Our 2CH-CRDS instrument is installed in an aluminium profile rack (∼ 55 cm × 55 cm × 150 cm), weighs approximately 50 kg and uses two almost identical channels (where the term “channel” means cavity plus associated inlets) to measure NO3 as well as NO3 + N2O5 after thermal dissociation (TD) of N2O5. In order to avoid transfer of mechanical stress from the rack to the cavities, the mirror mounts (see below) were attached to the aluminium rack via rubber buffers (60 Shore A, 189-3264, RS PRO) supporting the upper end of each cavity only. To mitigate the rotational force created by the weight of the “hanging” cavity the lower mirror mount is gently pressed against a rubber covered aluminium plate.

The instrument is illustrated in Fig. 1. Both channels are vertically aligned and accessible from the front of the instrument for adjusting the cavity mirrors. The cavities are insulated with custom-made heating/insulation sleeves (HORST GmbH) that match the T-shape geometry (arm lengths of 2×30 cm and 1×25 cm) of the glass tubing (1/2 in. outer diameter, 9 mm inner diameter). While the shorter arm is used as an entrance, the longer arms define most of the absorption path length. The inner walls are coated with Teflon (FEPD-121, Chemours) to reduce wall losses of NO3 (more details will be given in Sect. 3.6).

With the left-hand cavity, we detect NO₃ at room temperature. In the channel on the right-hand side, the absorption path is maintained at 105 °C (outer glass wall temperature) and samples via a 20 cm long TD inlet held is at 170 °C. All temperature read-and-control units (diraView and Quantrol series, Jumo GmbH) are operated with standard PT1000 sensors (RS PRO) attached to the outside of the glass but within the thermal insulation. While the sensor for the TD inlet is located directly at its center, both cavity temperatures are measured at the midpoint of the length of one of the longer arms. For the heated channel, we use an additional sensor at the T-shape intersection. As the cavities are heavily insulated and about 20 cm apart, there is no significant transfer of heat from the “hot” cavity to the “cold” one.

The rear side of the instrument houses a laboratory power supply, mass flow controllers (MFCs), an automatic filter changer and all electronics required for data acquisition.

The flow of air through the instrument is depicted by black arrows in Fig. 1. Regulated flows through the instrument are exhausted into a vacuum manifold when the instrument is operated in our measurement container or into an external membrane pump (MD 4 T, Vacuubrand) during laboratory operation. Alternatively, the instrument can use a small, in-rack membrane pump (MD 1 VARIO-SP, Vacuubrand). Unless otherwise stated, we use 1/4 in. PFA tubing and PFA fittings (Swagelok) for all connections. Air enters the instrument via an automatic filter changer and PTFE filters (details in Sect. 3.5), which prevent particles from contaminating the instrument. We use the same filter changer device that was presented by Sobanski et al. (2016a) (which operates similar to that described by Dubé et al., 2006) and renewed the FEP coating on the inner walls. The device uses a three-station rotary turret plate: by successive rotation of this plate by 120° a new filter is inserted from a supply stack, then used within the sampled air and finally ejected. During filter changes, the sample air flow is interrupted to prevent contamination.

Directly underneath the filter changer, NO for zero measurements (titration of NO3, Reaction R3) is added via a T-piece. Similar to Schuster et al. (2009), we use two MFCs (IQF-200C, Bronkhorst GmbH) and 1/8 in. PFA tubing to reduce the switching time between ambient and zero measurements. The first MFC continuously flows 7 standard (STP) cubic centimetres per minute (sccm) of NO (200 ppmv (parts per million by volume) in N2, Air Liquide) while the second MFC (set to 9 sccm) is connected to a solenoid valve and an exhaust system. During zero measurements, the solenoid valve is closed and NO is added to the sampled ambient air. Opening the valve directs the NO (plus 2 sccm of sampled air) to the exhaust, allowing NO3 and N2O5 to be measured. This setup results in a switching time of ∼ 2 s (depending on the flow of sampled air) and zero measurements are performed every 3 min for 10–15 s.

