Development of a broadband cavity-enhanced absorption spectrometer for simultaneous measurements of ambient NO₃, NO₂, and H₂O

We describe the characteristics and performances of our newly built broadband cavity-enhanced absorption spectrometer for measurements of nitrate radical (NO₃), nitrogen dioxide (NO₂), and water vapor (H₂O). A customized vibration-resistance cavity layout incorporated with N₂ purging on high-reflection mirror surfaces was implemented with a red light-emitting diode (LED) as a light source. In general, this system achieved over 40 km (up to 101.5 km) of effective light path length at 662 nm from a 0.52 m long cavity. For accurate NO₃ measurement, the measured absorption spectrum of H₂O was used for simultaneous concentration retrievals with the other species instead of being treated as interferences to be removed or corrected prior to NO₃ detection. Synthesized N₂O₅ crystals under atmospheric pressure were used for performance tests of linear response and transmission efficiency. From the standard injection experiments of NO₃, NO₂, and H₂O, high linearities were observed (R²≥0.9918). The total NO₃ transmission efficiency through the system was determined to be 81.2 % (±2.9, 1σ) within the residence time of 2.59 s. The precisions (1σ) of NO₃, NO₂, and H₂O in 1 Hz measurement from a single pixel on the charge-coupled device (CCD) were 1.41 pptv, 6.92 ppbv, and 35.0 ppmv with uncertainties of 10.8 %, 5.2 %, and ≥20.5 %, respectively, mainly from the errors in the literature absorption cross-section. The instrument was successfully deployed aboard the Korean icebreaker R/V Araon for an expedition conducted in the remote marine boundary layer in the Arctic Ocean during the summer of 2021.

1 Introduction

The nitrate radical (NO₃) has drawn considerable attention due to its significant influence on nocturnal nitrogen oxide chemistry after the first observation in the troposphere (Noxon et al., 1980; Platt et al., 1980; Brown and Stutz, 2012). NO₃ is mainly produced from the oxidation of nitrogen dioxide (NO₂) by ozone (O₃, Reaction R1) and is in thermal equilibrium with dinitrogen pentoxide (N₂O₅) from its further combination reaction with NO₂ (Reaction R2).

The role of NO₃ as an oxidant especially for the unsaturated volatile organic compounds (VOCs) becomes more critical at night not only because of its extremely low abundance due to the losses by rapid photolysis (Reaction R3; Stark et al., 2007) and reaction with NO (Reaction R4) but also because of the negligible amount of photochemically induced hydroxyl radical. Particularly, alkenes with more than two double bonds from biogenic sources (e.g., isoprene and terpenes; Winer et al., 1984; Ng et al., 2017) and reduced sulfur compounds like dimethylsulfide (DMS; Allan et al., 2000) are susceptible to be oxidized by NO₃ (Reaction R5).

$$\begin{array}{}\text{(R1)}& {\displaystyle }{\mathrm{NO}}_{\mathrm{2}}+{\mathrm{O}}_{\mathrm{3}}\to {\mathrm{NO}}_{\mathrm{3}}+{\mathrm{O}}_{\mathrm{2}}\text{(R2)}& {\displaystyle }{\mathrm{NO}}_{\mathrm{3}}+{\mathrm{NO}}_{\mathrm{2}}+M\leftrightarrow {\mathrm{N}}_{\mathrm{2}}{\mathrm{O}}_{\mathrm{5}}+M\text{(R3)}& {\displaystyle }\begin{array}{rl}& {\mathrm{NO}}_{\mathrm{3}}+h\mathit{\upsilon }\to {\mathrm{NO}}_{\mathrm{2}}+\mathrm{O}{(}^{\mathrm{3}}\mathrm{P})(<\mathrm{587}\mathrm{nm})\\ & {\mathrm{NO}}_{\mathrm{3}}+h\mathit{\upsilon }\to \mathrm{NO}+{\mathrm{O}}_{\mathrm{2}}(<\mathrm{714}\mathrm{nm})\end{array}\end{array}$$

$$\begin{array}{}\text{(R4)}& {\displaystyle }{\mathrm{NO}}_{\mathrm{3}}+\mathrm{NO}\to \mathrm{2}{\mathrm{NO}}_{\mathrm{2}}\text{(R5)}& {\displaystyle }{\mathrm{NO}}_{\mathrm{3}}+\mathrm{alkenes},\mathrm{aromatics},\mathrm{DMS}\to \mathrm{product}\text{(R6)}& {\displaystyle }\begin{array}{rl}& {\mathrm{N}}_{\mathrm{2}}{{\mathrm{O}}_{\mathrm{5}}}_{\left(\mathrm{g}\right)}+({{\mathrm{H}}_{\mathrm{2}}\mathrm{O}}_{\left(\mathrm{aq}\right)}\mathrm{and}/\mathrm{or}{{\mathrm{Cl}}^{-}}_{\left(\mathrm{aq}\right)})\\ & \to (\mathrm{2}-\mathit{\phi }){{\mathrm{NO}}^{-}}_{{\mathrm{3}}_{\left(\mathrm{aq}\right)}}+\mathit{\phi }{{\mathrm{ClNO}}_{\mathrm{2}}}_{\left(\mathrm{g}\right)},(\mathrm{0}\le \mathit{\phi }\le \mathrm{1})\end{array}\end{array}$$

As shown in Reaction (R6), N₂O₅, a reservoir species of NO₃, can undergo heterogeneous reaction to form gaseous nitryl chloride (ClNO₂) and/or aqueous nitrate (Roberts et al., 2008). This uptake on aerosol can act as a dominant nitrate formation path under haze events (Chang et al., 2018; McDuffie et al., 2019; Lin et al., 2020; Liu et al., 2020). Meanwhile, ClNO₂ can be photolyzed to NO₂ and Cl radical after following sunrise; consequently, it not only conserves NO_x (=NO + NO₂) but also accelerates oxidation speed by adding the Cl radical into the atmosphere (Osthoff et al., 2008; Le Breton et al., 2018).

