.3. Data Processing and Analysis Techniques
 The TDLAS system acquires an absorption spectrum and analyzes it by performing an advanced type of sweep integration. In this method, the full infrared spectral transition or group of transitions is acquired by rapidly scanning the current applied to the diodes, and then the area under the transitions is integrated using nonlinear least squares fitting to the known spectral line shapes and positions. At the beginning of each scan, the software turns the laser on and its frequency is swept across the desired transition frequency using a software generated voltage ramp. At the end of the scan, the laser current is dropped below threshold to determine the detector voltage corresponding to the absence of laser light. The sweep rate can be as fast as 20 kHz for a 150-point spectrum. Spectra are coaveraged in a background process while maintaining a 100% duty cycle. The resultant spectrum is fit to a set of Voigt line shape functions that are determined by the pressure and temperature. The baseline is treated as a slowly varying polynomial typically of third order.
 There are several advantages to this approach. First, absolute species concentrations are returned from the nonlinear least squares fits using predetermined line strengths, positions and broadening coefficients. The species concentrations are tied to absolute spectroscopic data found in the HITRAN database [Rothman et al., 1998]. With proper attention to laser mode purity and laser line width, the instrument does not require calibration, eliminating the need for calibration gas mixtures in the field. Second, the line shape functions are known from theory and can be calculated given the pressure and temperature of the sample. This is particularly important for field measurements where the ambient pressure and temperature may vary rapidly. Finally, the sweep integration approach allows up to 100 individual lines to be used in the “fingerprint” fit for each species and can fit up to eight species in a spectrum. Compared to monitoring at a single absorbing wavelength, this approach makes the retrieved concentrations less susceptible to potential interferences from other species absorbing in the same region, as well as those weak interference fringes inherent to such optical systems. Monitoring several transitions for one species can also enhance the specificity and sensitivity. Additionally, the capability of monitoring multiple species simultaneously can allow one to fit unknown lines which overlap the desired spectrum as a method of removing background absorption from unknown species. This advantage is particularly useful in the present application to the analysis of HCHO and NO2 with unknown lines overlapping the monitored absorption features. The corresponding discussion will be given in the following section.
3. Experimental Section
3.1. Absorption Feature and Measurement Selectivity
HCHO: The strong HCHO absorption feature at 2826.7 cm−1 was employed to measure HCHO concentration in the field campaign. The strongest line of 2826.7102 cm−1 has an integrated absorption cross section of 3.52 × 10−20 cm2 molecule−1 cm−1. Figure 2 depicts an HCHO spectrum, acquired from the 5 cm long reference cell with 30 Torr HCHO vapor present. As seen from Figure 2, the HCHO absorption feature includes six absorption lines, comprising an easily identified and very strong absorption feature in HCHO spectrum map.
 In the characterization of the HCHO laser diode, it was found that some HCHO spectra cannot be matched using the spectroscopic simulation with the line parameters from HITRAN database. This suggests that there must be errors in the HCHO line parameters in the HITRAN database. In the 2826.7 cm−1 region, two errors in HITRAN database have been identified. The first error is the incorrect line strength of line 2826.6817 cm−1, which is also discussed by S. C. Herndon et al. (Comparison of selected transitions in the v2 to the v1 and v5 bands of H2CO, submitted to Journal of Quantitative Spectroscopy and Radiative Transfer, 2003, hereinafter referred to as Herndon et al., submitted manuscript, 2003). The listed value in the HITRAN database is 2.510 × 10−20 cm2 molecule−1 cm−1. If one corrects the line strength to 1.255 × 10−20 cm2 molecule−1 cm−1 without changing any other parameters, all HCHO absorption features can be perfectly fit. The second found error is a missing HCHO line at 2826.6341 cm−1 in the HITRAN database. To perform accurate measurements, this missing line was added to the database used in our fingerprint fitting procedure. Neglecting this missing line in the fitting procedure introduces a negative interference to the HCHO measurements since its presence pulls down the polynomial base line. In Figure 2, the dash line represents the simulated fingerprint fit using the original line parameters in HITRAN database, and the solid line is the actual fingerprint fit with the corrected parameters.