Following the NO addition point, the sampled ambient air is split into both channels, each regulated by a MFC (FG-201CV, Bronkhorst) set to 7200 sccm. The major contribution of this total flow is 7000 sccm sampled air, which again splits into two directions, with flows of 3500 sccm towards each cavity mirror. The cavity mirrors are protected against contamination by a 200 sccm “counter” flow of purge gas (light-blue arrows, clean dry air, IQF-200C, Bronkhorst GmbH), which mixes with the sampled air in the “pre-exhaust chamber” and is exhausted together as total flow. For each channel, the cavity pressure (typically 900–950 mbar) is measured using a piezo-resistive manometer (IQP-500C, Bronkhorst GmbH). Operating at these “normal” conditions, the total residence times within the instrument (filter changer exit until exhausted) are 0.43 s for the NO3 channel and 0.34 s for the heated cavity.

The mirror mounting system is illustrated in Fig. 2. Each cavity mirror is installed in a custom-made titanium mirror mount and secured with a silicone O-ring plus retaining ring. Titanium was chosen because of its mechanical properties and low thermal expansion coefficient. The tilt adjustment of each mirror is achieved by three fine adjustment screws (M4AS series, Thorlabs) that apply pressure at the locations highlighted with red dots in Fig. 2a. The use of long arms to connect the inner mirror-securing ring with the outer sealing part means that applying pressure induces bending of the arm and allows the mirror to tilt. The flexural modulus of the material enables the mirror to return to its original position when the pressure is reduced (i.e., the screw is retracted). Different metals such as stainless steel or spring steel and different arm thicknesses can achieve similar results, but poor adjustability (too high rigidity) and irreversible deformation (elastic limit exceeded) are design challenges. To maximize the stiffness of the mirror mount, the arm thickness needs to be as high as possible. Our solution was to use a thickness of 2.5 mm which was milled from a 4 mm solid titanium disk. This design offers an optimal balance between easy adjustment and very stable positioning, which is critical for long-term cavity stability during field campaigns.

Figure 2b shows the mirror mount after its installation according to the schematic in Fig. 1. The mirror (red border) is secured in a separate chamber, which is continuously purged with clean air that enters the pre-exhaust chamber (indigo border, 46 mm inner-diameter) after traversing a 42 mm long, 4 mm inner-diameter metal tube. The diameter of the tube is sufficient to guide the laser beam through, but small enough to generate a high flow velocity of ∼ 13 cms-1 compared to the sample air (3.5 cms-1) within the pre-exhaust chamber that prevents back diffusion. After mixing, ambient and zero-air are exhausted as previously described. In this system, even the planar side of the mirrors is within the purged volume as a window (WG10530-A, Thorlabs) prevents direct contact with ambient air and reduces the impact of changes in ambient pressure on mirror alignment. The window and all metallic components are sealed to the outside using silicone O-rings (magenta circles/lines in Fig. 2b), which are recessed into designated grooves.

Figure 2c shows a photo of the combined mirror mount assembly within the instrument. In the centre, the laser enters through the aforementioned window, while the adjacent fine adjustment screws set the mirror alignment and four wing screws are used to easily disconnect the mirror mount for cleaning. When reinstalling, only very minor adjustments are required to optimize the ring-down signal. At the top, the laser mount and 4 out of 8 of its fine-adjustment screws are shown. While the upper screws are used to adjust the position of the collimation lens relative to the laser diode, the lower ones are used to adjust the position of the laser mount for slight (< 1 mm) off-axis alignment to avoid back reflections into the laser diode, which can perturb the laser emission spectrum.

In order to decrease thermal expansion related effects, each cavity uses three carbon-fibre reinforced polymer (CFRP) tubes (1 m in length, 20 mm outer diameter, 1 mm wall thickness) mounted on custom-made, triangle-shaped, aluminium adapter plates to stabilise the relative positions of the mirrors. This system is almost unaffected by vibrations or shocks and even after strong impacts during container shipment only minor adjustments were required for optimal alignment.

Two laser diodes (HL6545MG mounted in LDM21, Thorlabs) equipped with collimators (C610TME-A, Thorlabs) and optical isolators (IO-3D-660-VLP, Thorlabs) provide light at ∼ 662 nm for the cavities. Even with optical isolators in place, back-reflection from the first cavity mirror into the laser diode resulted in random changes in transmitted laser intensity and/or variations to the emission spectrum. Both caused additional noise to the ring-downs and therefore we avoid perfect on-axis alignment with this instrument. The central emission wavelength is matched to the peak absorption cross-section (λmax⁡≈ 662 nm) of NO3 by heating the laser diodes to 38–40 °C (TTC001 temperature controller, Thorlabs), as shown in Fig. 3. The square-wave modulated laser current (Sect. 2.5) is driven by a control board (MLDEVAL with MLD203CHB, Thorlabs) and results in an average optical power output of 8 mW (at 25 % duty cycle) after the optical isolator.