In addition, the important roles of daytime NO₃ and N₂O₅ have been reported in terms of their contributions to VOC oxidation and aerosol evolution (Geyer, 2003; Brown et al., 2005, 2016, 2017; Osthoff et al., 2006; Wang et al., 2014; Wang et al., 2020; Foulds et al., 2021). Brown et al. (2017) observed a non-negligible amount of N₂O₅ (up to 35 pptv) during the daytime under high NO_x conditions, and Foulds et al. (2021) found competitive NO₃ loss by VOC oxidation with photolysis, even in daytime. These findings indicate that the impacts of NO₃-driven chemistry are not limited to nighttime.

Because of high reactivity and thus short lifetime and low mixing ratio (a few to several parts per trillion by volume), observation of the ambient NO₃ is challenging. To our best knowledge, no commercial instruments are available at present, and only a few in situ measurement techniques have been used. Systems based on laser-induced fluorescence (LIF; Wood et al., 2003; Matsumoto et al., 2005) and matrix isolation electron spin resonance (MIESR; Mihelcic et al., 1993) have been reported. However, most of the contemporary instruments in active use even in intercomparison work (Dorn et al., 2013) are based on absorption spectroscopic techniques capturing the strong B²E'–X²A'₂ electronic transition of NO₃ at 662 nm (Yokelson et al., 1994). Differential optical absorption spectroscopy (DOAS), characterized by a long physical path length, has been widely used for several decades (Platt et al., 1980; Heintz et al., 1996; Allan et al., 1999; Geyer et al., 2001b; McLaren et al., 2004; Stutz et al., 2004; Vrekoussis et al., 2004; Li et al., 2007; Sommariva et al., 2007; Asaf et al., 2009; Wang et al., 2013; Lu et al., 2016). Meanwhile, more compact instruments with an optical cavity have been recently developed for NO₃ measurements; cavity ring-down spectroscopy (CRDS; King et al., 2000; Brown et al., 2001; Simpson, 2003; Ayers et al., 2005; Nakayama et al., 2008; Schuster et al., 2009; Flemmer and Ham, 2012; Hu et al., 2014; Wang et al., 2015; Li et al., 2018; Wu et al., 2020) and cavity-enhanced absorption spectroscopy (CEAS) are the most representative techniques. In particular, CEAS instruments with broadband light sources for NO₃ measurement were developed for field (Langridge et al., 2008; Varma et al., 2009; Kennedy et al., 2011; Wang et al., 2017; Suhail et al., 2019; Wang and Lu, 2019) and laboratory studies (Venables et al., 2006; Wu et al., 2014; Fouqueau et al., 2020) with the advantages of a capability for simultaneous measurements of multiple species and the applicability of cheap light sources (e.g., light-emitting diode (LED), arc Xe lamp, and supercontinuum radiation source).

One of the main difficulties for accurate NO₃ measurement by CEAS is H₂O treatment. Due to the strong but narrow absorption by H₂O around 660 nm (Ball and Jones, 2003), the contribution of H₂O on light extinction needs to be well characterized. Several methods such as a look-up table (Langridge et al., 2008; Varma et al., 2009; Suhail et al., 2019), iterative calculation (Kennedy et al., 2011), and frequent NO₃ zeroing via NO titration (Wang et al., 2017) were adopted to overcome this issue.

Based on the capability of species-specific absorption cross-section measurements by CEAS (Axson et al., 2011; Chen and Venables, 2011; Young et al., 2011, 2014; Kahan et al., 2012; Sheps, 2013; Thalman and Volkamer, 2013; Prakash et al., 2018; He et al., 2021; Wang et al., 2022), we suggest a new approach using measured H₂O spectrum for the simultaneous quantification of NO₃, NO₂, and H₂O, which is simple and efficient enough for atmospheric application. Through this paper, we present not only our newly built broadband cavity-enhanced absorption spectrometer (BBCEAS) with detailed descriptions of design and performances but also linearity test results with H₂O in atmospherically relevant ranges. Moreover, we also show the results of the shipborne measurement of NO₃ in the Arctic Ocean, indicating successful performance of the instrument in field application.

2 Instrumental setup

BBCEAS is a sensitive instrument to directly measure the abundance of a target species and/or its optical extinction properties, introduced by Fiedler et al. (2003). Details on the working principle can be found in Ball and Jones (2003) as well as Gagliardi and Loock (2014). Briefly, this technique is based on the measurement of light extinctions in a relatively broad wavelength range. A system based on this method basically consists of a broadband light source, a high-finesse optical cavity formed by a pair of high-reflection (HR) mirrors, and detector(s). The light from the source resonates inside the cavity and rapidly reaches a steady state with the wavelength-specific attenuated intensity through the loss processes of (1) transmission, diffraction, and absorption by HR mirrors, (2) scattering by particles and gases (i.e., Mie and Rayleigh scattering), and (3) absorption by sampled trace gases in the cavity. The transmitted light from the cavity is then detected by the spectrometer and charge-coupled device (CCD) or photodiode array to monitor the light extinction spectrum.

The schematic diagram of our single-channel BBCEAS is shown in Fig. 1. The material, dimensions, and design of the novel cage system on optical mounts for stable performance with respect to vibration and pressure changes during the field deployment were adopted from the glyoxal (CHOCHO) and nitrous acid (HONO) instrument described in Min et al. (2016). Detailed descriptions of the optical layout, flow system, and data acquisition system are given in the following section.

Figure 1 Schematic of the NO₃ BBCEAS system with light path (shaded in red) and gas flows (black arrows). High-reflection mirrors (M), band-pass filter (BPF), fiber collimator (C), mass flow controllers (MFCs), and sensors of precision pressure transducer (P) and thermocouple (T) are marked.