 In the HCHO absorption feature region, the nearest H2O line is 2826.7810 cm−1, and its line strength is weaker than HCHO feature by over six orders of magnitude. In the fingerprint fitting procedure, the H2O absorption feature was fit as a second species to eliminate its potential interference to the HCHO measurement. In the HITRAN database spectroscopic simulation, H2O is the only interference species near the HCHO absorption feature. However, in the field campaign, one unidentified interference line next to the HCHO absorption line of 2826.6817 cm−1 was observed. This is indicated in Figure 3, which displays a 1-hour averaged HCHO spectrum sampled from ambient air at the field site. This interference line was fit as an unknown species in the fingerprint fit so that its presence did not affect the accurate measurements of HCHO.
 The mode property for the laser was characterized with a monochromator, which can produce a single spectral line from a broadband (multiwavelength) source. By scanning the monochromator through a wide wavelength range and measuring the corresponding laser output, a minor mode, about 11 cm−1 lower in frequency, was identified to account for 5% of the total diode laser output. The minor mode fraction thus determined was also checked by fitting the HCHO spectrum acquired from the reference cell with and without the monochromator, and then comparing the corresponding fitting results. Because of the impure mode, the HCHO measurements using the employed HCHO absorption feature may be low by approximately 5%. It is also possible that the unknown peak in Figure 3 could result from a strong absorption line in this weak mode. The mode structure of the laser appeared to be stable throughout the measurement campaign during which the laser temperature remained constant.
NO2: The NO2 absorption feature at 1593.3 cm−1 was employed. In this region, two strong absorption lines at 1593.2804 cm−1 and 1593.3120 cm−1 have similar line strengths, 7.466 × 10−20 cm2 molecule−1 cm−1 and 8.172 × 10−20 cm2 molecule−1 cm−1, forming a distinctive doublet. Figure 4 shows a NO2 spectrum with a nonlinear least squares fit, sampled from 5 cm long NO2 reference cell. Near the NO2 doublet, there are other six weaker NO2 lines, among which only 1593.2250 cm−1 line has a comparable absorption feature. In the data analysis, all of the NO2 absorption features in the monitored region were included in the fitting procedure.
 In the HITRAN database spectroscopic simulation, the employed NO2 absorption spectral feature is clear of known interferences. The nearest H2O line is 1593.1318 cm−1, and its line strength is 4 orders of magnitude weaker than that of NO2 doublet. Interference tests and ambient air measurements carried out in the laboratory before the field campaign revealed no other interferences. However, during the field campaign, an additional absorption feature between NO2 double lines was observed. Figure 5 shows a 30-min averaged NO2 spectrum, acquired during the field study on 24 July 2002. The extra interference line between NO2 lines is indicated in the figure and has an apparent absorbance of 2 × 10−4. In our data acquisition procedure, the interference line was included in the fit as an unknown additional species so that accurate measurements of NO2 were made in spite of its presence.
 After the campaign, a laboratory experiment was set up to identify this interference absorption feature. By challenging the system with water vapor, the interference peak was identified to be a strong water line in a minor mode. Characterizing the mode property with a monochromator proved that a low frequency minor mode, about 3 cm−1 away from the employed NO2 absorption feature, accounted for about 10% of the total diode laser output. It is possible that the mode purity changed before and after the field campaign since better mode purity (more than 95%) had been determined before the field campaign. Therefore, the NO2 measurements using this absorption feature may be underestimated by between 5 and 10%. We opted not to use the monochromator to isolate the primary mode during the field measurements since it decreases the amount of light transmission and increases optical fringes in the measurement system resulting in decreased detection sensitivity.