The laser diode emission is recorded using a compact spectrometer (630–684 nm, HR4000, OceanOptics). Its spectral resolution (∼ 0.06 nm) was determined by calculating the full width at half maximum (FWHM) of seven spectral lines of neon from a low-pressure “Pen-Ray” lamp. We use an optical Y-fibre (BFY50HS02, Thorlabs) in front of the optical isolators to collect a fraction of the light intensity at the edge of the laser. This enables us to monitor the output of both diodes simultaneously while individual emission spectra can be acquired by turning off the other laser. During normal operation, this process takes less than 2 s and is automated using a programmable USB-hub (EX-1596HMVS, EXSYS) which supplies power to the laser current drivers. By regularly recording the laser diode spectra (usually once per hour during field deployment), we can account for changes in the overlap of the laser emission and the NO3 absorption cross-section, discussed in Sect. 3.2.

At the other end of the cavity, transmitted light is detected by a photomultiplier tube (H10492-012, Hamamatsu) screened by an optical bandpass filter (10 nm FWHM at 660 nm, 86–089, Edmund Optics) and located in a custom-designed housing to minimize stray light.

For data acquisition, signal generation and initial visualization, we use a PXIe-1082 embedded computer including a PXIe-8840 Quad-Core processing unit manufactured by National Instruments (NI). The instrument's system control has been programmed in LabVIEW and is based on the “Modular Multiple-Loop Application Framework” described in “The LabVIEW Style Book” (Blume, 2007).

In the same computer chassis, a PXIe-5413 function card is used to generate the modulation signal for the laser current drivers. We use a square-wave function (625 Hz, 25 % duty cycle) at a sample rate of 2 MSs-1. Data acquisition is performed by a PXI-6132 card (with a TB-2709 adapter for SMB connectors) running at the same sampling rate and is triggered by the rising edge of the modulation signal. Additionally, we use an external oscilloscope (TBS1104, Tektronix) to visualize the signals during alignment and operation.

For each recorded ring-down, we reject the initial 5 µs after switching off the laser and use the following 600 µs for fitting. We pre-average 125 individual ring-down traces to smoothen the signal and reduce computational load. Subsequently, we use the “linear regression of the sum” (LRS; Everest and Atkinson, 2008) algorithm to calculate the ring-down constant as used for similar instruments (Sobanski et al., 2016a; Wagner et al., 2011). Afterwards, we use the average of five τ measurements to report 1 s data. We did not see a significant difference in the sensitivity of the instrument for different batch sizes of pre-averages.

For most measurements we generated NO3 and (mainly) N2O5 via the oxidation of NO with O3 (Reactions R1, R2, and R5) in a cylindrical, FEP-coated glass vessel (approx. 3 L volume, 60 cm length). As shown in Fig. S1 in the Supplement, 30–100 sccm of NO (1 ppmv) and 300–400 sccm of clean air (purified with CAP series, different manufacturers) with > 15 ppmv of O3 flowed into one end of the reactor where the mixture resided for more than 6 min in which time all NO was converted to either NO2, NO3 or N2O5. O3 was generated by passing the clean air flow over a partially masked, 2 in. low-pressure mercury “Pen-Ray” lamp (10 mA, 78-2046-2, Jelight). After exiting the reactor, the N2O5 could optionally be thermally dissociated into NO3 (and NO2) by passage through a heated PFA segment (35 cm, 130 °C) directly before being diluted by about 17 slm (standard litres per minute) of clean air. The 17 slm were then sampled by both the 2CH-CRDS instrument and an ozone monitor (Model 205, 2B Technologies). We also used a 1 m3 environmental chamber into which N2O5 was flowed by passing clean air over N2O5 crystals held at about 200 K. This way both N2O5 and NO3 in equilibrium were generated along with NO2. Alternatively, to generate NO3 in the absence of N2O5, the photolysis of ceric ammonium nitrate was used (Lambe et al., 2023). The chamber sources were found to have disadvantages including less stable mixing ratios over long time periods (when using N2O5 crystals) and the formation of high levels of water and HNO3 when using the ceric ammonium nitrate source.