2.1 Optical layout

An LED (LZ1-10R202-0000, LedEngin, Germany) centered at 660 nm is used as a broadband light source and is mounted on a home-built module with copper plate, thermoelectric cooler, heat sink, XYZ translator (LP-1A-XYZ, Newport Corp., USA), and fan to control the position and temperature (20±0.1 ^∘C) precisely. The light from the LED is collimated by an off-axis parabolic mirror (50328AU, Newport Corp., USA) and enters the cavity. The HR mirrors (FiveNine Optics, USA) for high-finesse optical cavity are mounted at a distance of 51.5 cm occupied by a Teflon cell (1 in. outer diameter (o.d.) and 1 mm thickness). The light exiting the cavity is then focused by another parabolic mirror (50331AU, Newport Corp., USA), filtered through a colored band-pass filter (660±5 nm, FB-660-10, Thorlabs Inc., USA) and coupled to a fiber collimator (74-UV, Ocean Optics Inc., USA). The customized optic fiber (Seokwang Optical Co., Korea) with linearly assembled seven 200 µm diameter cores is aligned along the slit axis of the spectrometer (HRS-300MS-NI, Princeton Instruments Inc., USA).

Light is then transmitted to a diffraction grating (1200 G mm⁻¹, 750 nm blaze, 68 mm × 68 mm) and dispersed with respect to its wavelength. A CCD is used as a detector (PIXIS-2KX, Princeton Instruments Inc., USA) to monitor the spectra of the final transmitted light intensity and is cooled to −70 ^∘C to minimize the dark current. The wavelength coverage from 632 to 691 nm and the spectral resolution of 0.47 nm as full width at half maximum (FHWM) were calibrated with the narrow Ne emission lines (NE-2, Ocean Optics Inc., USA). The entire optical layout was housed in a temperature-controlled optic box to maintain constant performances regardless of environmental changes.

The overall configuration of the optics is similar to the system described by Min et al. (2016); however, the main difference is that we have an independent cavity for an individual channel rather than two channels sharing one parabolic mirror plate. This modification minimizes potential interferences of light leaking from the adjacent cell and provides convenience in operation and maintenance aspects (i.e., HR mirror cleaning and LED swapping).

2.2 Flow and data acquisition system

The air sample is drawn into perfluoroalkoxy alkene (PFA) inlet tubing with a constant flow rate of 2.5 slpm (standard liter per minute) by a mass flow controller (Alicat Scientific, USA) and a scroll pump (IDP-3, Agilent Technologies Inc., USA). The design of the coaxial inlet following Min et al. (2016) is used to minimize the pressure change during the mirror reflectivity measurements and sampling cycles (see Sect. 3.1.1). On the downstream of the inlet, a 2 µm polytetrafluoroethylene (PTFE) membrane filter (R2PJ047, Pall Corporation, USA) is used to minimize the light extinction owing to Mie scattering by sampled aerosols in the cavity as well as to prevent reflectivity degradation of HR mirrors. The air then passes through 1/16 in. inner diameter (i.d.) PFA tubing into the cavity to minimize NO₃ loss by shortening the residence time with the scheme of reduced pressure operation as in Fuchs et al. (2008). The temperature and pressure inside the cavity are measured at the outflow of the cell.

Another difference with the cavity system in Min et al. (2016) is the addition of a purging system. As an active strategy to prevent sampled air contact on the HR mirror surface, ultra-high-purity (UHP) N₂ flows (>99.999 %, 20 standard cubic centimeter per minute, sccm) are introduced on each side of the cavity mirrors via custom-designed PFA cell flanges with orifices (50 µm, SS-1/8-Tube-50, Lenox Laser, USA). The effect of purge flow on the volume occupied by the air sample in the cavity was estimated as in Sect. 3.1.2.

Our instrument is operated and controlled automatically with the customized software programmed by LabView (National Instruments) for flow rates and temperatures of LED and CCD. The transmitted light spectra and other auxiliary data including temperatures, pressures, and flow rates for the extinction spectrum calculation are acquired via this program as well.

3 Characterization

From the spectrum of measured light extinction inside the cavity, α(λ), the number density (x_i) of species i can be calculated from Eq. (1).

$$\begin{array}{}\text{(1)}& \begin{array}{rl}\mathit{\alpha }\left(\mathit{\lambda }\right)& =\sum T_{\mathrm{i}}{\mathit{\sigma }}_i\left(\mathit{\lambda }\right)x_i+{\mathit{\alpha }}_{\mathrm{Mie}}+{\mathit{\alpha }}_{\mathrm{Rayleigh}}\\ & =\left({\displaystyle \frac{\mathrm{1}-R\left(\mathit{\lambda }\right)}{d_{\mathrm{eff}}}}\right)\left({\displaystyle \frac{I_{\mathrm{out},\mathrm{0}}\left(\mathit{\lambda }\right)}{I_{\mathrm{out}}\left(\mathit{\lambda }\right)}}-\mathrm{1}\right)\end{array}\end{array}$$

Here, T_i is the transmission efficiency from the inlet to the cavity, σ_i corresponds to the absorption cross-section, R stands for mirror reflectivity, and d_eff is effective cavity length. The α_Rayleigh and α_Mie refer to the optical extinctions due to Rayleigh and Mie scattering, while we neglect α_Mie owing to the aerosol filter on the inlet airway. The intensities of light transmitted from the cavity with and without the absorbing species are symbolized as I_out and I_out,0, respectively. Here, we define I_out,0 as the light intensity when the cavity is filled with dry zero air (ZA) only. Thus, for the accurate quantification of x_i based on Eq. (1), not only the reference spectrum σ_i but also the instrumental parameters such as d_eff, R, I_out,0, and T_i should be characterized beforehand.