SO2: The SO2 absorption spectral feature used in this study is also composed of two strong absorption lines, 1353.2537 cm−1 and 1353.2918 cm−1. The absorption feature is clear of any interference in the laboratory and none was detected during the field campaign. Figure 6 shows an SO2 spectrum acquired during the campaign. Near the SO2 double lines, there is a strong CH4 absorption feature, 0.0947 cm−1 lower in frequency. This CH4 absorption feature does not interfere with SO2 measurement, but it does provide a good marker for locating this desired SO2 absorption feature among the rich SO2 absorption spectrum. The mode purity for the employed SO2 absorption feature was characterized to be more than 99%.
3.2. Setup of Measurement Parameters
 The TDL Wintel data acquisition program, as a central component of the dual TDLAS system, has flexible options for setting up experimental parameters for laser control and data acquisition. In the dual TDLAS system, two laser diodes independently run on two TDL Wintel data acquisition programs. In order to simultaneously operate two laser diodes for the accurate measurements of two species (or more), the selection and optimum setup of the experimental parameters was established through a thorough laboratory characterization. The experimental setup includes the frequency tuning rate of each laser diode, laser frequency locking, the schedule of liquid nitrogen refill, the employment of background subtraction, and the selection of data acquisition cycle.
3.2.1. Frequency Scan of a Laser Diode
 The laser frequency scan is set with two adjustable parameters: the central value and the slope of the ramp determining the modulated current to the laser. The central value step is the absolute value of the analog output at the center of the scan, and slope or ramp range is the number of steps of digital to analog converter (DAC) used in one complete scan of the diode laser. Since each laser diode has its own operation conditions, the selection of these parameters for each diode can be different. The Table 1 lists the experimental parameters for three laser diodes in generating the corresponding modulation waveform. The values listed for ramp range (−20 V to 20 V) and central D/A step (−10 V to +10 V) are the output voltages from the DAC (−10 V to +10 V), equivalent to the resulting injection current produced by the laser components L5830 controller (−200 mA to +200 mA). The second row in Table 1 gives the number of points for each laser diode scan, which is the number of data points acquired from the detector during a single frequency scan. As can be seen, the frequency scan for each laser has a different number of points. This ensures that two diodes are modulated asynchronously so that the small optical cross-talk from the absorption cell and electronic cross-talk from the detectors contribute only a negligible amount of random noise. It is noted that the central D/A steps in the Table 1 are initial number given at the beginning of the experiment. The program adjusts this value once the frequency-lock feature is applied.
Table 1. Experimental Parameters for Three Laser Diodes in Generating the Corresponding Modulation Waveform
Number of points
Ramp range in D/A step (−20 V to +20 V)
Central D/A step (−10 V to +10 V)
3.2.2. Setting Up a Spectrum Fit
 As described in section 2.3, the absorption spectra are fit using fingerprint fit procedure where the baseline is represented as a slowly varying polynomial of adjustable order. In setting up a spectral fit, the corresponding fitting parameters must be defined. These fitting parameters include species-specific fitting parameters and the laser tuning rate. The peak position corresponding to the fingerprint fit frequency is usually the strongest line in the spectral scan. In the present measurements, the absorption lines of 2826.7102 cm−1, 1593.3120 cm−1 and 1353.2918 cm−1 are used for the fingerprint fit frequencies of HCHO, NO2 and SO2, respectively, and the peak positions are fixed. The spectroscopic parameters of line frequency, line strength, broadening coefficient, and lower state energy are taken from the HITRAN database. In the measurement mode, the absorption cell pressure and temperature are used to calculate the Voigt line shapes for the fingerprint fit. In most cases, the frequency of the laser does not vary linearly with the applied voltage ramp. In the TDLAS system, the laser tuning rate (TR) is described as a three parameter model as follows:
Where x and y are in cm−1/channel, c is the channel number and z is dimensionless. In this model x is the linear tuning rate, y is the nonlinear tuning rate and z is the decay constant determining the range of influence for the nonlinear effects. These three parameters may be set by hand (using a reference spectrum) or they can be determined using a germanium etalon with a free spectral range of ∼0.05 cm−1.