3.1 Determination of cavity parameters

3.1.1 Mirror reflectivity, R(λ)

R(λ) can be derived from the well-known Rayleigh scattering differences in two species, and we selected helium (He) and ZA as shown in Eq. (2). For our instrument, a 2.75 slpm flow of He or ZA (99.999 % each, Daedeok Gas Co. Ltd.) is overflowed into coaxial inlet tubing so that 2.5 slpm flow is introduced into the cavity, the same as the flow rate of the sample, while the rest of the flow is streamed towards the outside of the inlet to minimize the pressure change inside the cavity.

$$\begin{array}{}\text{(2)}& \begin{array}{rl}& R\left(\mathit{\lambda }\right)=\mathrm{1}-d\\ & \left[{\displaystyle \frac{I_{\mathrm{ZA}}\left(\mathit{\lambda }\right){\mathit{\alpha }}_{\mathrm{Rayleigh}{,}_{\mathrm{ZA}}}\left(\mathit{\lambda }\right)-I_{\mathrm{He}}\left(\mathit{\lambda }\right){\mathit{\alpha }}_{\mathrm{Rayleigh}{,}_{\mathrm{He}}}\left(\mathit{\lambda }\right)}{I_{\mathrm{He}}\left(\mathit{\lambda }\right)-I_{\mathrm{ZA}}\left(\mathit{\lambda }\right)}}\right]\\ & =\mathrm{1}-{\displaystyle \frac{d}{L_{\mathrm{Light}}\left(\mathit{\lambda }\right)}}\end{array}\end{array}$$

In Eq. (2), I_i refers to the transmitted spectral intensity filled with gas species i. L_Light is the theoretically calculated effective light path length under the assumption that light attenuation is solely driven by the mirror and the cavity length, d. The ${\mathit{\alpha }}_{\mathrm{Rayleigh}{,}_{\mathrm{i}}}$ was calculated from the literature Rayleigh scattering cross-sections of N₂ (Bodhaine et al., 1999), O₂, and He (Shardanand and Rao, 1977) considering pressure and temperature changes.

Figure 2 Cavity characteristics of 1 min averaged (a) transmitted light intensities with zero air (ZA) and He injection, (b) mirror reflectivity, R (dashed black line is fitted with fourth-order polynomial), and (c) effective light path length, L_Light. Shaded area in light red represents fitted range for the number density retrieval.

Figure 2 shows the cavity characteristics such as the I_ZA(λ), I_He(λ), R(λ), and L_Light(λ) as examples acquired during the $T_{\mathrm{E},{\mathrm{NO}}_{\mathrm{3}}}$ quantification experiments (described in Sect. 3.3.2). Although the difference of the transmitted light intensities between He and ZA injections was largest near 662 nm (Fig. 2a), the wavelength corresponding to the highest R(λ) value was shifted to 672.5 nm due to the incorporated wavelength-dependent characteristics of Rayleigh scattering and mirror reflectivity (Fig. 2b). L_Light exceeded 98 km (Fig. 2c) within the wavelength range for fitting (shaded in light red in Fig. 2). Especially at 662 nm, where NO₃ absorption peaks, R and L_Light were 99.9995 % and 101.5 km, which are superior values to other reported ones from previous BBCEAS studies (Table 1 in Sect. 3.5). The uncertainty for R(λ) estimation is 2.2 %, mainly from an error in the Rayleigh scattering spectrum of ZA (2 %; Washenfelder et al., 2008) and rarely from the temperature and pressure measurements (0.7 % and 0.5 %, respectively).

Table 1 Comparison of BBCEAS performances for NO₃ measurement.

^a U – urban region; SU – suburban region; M – marine region. ^b Systems with open cavity. ^c Ambient NO₃ was well below the detection limit through the whole measurement period.
^d Average values of reflectivity are noted instead of the maximum value. ^e NA – not available.

3.1.2 Effective cavity length, d_eff

The volume occupied by the sample in the cavity should be accurately evaluated. Due to the reduction in the sample volume by addition of the purge flow, the d_eff differs from both the cavity length defined by mirror separation (51.5 cm) and the physical displacement of sample in and out of ports (47.5 cm). This parameter has been commonly quantified by the injection of known amounts of standard gas such as H₂O (Kennedy et al., 2011), O₃ (Dubé et al., 2006; Fuchs et al., 2008), and NO₂ (Schuster et al., 2009; Wang et al., 2017). In this study, we filled the cavity with 5 ppmv NO₂ standard (in N₂, Korea Research Institute of Standards and Science, KRISS) and compared the retrieved NO₂ number densities with and without the purge flow. From this experiment, we concluded that the purge volume takes 5.25 cm³ in the cylindrical cell (1 in. o.d. and 1 mm thickness) and thus estimated d_eff as 50.28 cm with an uncertainty of 5.2 % (1 % from the standard gas, 2.9 % and 2.6 % from retrieval errors, and 3.2 % from ${\mathit{\sigma }}_{{\mathrm{NO}}_{\mathrm{2}}}$ reported by Bogumil et al., 2003).

3.2 Retrieval of number density

3.2.1 Absorption cross-section, σ_i

Number density x_i is determined from the optimal fit of σ_i(λ) on the light extinction spectrum, α(λ), within broad spectral ranges. For ${\mathit{\sigma }}_{{\mathrm{NO}}_{\mathrm{3}}}$, the convolved spectrum of the literature cross-section from Yokelson et al. (1994) was used because of its simple outstanding absorption features as shown in Fig. 3. Its uncertainty was reported as 10 %.

Figure 3 Absorption cross-sections of NO₃, NO₂, and H₂O used and fitting range (shaded in light red) for number density retrievals.

Measured spectra were used for H₂O and NO₂ to compensate for the imperfections in line-shape determination. These could originate from the astigmatic bias in CCD pixels and from the environmental changes (i.e., pressure and temperature) in instrumental operation conditions. In order to minimize the impacts of these accumulated errors on concentration retrievals, measured spectra were used in previous studies (Min et al., 2016; Liang et al., 2019; Barbero et al., 2020) based on the fact that the CEAS technique is widely used to characterize the wavelength-dependent light extinction properties of chemical species (Thalman and Volkamer, 2010; Axson et al., 2011; Chen and Venables, 2011; Young et al., 2011, 2014; Kahan et al., 2012; Sheps, 2013; Thalman and Volkamer, 2013; Prakash et al., 2018; Jordan et al., 2019; He et al., 2021; Wang et al., 2022).