3.2.3. Frequency Locking
 Like the frequency scan of a laser diode, the TDL Wintel program supplies several options for the frequency lock approach dealing with the laser frequency drift. In the present study, a reference spectrum acquired from the reference cell is always used as frequency standard to lock the signal peak position. In this study, the continuous frequency lock approach was chosen, meaning that the laser frequency is continuously checked. The program automatically acquires a reference spectrum and compares it to the previous reference spectrum. Any frequency drift is noted and used to adjust the peak position parameter in the fits. If the frequency drift is detected by one channel, it is compensated by modifying the DC offset of the modulation waveform. The modification of the DC offset is reflected by the variation of central D/A steps, an indication of the laser frequency drift during the experimental measurements.
3.2.4. Schedule of Liquid Nitrogen Refill
 The dual TDLAS system is designed for unattended, automated operation. The liquid-nitrogen-cooled Dewar is refilled automatically using a timer to trigger the cryogenic solenoid valve connected to a liquid nitrogen tank. During the field campaign, the liquid nitrogen refill system was scheduled to fill the liquid-nitrogen-cooled Dewar every six hours, and the duration for the liquid nitrogen fill was 6 min. The TDL Wintel was programmed to stop the data acquisition procedure 2 min prior to the start of liquid nitrogen refill, and then resume the normal data acquisition procedure 6 min later after the refill is completed. Filling the liquid nitrogen into the Dewar can cause the thermal disturbance to optical table temperature and optical alignment, resulting in rapid laser frequency drift and subsequent failure of laser frequency locking. The actual procedure prevents the cryogenic solenoid valve from being triggered too early and provides enough time for the system to be stabilized before the TDL Wintel starts to implement the frequency locking and data acquisition.
3.2.5. Rapid Background Subtraction
 In the setup of automatic background subtraction, the background interval and starting time of background subtraction must be coordinated between the two TDL Wintel programs. The interval is simply the time between successive background subtractions, which is set to be 60 s for all species in this study. The optimum selection of this parameter is based on the Allan variance [Werle et al., 1993] and will be discussed later. The backgrounds begin at 20 s after the minute on the system clock so that both instances of the program may be coordinated.
3.2.6. Data Acquisition Sequence
 In setting up data acquisition sequence, the program allows user complete flexibility to change any aspect of data acquisition operation procedure, including the optical cell flushing time, ambient integration time, and the frequency of background measurements relative to ambient measurements. The time resolution of the archived time series data is one second in this field study. In a typical sampling sequence, automatic background subtraction begins at 20 s after each minute, and the three-way solenoid valve is activated for dry nitrogen gas to flush the optical cell for 6 s. After this flushing period, a background spectrum is acquired for 15 s and averaged in real time, followed by a 6 s optical cell flush period with ambient air. After flushing with ambient air, background subtraction ambient spectra are acquired for 32 s. Each background subtraction spectrum is acquired by subtracting the average background spectrum from ambient spectrum point by point, and fit in real time. When the system clock turns to 20 s of the next minute, the second round of background spectrum acquisition begins with activating the solenoid valve to flush the optical cell with dry nitrogen. In this manner, the background acquisition-flush-ambient spectrum acquisition cycles are repeated, and a single data block in the time series data stream takes 1 min. Along with the time series data stream, all averaged spectra, including background subtraction ambient, background and reference, are saved for future reanalysis and data quality assurance tests.
3.3. Instrument Performance
 The instrument performance of almost all TDLAS systems is limited by optical noise (or optical fringes) generated by light scattering from various optical elements. In many cases, such optical noise contains multiple frequencies, amplitudes, and time constants, and its presence would limit the detection to optical densities of about 10−5 [Kormann et al., 2002]. By far the most dramatic improvement in suppressing the optical noise has been achieved by using rapid background subtraction. If carried out correctly, rapid background subtraction can effectively capture and remove optical fringes [Werle et al., 1993; Horii et al., 1999, Fried et al., 1998]. In order for background subtraction to be effective, acquisition of sample and background spectra along with associated cell flush times need to be accomplished within a characteristics system stability period, topt. We have characterized this stability period by using Allan variance technique [Werle et al., 1993]. In Figure 7, Allan plots for HCHO, NO2 and SO2 are generated by sampling relatively constant room air with an integration time of 1 s for all species. For each gas species, groups of stream data acquired for more than 5 min are analyzed to generate Allan plots, and at least three Allan plots are produced and evaluated. The measured mixing ratios for three gases shown in Figure 7 are 1.3 ppbv, 0.71 ppbv and 0.20 ppbv for HCHO, NO2 and SO2, respectively. As can be seen from the Allan plots, the three laser diodes behave similarly. First, the Allan variance decreases proportionally to the integration time, and then increases after the transition between white noise-dominated and drift-dominated regimes. The plots have indicated the optimum integration time topt of approximately 50 to 70 s before drifts become prevalent, which is comparable to other results published in literature [Werle et al., 1993; Fried et al., 1998; Kormann et al., 2002; Wert et al., 2003] for similar TDLAS systems. As stated earlier, our rapid background subtraction cycle of 1 min employed in the field campaign is based on the Allan variance plots.
 The square root of the minimum Allan variance in Figure 7 indicates the optimal precision with 60 s averaging time that can be obtained for the TDLAS system under ideal conditions in a laboratory environment. These are 23 ppt, 7 ppt and 13 ppt for HCHO, NO2 and SO2 respectively. For HCHO, at a sampling cell pressure of 25 Torr and path length of 153.5 m, 23 ppt corresponds to an absorbance of 0.8 × 10−6. This is similar to the result obtained by Fried et al.  who reported a HCHO optimal precision of 35 ppt with the integration time of 20 s, corresponding to an absorbance of 1.0 × 10−6. Fried et al.  system utilizes a low frequency wavelength modulation approach with a cell pass length of 100 m. More recently, Wert et al.  improved the precision of the NCAR TDLAS system and determined a stability period of 120–360 s and an optimal precision of 20 pptv, which is comparable to the precision obtained in this work.
 The minimum in the Allan variance may be used to predict the performance of the instrument under measurement conditions where each sample is compared with background by accounting for the duty cycle. In the background subtraction cycle where 30 s are for the ambient data acquisition and 15 s for the background, the expected reduction in noise due to averaging over this cycle would be ((1/30) + (1/15))1/2 = 0.3 instead of (1/60)1/2 = 0.13. Thus the replicate precision of a series of 60 s measurements with this duty cycle would be 2.5 greater than the minimum from the Allan variance plots, or 58 ppt, 18 ppt, and 33 ppt for HCHO, NO2 and SO2, respectively. During the field measurements, the actual replicate precisions (1 standard deviation) for consecutive 60 s cycles were obtained by sampling relatively constant ambient HCHO, NO2 and SO2 for 15 min. The 60 s replicate precisions based upon numerous such time periods are 80 ppt (1σ), 30 ppt (1σ) and 40 ppt (1σ) for HCHO, NO2 and SO2. Thus, the actual measurement precisions are greater than those predicted from the minima in the laboratory Allan variance plots by factors of 1.2 to 1.6.
 The laboratory and field study has proved that the temperature control system designed for our TDLAS system has worked well in minimizing thermal gradients on the optical table, which can result in optical fringe drift, adding noise and uncertainty to measurement. In the field measurements, we also found that the temperature fluctuations following each liquid nitrogen fill were significant in degrading the instrument performance. The effect of such temperature fluctuations often lasted from 10 min to one hour. It seemed to be that HCHO diode was more sensitive to this temperature fluctuation than NO2 and SO2 diodes. The resulting effect on HCHO measurements will be discussed later.