As mentioned before, the treatment of H₂O absorption features is crucial for accurate NO₃ measurement due to its strong but narrow absorption lines around 660 nm. To prevent this issue, the measured H₂O spectrum scaled with the literature, and a relative humidity (RH) probe was used as ${\mathit{\sigma }}_{{\mathrm{H}}_{\mathrm{2}}\mathrm{O}}$. For that, H₂O was injected via the constant flow of ZA through a deionized water bubbler at a temperature of 22.7 ^∘C and pressure of 991.5 hPa with 12.3 % relative humidity as averages, while an activated carbon denuder (6 mesh, Ecotech Pty Ltd., Australia) as well as a Drierite filter (8 mesh, Thermo Fisher Scientific, USA) were installed upstream of the bubbler to remove possible contaminants in the ZA cylinder. An averaged H₂O spectrum from 15 min injection was scaled with the literature spectrum from the HITRAN2020 database (Gordon et al., 2022) in the range of 659.28–671.94 nm and the humidity transmitter (HMT337, Vaisala, Finland; uncertainty: 1 %) data.

Due to the weak but complex absorption features of NO₂ in the fitting range, we measured the absorption spectrum produced from 3 min injections of NO₂ standard gas (10 ppmv in N₂, uncertainty 1 %, KRISS) passing through a trap immersed in dry ice to ensure H₂O-free conditions. The acquired spectrum was scaled to the literature one from Bogumil et al. (2003) for the range of 649.64–672.79 nm, where the most apparent NO₂ absorption features exist with minimum fitting error.

Although the application of the measured spectra improves the fitting performance in general, the uncertainties of ${\mathit{\sigma }}_{{\mathrm{H}}_{\mathrm{2}}\mathrm{O}}$ and ${\mathit{\sigma }}_{{\mathrm{NO}}_{\mathrm{2}}}$ inevitably increased (≥20.1 % and 3.3 %, respectively) compared to the literature's (≥20 % and 3.2 %), owing to the additional errors caused by the fitting procedure (1.3 % and 0.036 %) and number density calculation (0.86 % for both). Note that we only provide the lower limit error in ${\mathit{\sigma }}_{{\mathrm{H}}_{\mathrm{2}}\mathrm{O}}$ propagated from the absorption line-by-line uncertainties in HITRAN2020 (≥20 %).

3.2.2 Spectral fitting

The Levenberg–Marquardt least-squares fitting software, DOASIS (Kraus, 2006), was applied for the spectral fitting of the extinction spectrum between 659.28 and 671.94 nm. A fourth-order polynomial was applied to account for the optical drift and/or unaccounted extinctions such as absorption by ambient ozone. Fit order was selected based on the resulted fitting statistics (i.e., fit coefficient uncertainties, root mean square of residuals, and chi square of residuals; Fig. S1 in the Supplement), which need to be verified for different measurement applications. The reference spectra were allowed to be shifted within ±1.0 nm and squeezed freely for the ${\mathit{\sigma }}_{{\mathrm{NO}}_{\mathrm{3}}}$, but the measured ones were set to share the degree of horizontal shift and squeeze. Figure 4 shows an example of simultaneous retrievals of 5.45 pptv NO₃, 5.75 ppbv NO₂, and 5620 ppmv H₂O with polynomial and fit residual from the ambient measurement during the Arctic shipborne mission (acquired on the open ocean on 26 August 2021, 17:11:41 UTC, described in Sect. 4) with 2 s integration time.

Figure 4 Spectral fitting example (2 s average) of ambient data; measured (red) and fitted (black) (a) total extinction, (b) NO₃, (c) NO₂, (d) H₂O, and (e) polynomial (green) with residual (cyan), respectively. Data were acquired on 26 August 2021 (UTC) from shipborne measurement on the open ocean in the Arctic region.

3.3 NO₃ wall loss evaluation

Even though CEAS is able to measure the target species without concentration calibration using chemical standards in regular operation, indeed, the loss of NO₃ along the airway needs to be evaluated. Many previous works carefully characterized transmission efficiency of NO₃ for their instruments (Aldener et al., 2006; Dubé et al., 2006; Fuchs et al., 2008; Schuster et al., 2009; Kennedy et al., 2011; Wagner et al., 2011; Hu et al., 2014; Wang et al., 2015; Sobanski et al., 2016; Wang et al., 2017; Li et al., 2018). In line with this, we conducted a series of NO₃ injection experiments with a custom-built NO₃ generation system.

3.3.1 NO₃ generation

NO₃ was delivered from the synthesized N₂O₅ crystal under atmospheric pressure with the home-built system (Fig. 5). To produce gaseous N₂O₅ via Reactions (R1) and (R2), 5 % to 10 % of O₃ was generated by corona discharge (Nano 15, Absolute Systems Inc., Canada) of O₂ (300 sccm of 99.999 % O₂ in N₂, Daedeok Gas Co. Ltd.). NO₂ (2 % in N₂, AirKorea Co. Ltd.) was added in two different positions (500 and 200 sccm, respectively) into a quartz reactor (5 cm i.d. and 50 m length) for efficient production of N₂O₅ crystals. To minimize HNO₃ formation in the reactor, any H₂O which can be present in the gas supplies as well as on all surfaces in the generation system was removed by heating the reactor (up to 120 ^∘C) before the injection and by applying dry-ice traps in front of the reactor during the synthesis. After the synthesis, additional O₃ was introduced on white N₂O₅ crystals to flush out the remaining HNO₃ for at least 30 min. The crystals were used immediately or trapped with dry ice and stored at −78 ^∘C for later use.

Figure 5 Schematic diagram of NO₃ generation system with flow paths (arrows). Dry-ice cold traps for H₂O removal (Tr) and for sample collection (Tr^*) are also shown.

Amounts of NO₃ below parts per billion by volume to a few tens of parts per billion by volume were produced by thermal equilibrium, with the sublimated N₂O₅ introduced with a small flow of dry ZA (15–40 sccm) as a carrier gas passing the trap. Unlike previous studies (Fuchs et al., 2008; Kennedy et al., 2011; Odame-Ankrah and Osthoff, 2011; Wang et al., 2015; Wu et al., 2020), we did not provide any additional heat to shift the equilibrium towards NO₃ because the amount of NO₃ through this method was large enough to cover the ranges of typical atmospheric NO₃ mixing ratios in urban night conditions.

3.3.2 NO₃ transmission efficiency, $T_{{\mathrm{NO}}_{\mathrm{3}}}$

NO₃ losses ($T_{{\mathrm{NO}}_{\mathrm{3}}})$ of individual parts along the airway before the detection region (i.e., overflow inlet, filter, and cavity cell) were quantified via continuous injection (at least 5 min) of the synthesized NO₃ under dark condition. Differences in NO₃ concentrations with and without each part of the flow system were acquired by periodic switching between two conditions under the same residence time using a three-way solenoid valve. In order to determine the effect of NO₃ loss on the aerosol-accumulated filter, filters with total ambient suspended particle loadings of 2480, 4075, and 12 308 µg were compared with a clean one.

Figure 6 presents the results of the $T_{{\mathrm{NO}}_{\mathrm{3}}}$ experiments for each of the test components with relative NO₃ concentration changes (concentrations in each step are normalized by the maximum value for every experiment resulting in ranges from 745 pptv to 169 ppbv) since a slow but steady increase in NO₃ concentrations was observed for all the experiments. We presumed that this drift may have been mainly due to the changes in temperature in dry-ice bath where N₂O₅ crystals were placed and/or the variations in contact of ZA with the crystal surface. The comparisons of the retrieved NO₃ with and without each test part were achieved after reflecting the changes in slow increase by linear interpolations for the concentrations acquired in the same condition (shown as light red and gray dashed lines in Fig. 6).

Figure 6 Relative NO₃ concentrations with (black) or without (red) test sections in (a) overflow inlet (averages as circle and 1σ as error bars), (b) cavity, and (c) clean and (d) used filter (aerosol loading of 12 308 µg). Light red and gray dashed lines represent concentration drifts inferred from linear interpolation of each condition.

The $T_{{\mathrm{NO}}_{\mathrm{3}}}$ values of individual parts are 98.9 (±1.9, 1σ), 88.1 (± 2.6), and 93.1 (± 0.3) % for coaxial overflow inlet (residence time <0.03 s), cavity cell (residence time <2.56 s), and filter assembly (clean case), respectively. The largest loss was observed within the cavity cell due to its relatively large surface area and long residence time compared to the other parts of the flow system. The quantified NO₃ loss on a clean filter surface (93.1±0.3 %) was similar to previous studies (85 (±10, 1σ) % on Aldener et al. (2006); 93 (±2) % on Dubé et al. (2006); 95 (±2) % on Fuchs et al. (2008); 84.8 (±10) % on Schuster et al. (2009); 92 (±3) % on Wang et al., 2015). Interestingly, the used filters showed no significant differences compared to the clean one regardless of the ambient aerosol loadings within the experimental range (93.1 (±0.1, 1σ), 92.9 (±0.1), and 92.7 (±2.0) % for 2480, 4075, and 12308 µg, respectively) which agree with Fuchs et al. (2008) and Zhou et al. (2018). From the results, total $T_{{\mathrm{NO}}_{\mathrm{3}}}$ was quantified to be 81.2 % (±2.9, 1σ) under 2.5 slpm sampling conditions.

3.4 Linearity tests

To test the linearity in signal response against the concentrations of species, standard injection experiments were performed. Multiple mixing ratios were achieved by regulating the degree of dilutions in synthesized N₂O₅ crystals, NO₂ standard (5 ppmv in N₂, KRISS; uncertainty: 1 %), and deionized water from the bubbler. As described in Sect. 3.3.2, a slow and steady increase in NO₃ was observed, varying from 746 to 1045 pptv under the constant dilution ratio of 1:150 throughout the experiment (Fig. S2). For tracking this drift in NO₃ standard, we alternated various dilution conditions with the base one (dilution ratio set as 1:150, black markers in Fig. 7a, b) and applied the linear interpolation of retrieved NO₃ concentrations in those conditions (gray dotted line in Fig. S2a). This baseline which depicts the changes in NO₃ source drift was subtracted from the data for both constant (black) and different (red) dilution conditions.

Figure 7 NO₃ (a, b), NO₂ (c, d), and H₂O (e, f) mixing ratios with elapsed times and other independent abundance evaluation parameters in standard addition experiments. For NO₃, data with constant (black) and varying (red) dilution conditions were corrected for the steady drift in NO₃ source by linear interpolation (Fig. S2). In (b), (d), and (f) axes of the ellipses represent 1σ variability for 2 s integration data (vertical) and uncertainty of each variable (horizontal), while dashed lines show linear correlations.

Figure 7 shows the results of standard additions with respect to the elapsed time (left) and other independent concentration evaluation parameters (right; i.e., dilution ratio for NO₃, nominal concentration for NO₂ and independent RH measurement for H₂O). Dashed lines in Fig. 7b, d, and f represent the correlations considering their errors (uncertainty of the parameter on x axis and 1σ variabilities in measurements on y axis). The dilution ratio of NO₃ is calculated from the flow rate of ZA passed over the source divided by the total flow rate and NO₂ standard concentration is from the nominal concentration on the manufacturer's specification with the dilution ratio. RH was measured at the inlet tip by the humidity transmitter (HMT337, Vaisala, Finland) with the measurement uncertainty of 1 %.