 The absolute accuracy of concentration measurements performed using tunable diode laser differential absorption spectroscopy is fundamentally determined by how well the line strengths used in the spectral fitting procedure are known. Additional factors which contribute to a systematic error in these measurements are as follows: path length, pressure, temperature, the line shape model used in the fitting procedure and diode mode purity. These factors for the overall instrument will be discussed first, and then the specific line strength uncertainty will be detailed. The uncertainty in the path length for the 174 spot/153.5 m astigmatic Herriot cell is estimated to be less than 0.1% or 15 cm [McManus et al., 1995; Herndon et al., submitted manuscript, 2003]. The derived quantity from the spectral fitting is a number density measurement; however, for convenience, this is immediately converted to a mixing ratio, which requires an accurate measurement of both pressure and temperature. The systematic uncertainty in the pressure and temperature measurements directly contribute to the accuracy of the measurement; however, both pressure and temperature play a subtler role in the spectral fitting. A Voigt line shape model [Humlicek, 1979; Armstrong, 1967] is used to fit the spectral data and extract concentrations for each of the species specified in a HITRAN style [Rothman et al., 1998] input file. With a “well-behaved” diode the uncertainty associated with the fitting procedure can be shown to be less than 1% (Herndon et al., submitted manuscript, 2003). When the operator has performed the experiments needed to demonstrate that the diode characteristics are within certain specifications; apparent laser line width <0.005 cm−1 and high mode purity >98%. When the diode is operating outside of these typical limits, the potential systematic error in the fit concentration increases. However, these arguments also assume that the concentration and path length are such that the minimum transmission is greater than 50% or that “optically thin”. The line strengths present in the HITRAN database for the 3.5 μm band of H2CO (ν1 + ν5) come largely from the determination of Brown et al.  who estimate the individual line strength assignments to be good to 5%. For SO2 in the 8 μm band (ν1 + ν3) the most recent work which the HITRAN line listings are based on the uncertainty estimate at 10% [Chu et al., 1998]. In the case of the 6.2 μm, ν3 band of NO2 the current estimate of the uncertainty in the band strength is 4% [Smith et al., 1985]; however, the most recent update to the HITRAN is based on the work which estimates the uncertainty of the line intensities to be 5% [Mandin et al., 1997].
4. TDLAS System Deployment in Whiteface Mountain Campaign
 Field experiments were conducted at Whiteface Mountain base station of the Atmospheric Sciences Research Center, University at Albany, located in the northeastern Adirondack Mountains. The sampling site for the TDLAS system and all other instruments was at the base lodge (2080 ft above msl), situated northeast of the summit on Marble Mountain. The base lodge is at the end of a dead end street. The traffic in the approach road to the base station is minimal and was generally limited to the participants of the campaign. An uphill graveled road (approximately 200 ft) leads to the sampling location from the parking lot. Several trailers, housing a wide array of instrumentation, were located in a forest clearing approximately 300 ft × 100 ft in area. The surrounding forest is identified as “Transition Zone” forest and is comprised of a mix of hardwood and conifer species. During the summer months the prevalent wind direction at Whiteface Mountain is southwesterly. Downslope winds develop in most summer evenings. Relatively low wind speeds were predominant at the Lodge base station (average ∼2 m s−1).
 The TDLAS was housed in an air-conditioned trailer. The sampling line is 1/4″ Teflon tubing with an inverted plastic funnel at the sample intake point. The tube extends 2 m above the roof of the trailer. The liquid nitrogen supply and the mechanical pump to draw sampling air into the multipass absorption cell were stored outside the trailer. The sample was maintained at reduced pressure (typically 25 Torr) by a rotary vane (Busch) pump with the typical pumping speed of 5 l/s, resulting in a residence time of ∼1 s in the multipass absorption cell. The TDLAS was configured to measure HCHO and SO2 from 10 to 21 July 2002 and HCHO and NO2 from 21 July to 7 August 2002.
5. Experimental Results and Discussions
5.1. Measurement of HCHO
 In the field study, one channel of the dual TDLAS was exclusively set to measure HCHO over the entire campaign from 10 July to 7 August 2002. Data coverage during this interval was greater than 90%. The missing data are due to the data dropouts during the liquid nitrogen refill, routine optical alignment optimization, line position locking failure, power surges and laser diode switching on the other channel in the middle of campaign.
 As mentioned in the instrument performance section, the measurement precision of the 1-min average data with the background subtraction employed is 0.078 ppb (1σ), which was obtained under the best instrument performance conditions. However, in most cases, the HCHO measurements demonstrated much worse instrument performance in the field. Such instrument degradation could be related to the hardware on this channel. Compared with NO2 and SO2 diodes, HCHO diode seemed to be more sensitive to the temperature fluctuation of the optical table during the liquid nitrogen fill. Figure 8 shows a part of HCHO time series in the original 1-s time interval, obtained in the field campaign. The data gap indicated in Figure 8 is as a result of the data dropout during the liquid nitrogen fill starting at 6 pm. The oscillation data pattern shown in the figure is the HCHO measurement results after the liquid nitrogen refill. Basically, the similar HCHO data oscillation pattern occurred each time after the liquid nitrogen fill, and the duration lasted from 10 min to 1 hour. These oscillations are due to thermal instability in the optical system resulting in the baseline fringes moving through the spectrum on a timescale shorter than the background subtraction cycle. In consideration of data quality, such data were initially believed to be of limited value. However, a refitted average spectrum over that period, shown in Figure 3, has exhibited a high signal-to-noise absorption feature, and the corresponding spectrum fit (1.78 ppb) agrees with the average concentration generated from the time series data, represented by the solid line in Figure 8. Refitting all the saved spectra has confirmed the agreement between the average spectrum fit and the corresponding average value from the time series data. It is strong evidence that the periodic noise caused by the temperature fluctuations can be effectively averaged with time. Therefore, the 30-min average data are reported for this field study.
Figure 9 presents the HCHO time series in 30-min time resolution. As can been seen, the HCHO concentration shows a high variability from below the detection limit to a highest value of 4.7 ppb during the campaign. The lowest value was observed at 5 pm, 27 July, and the highest value occurred at 5:30 am, 5 August. The average HCHO concentration is 1.4 ± 0.8 ppb over the entire campaign. A typical diurnal pattern can be seen from Figure 10, showing the 10-min average time series acquired from 19 to 21 July 2002. The diurnal cycle indicated in Figure 10 illustrates the higher HCHO concentrations in late afternoon and early morning hours. The photochemical oxidation processing of HCHO precursors, in particular VOCs might account for the higher concentrations in late afternoon hours. The higher HCHO observed in the earlier morning hours can be ascribed to the upslope wind since the meteorology data observed at the sampling site confirms that the wind direction was reversed from northwest/west to south/southeast after the sunrise.
5.2. Measurement of NO2
 Measurements of NO2 were made from 23 July to 7 August 2002, until the end of the campaign. The measurement results with 91% data coverage were obtained in original 1-s time interval. During the measurement period, the measurement precision level upon sampling relatively constant ambient air was consistent with the laboratory-measured value. As shown in Figure 11, presenting the NO2 time series data at its original 1-s time resolution, the measured NO2 concentrations varied from below the detection limit to the maximum of 24 ppb. The average concentration was 0.73 ± 0.59 ppb (1σ). The wide range of NO2 values is typical of the site, which samples very clean, local mountain flow, automobile exhaust plumes as well as polluted, westerly flow from Lake Placid urban area.