All species show good linearities (R² of 0.9918, 0.9985, and 0.9980 for NO₃, NO₂, and H₂O, respectively) indicating the feasibility of atmospheric applications to those species. For NO₂, the intercept of 0.12 ppbv is insignificant considering the limit of detection (which will be discussed in Sect. 3.5). However, the intercept of −143 pptv in NO₃ is larger than the observed precision of the instrument (Sect. 3.5). This is likely due to the variabilities in our NO₃ source and/or the variations in offsets of slow drift correction since the source has the minimum flow rate requirements to operate. In addition, the negligible retrieved NO₃ (0.077±1.46 pptv, average and 1σ for 1s integration data) during the N₂ injection experiment, described in Sect. 3.5, can be used as an alternative to evaluate the zero offsets.

For H₂O, 84 ppmv of the intercept was found, which is in a similar order of magnitude as with its precision (35 ppmv for 1 s integration time). This may be attributable to not only the random noise in detection but also the zero offset in the humidity transmitter (uncertainty: 1 %). From this highly linear response of our instrument in H₂O measurements in varying atmospherically relevant ranges, we would like to emphasize that the difficulties in retrieval in NO₃ measurement due to H₂O can be alleviated by simultaneous measurement with an instrument-specific absorption spectrum of H₂O without any pre-treatment to remove it.

3.5 Precision and accuracy

The Allan deviation method is often used to determine the instrumental precision and the optimal integration time (Allan, 1966; Werle et al., 1993). Minimum detectable extinction for each pixel of the CCD was extracted from 1 h injection of UHP N₂ with 1 s integration time. Figure 8a shows the time series of the light extinction (${\mathit{\alpha }}_{{\mathrm{N}}_{\mathrm{2}}})$ at 662 nm (corresponding to the 1024th pixel) where NO₃ absorbance peaks. ${\mathit{\alpha }}_{{\mathrm{N}}_{\mathrm{2}}}$ shows no significant time-dependent changes and it deviates around zero, which is likely dominated by white noise. Figure 8b shows the Allan deviation for the single pixel corresponding to 662 nm. Up to around 900 s, 1σ precision generally follows the statistical limit which implies that there are no significant integration time-dependent systematic errors up to 900 s, but it starts to gradually diverge after that, which is likely due to the instrumental drifts such as changes in conditions of the light source and/or CCD.

Figure 8 (a) Time series of light extinction for 1 h of N₂ injection (1 s integration) and its (b) Allan deviations (1σ) for the single pixel at 662 nm. Yellow circles represent deviations for 1, 10, 60, and 900 s, respectively, and the dashed line indicates the theoretical limit representing white noise (slope: −0.5).

By only using the corresponding absorption cross-section on that single pixel, the detection limits for NO₃, NO₂, and H₂O are determined to be 1.41 (0.15) pptv, 6.92 (0.73) ppbv, and 35.0 (3.69) ppmv, respectively, for 1 (60) s averaging under 1 atm and 25 ^∘C conditions, by following Fouqueau et al. (2020). However, one should note that the spectral retrievals through the optimized fitting algorithm are likely to produce even lower detection limits than those from the single pixel because this method relies on the absorption features on broad wavelength ranges among hundreds of pixels.

Measurement uncertainties (1σ) for NO₃, NO₂, and H₂O are calculated to be 10.8, 5.2, and ≥20.5 %, respectively, by Gaussian propagation of the errors in the absorption cross-section (NO₃: 10 %; NO₂: 3.3 %; and H₂O: ≥20.1 %), effective cavity length (3.4 %), and HR mirror reflectivity (2.2 %). Note that the fitting errors are not included here because the mathematical error varies with target species abundance. If there are strong signatures of target species in the measurement, fitting errors are negligible. However, for the extremely low-abundance condition, the absorption features of target species weaken, and thus the fit result likely ends up producing a physically meaningless number with large error because of the limitations in numerical fitting algorithms. Hence, when the fitting error outweighs the abundance of target species, interpretations should be limited.

Table 1 summarized the cavity characteristics and performances of the existing BBCEASs for NO₃ measurement. Inferred detection limits with the same integration time are available in Table S1. Our instrument has the longest effective light path length even though the mirror displacement is relatively short. Since this system is able to observe the optical extinctions in the order of 10⁻¹⁰ cm⁻¹ within 1 s time resolution, we can conclude that our instrument is adequate for measuring ambient NO₃ abundances in terms of the sensitivity aspect. The capabilities of operation and utilization for actual application are described in the following section.

4 Field deployment

In order to demonstrate the feasibility of the instrument in field measurements, we deployed our system on the Korean ice breaker R/V Araon and operated it from late July to early September in 2021 for the expedition in the Chukchi Sea and the East Siberian Sea of the Arctic Ocean (Fig. 9a). The instrument was housed in a temperature-controlled seatainer placed on the compass deck (29 m above sea level). The inlet was installed on the window and covered by the weatherproof-designed stainless steel pipe (7.5 cm o.d.). To minimize the loss of NO₃ along the sampling line, air was subsampled from the center of the main flow (1 in. o.d., 1 mm thickness, PFA tubing, 20 slpm). The profile of the main flow was maintained to be steady and laminar (Reynolds number ≅1230) by the blower (DB-200, Manseung Electric Co., Korea). Total length and residence time inside the main flow was kept to be as short as possible to minimize the loss. However, due to the physical limitation of the instrument placement in the seatainer, the length of subsampled PFA tubes was elongated (length: <1 m; residence time: <1.5 s) and the total transmission efficiency of NO₃ for this deployment was 65.1 % (±2.14, 1σ), quantified by post-campaign experiments through the same method as described in Sect. 3.3.