 As seen from Figure 12, which displays a part of the NO2 time series data acquired from 2 am to 6 am on 2 August, many narrow spikes were often observed in the nighttime hours during the campaign. Since the occurrence of these narrow spikes were highly correlated with the meteorology data, and there is very low automobile traffic around the sampling site, the sources of these spikes were believed to be from a vacuum pump from one of the other instruments on site. Some broad spread NO2 plumes like the one in Figure 12 were captured with both the TDLAS and two other NO2 measurement techniques. The high correlation between the occurrence of these plumes and the dominant westerly wind direction suggests that these are aged air masses transported into the area from Buffalo/Toronto region.
5.3. Measurement of SO2
 SO2 measurements were carried out from 10 to 21 July 2002. The time series of SO2 measurements at 1-s intervals is showed in Figure 13. Over the entire measurement period, the data coverage for SO2 measurements is more than 90%. The data dropouts during the liquid nitrogen fill and routine optical alignment during the campaign mainly account for the missing data. The time series of SO2 data for first three days look noisy and some negative concentrations were measured. These negative values represent instances when the background and signal spectra no longer represents the same conditions in the sampling cell, usually due to temperature drifts or the pressure surge when the gas flow is switched between ambient air and zero gas during the background subtraction cycle. In this case, the noise data were due to 0.1 Torr pressure disparity, which was identified three days later and subsequently eliminated by resetting the sampling system.
 As shown in Figure 13, the SO2 concentrations varied from the maximum value of 9 ppb to below the detection limit, and the average concentration over the entire campaign was 0.75 ± 0.95 ppb (1σ). As expected, there was no diurnal pattern observed for SO2 measurements. Two broadly spread SO2 plumes indicated in Figure 13 were observed. Both episodes occurred in early morning hours when the predominant wind was northwesterly. This was likely due to transport from a large forest fire in Canada which occurred during this period.
6. Summary and Conclusion
 During the PMTACS-NY Whiteface Mountain field campaign, the dual TDLAS system was deployed to measure HCHO, NO2 and SO2 from 10 July to 7 August 2002. A thorough laboratory characterization of the instrument was undertaken prior to deployment in order to establish the optimum experimental conditions including the selection of absorption features and setup of experimental parameters.
 The Allan variance approach has been used to determine the optimal background subtraction cycle and assess the instrument performance. The Allan variance calculation for three laser diodes has indicated that the dual TDLAS system exhibited stability periods as long as 70 s before systematic drifts started to degrade the performance. The actual measurement precisions are greater than those predicted from the minima in the laboratory Allan variance plots by factors of 1.2 to 1.6. The absolute measurement accuracies for HCHO, NO2 and SO2 come largely from the line strength estimates, and they are 5%, 5% and 10%, respectively.
 More than 90% of data coverage for HCHO, NO2 and SO2 were achieved over the entire field campaign. The 30-min average time series of HCHO indicated a high variability of HCHO concentrations ranging from below the detection limit to 4.7 ppb. The diurnal cycle of HCHO measurements showed higher concentrations during the late afternoon and early morning hours. The observed higher HCHO concentrations during the late afternoon were a result of photochemical oxidation of HCHO precursors, and the higher HCHO concentrations in the earlier morning hours can be ascribed to the upslope wind. The measured NO2 concentrations varied from below the detection limit to the maximum of 24 ppb. Several broad NO2 plumes as well as many NO2 high spikes were observed. These broad spread plumes are aged air masses transported into the area from Buffalo/Toronto region. The measured SO2 results also show high variability with the average concentration of 0.75 ± 0.95 ppb (1σ).
 This work was supported in part by the New York State Energy Research and Development Authority (NYSERDA), contract 4918ERTERES99, the U.S. Environmental Protection Agency (EPA) cooperative agreement R828060010 and New York State Department of Environmental Conservation (NYS DEC), contract C004210. Although the research described in this article has been funded in part by the U.S. Environmental Protection Agency, it has not been subjected to the Agency's required peer and policy review and therefore does not necessary reflect the views of the Agency and no official endorsement should be inferred.