Figure 9 (a) A map with cruise track of R/V Araon and (b–d) representative time series of 1 min averaged H₂O (measured by our instrument on compass deck in blue and a Vaisala instrument on the mainmast in pink), radiation, O₃, NO₂, and NO₃ with a calculated production rate of NO₃ for 23 August 2021 at 17:00 UTC to 25 August 2021 at 15:10 UTC on the open ocean in the Arctic region.

Aerosol filters were replaced by an integrated auto-filter changer (Dubé et al., 2006) only during the early and later stage of the mission near the coastal region of northeast Asia for every 3 h and manually changed with intervals of 4 to 6 h in remote regions since the demands for changing the filter were scarce due to the low aerosol loading in the Arctic region. The R(λ) and I_out,0(λ) were checked every 2 h and instantaneously interpolated for real-time α(λ) calculation. During the campaign, R(λ=662 nm) were varying in the range from 99.9985 % to 99.9989 %, which are lower than the best performances of the instrument in the laboratory, mainly due to the difference in proficiency and environment for cleaning optics, but still high enough for ambient monitoring of NO₃. The negligible change in R is direct evidence that our vibration-resistant design is robust despite the strong vibrations in the platform due to the sea ice breaking activities. During the mission, the averages (±1σ) of temperature and pressure of the sample in the cavity were 22.02 (±0.90) ^∘C and 997.6(±25.9) hPa. And the changes in absorption cross-sections due to these variations were too small to be detected by our instrument.

Figure 9b–d show the time series of H₂O, NO₂, and NO₃ (1 min averaged), as well as O₃ monitored by the UV absorption instrument (49i, Thermo Fisher Scientific, USA), radiance (CNR4, Kipp & Zonen, Netherland) and production rate of NO₃, P(NO₃), calculated as Eq. (3). For H₂O, our measurements were compared to the calculated values from the pressure (PTB110, Vaisala, Finland), relative humidity, and temperature (HMP155, Vaisala, Finland) data measured on the mainmast of the icebreaker. Here we only show the selected period (23 August 2021 at 17:00 UTC to 25 August 2021 at 15:10 UTC) when the NO₃ signals were continuously observed well above the detection limit. H₂O mixing ratios measured by our instrument ranged from 4160 to 6510 ppmv (mean of 5580 ppmv) and average (maximum) value(s) of NO₂ and NO₃ were 3.21 (23.9) ppbv and 2.53 (9.51) pptv, respectively. H₂O concentrations measured by both instruments were in good agreement considering the uncertainty of H₂O for our instrument (≥20.5 %).

$$\begin{array}{}\text{(3)}& P\left({\mathrm{NO}}_{\mathrm{3}}\right)=k_{\mathrm{R}\mathrm{1}}\left[{\mathrm{NO}}_{\mathrm{2}}\right]\left[{\mathrm{O}}_{\mathrm{3}}\right]\end{array}$$

During the campaign, several fresh emissions from R/V Araon were observed, represented by sharp changes in O₃ and NO₂ as shown in Fig. 9c indicating that O₃ was titrated by NO and formed NO₂. These exhaust emissions lasted from a few minutes to several hours depending on the atmospheric conditions such as wind direction and atmospheric stability as well as movement of the vessel (i.e., sailing or anchoring in one place for marine survey).

P(NO₃) was generally small but rose up to 0.66 ppbv h⁻¹ (mean of 0.21 ppbv h⁻¹) depending on whether the sampled air masses were directly influenced by the ship plumes or not but were still lower than those observed in previous works during summertime: 0.31 ppbv h⁻¹ in rural areas (Geyer et al., 2001a) and 1.10–3.2 ppbv h⁻¹ in urban areas (Wang et al., 2013; Brown et al., 2017; Zhou et al., 2018). Within the plume condition, maximum NO₃ increased up to 1.85 pptv even though there was sunlight at that time. The trend of NO₃ concentration was well-matched with P(NO₃) for most of the period indicating the suitability of our instrument for ambient NO₃ measurements. Further analysis related to regional impacts of ship plume chemistry together with NO₃ oxidation assessment with observed VOCs and DMS would be interesting topics for future studies.

5 Conclusions

This paper describes our newly built cavity-enhanced absorption spectrometer for simultaneous measurements of ambient NO₃, NO₂, and H₂O from their absorption features in 659.28–671.94 nm. High performances in measurement capabilities and simplicity in maintaining and processing schemes were achieved by applying high-reflection mirrors (up to 99.9995 % at 662 nm), by integrating the mirror purge and cage system as well as by simultaneous quantification of H₂O using its measured spectrum. Generally, the light at 662 nm travels more than 40 km (up to 101.5 km) within the compact cavity cell (51.5 cm), which enables sensitive measurements of the target species. To overcome the difficulties in H₂O treatment for accurate NO₃ measurement, the measured absorption spectrum of H₂O was used and our instrument showed high linearity for varying atmospherically relevant ranges of H₂O. The transmission efficiency of NO₃ from the inlet tip to the detection region was evaluated as 81.2 (±2.9, 1σ) % within the residence time of 2.59 s from the prepared NO₃ addition experiments. Consequently, for NO₃, NO₂, and H₂O, the measurement accuracies and detection limit (1σ, 1 s, estimated from a single-pixel CCD) were determined as 10.8 %, 4.7 %, and ≥20.5 % with 1.41 pptv, 6.92 ppbv, and 35.0 ppmv, respectively, which are sufficiently low for ambient applications.

The instrument was successfully deployed aboard the Korean ice breaker R/V Araon and captured not only the background condition of the atmosphere over the open ocean in the Arctic but also the highly structured features of the plumes which originated from the vessel exhaust during the campaign. In addition, the trend of NO₃ concentration was well-matched with the calculated P(NO₃), which serves as a proof of the potential for active applications of this instrument in further studies not only in urban regions but also in pristine regions with any mobile platforms including aircraft and research vessels.