To derive NO3 mixing ratios according to Eq. (1), we need measurements of σ and RL. The effective absorption cross-section σ is calculated by the convolution of the emission spectra for each laser diode and the absorption spectrum of NO3. We use the temperature-dependent cross-sections reported by Orphal et al. (2003) and rescale the value at λmax⁡ to the IUPAC recommendation (IUPAC, 2026), which is 2.25 × 10⁻¹⁷ cm2molec.-1 at 298 K, see Fig. 3. The convoluted effective absorption cross-sections strongly depend on the gas temperature, with values of σ(25 °C) = 2.15 × 10⁻¹⁷ cm2molec.-1 and σ(100 °C) = 1.55 × 10⁻¹⁷ cm2molec.-1. As the laser diodes are temperature-stabilised, the effective cross-sections generally change by less than 5% over a period of several weeks. We estimate the uncertainty in the effective cross-sections to be 5 % and 8 % for the NO3 and (NO3 + N2O5) channel, respectively.

In order to determine RL, we conducted experiments using ozone which absorbs at 662 nm albeit with a cross-section that is a factor ∼ 10 000 smaller than that of NO3 at 298 K and 1 atm so that 10 ppbv (parts per billion by volume) of ozone is detected as ∼ 1 pptv NO3-equivalent. For this experiment, about 7.3 ppmv of O3 was produced by flowing 17 slm of clean air over a 7 in. mercury “Pen-Ray” lamp (10 mA, 78-2046-7, Jelight) and directed to the instrument. RL was then determined by comparing the signal when O3 was (1) present in both the purge gas flow and the cavity flow (i.e. absorption took place over the total mirror distance) or (2) was present only in the cavity flow (i.e. absorption took place over the actual absorption path length). This way RL was determined to be 1.11 ± 0.02 for the NO3 channel and 1.13 ± 0.06 for the (NO3 + N2O5) channel. Both values are very similar to the geometric ratio of approximately 1.12 (= 74 cm/66 cm).

Accurate determination of NO3 mixing ratios requires the complete removal of NO3 during zero measurements. The fractional removal of NO3 depends on the rate coefficient for the reaction between NO and NO3 (k3 = 2.6 × 10⁻¹¹ cm3molec.-1s-1 at 298 K; IUPAC, 2026), the amount of added NO and the reaction time. With typical NO concentrations of about 100 ppbv, more than 99.996 % of the NO3 is removed within the ∼ 0.16 s reaction time available before the flow enters the cavity associated with the unheated channel. For the heated inlet of the (NO3 + N2O5) channel, the NO3 formed by thermal dissociation of N2O5 has less time to react with NO, which partially results from the time taken for N2O5 to dissociate and partially from the larger linear velocity in this part of the tubing. In order to fully dissociate the N2O5 we found that a temperature of 170 °C was required. Setting the temperature in the thermal dissociation inlet to this level also resulted in a homogeneous temperature along the absorption path within the cavity (which was kept at 100 °C). At these NO concentrations and temperatures, we need to correct our NO3 + N2O5 measurements by a factor of 1.01 ± 0.01 in the heated channel. This accounts for a very small fraction of NO3 that is formed “late” from N2O5 dissociation and has insufficient time for reaction with NO during a titration cycle. While this small correction introduces a minor systematic uncertainty, it reduces the impact of cavity wall losses, which would increase for lower flow rates, see Sect. 3.6.

This discrepancy could be decreased by increasing the amount of NO added, but this would subsequently increase the production of NO2 via ambient O3 (Reaction R1) and affects NO3 measurements, since NO2 and O3 absorb (albeit weakly) at 662 nm. With a 662 nm cross-section of ∼ 4 × 10⁻²¹ cm2molec.-1 for NO2 at 294 K (Vandaele et al., 2002), its absorption cross-section is about twice that of O3 at the same wavelength (∼ 2 × 10⁻²¹ cm2molec.-1 for O3 at 293 K; Voigt et al., 2001). Using the rate coefficient for the reaction of NO with O3 of k1(298 K) = 1.9 × 10⁻¹⁴ cm3molec.-1s-1 (IUPAC, 2026) and taking an O3 level of 100 ppbv we calculate that about 2 ppbv of the O3 would have been converted into NO2. Given the relative 662 nm cross sections of NO3, O3 and NO2, this means that the loss of O3 results in a negative offset of about -0.2 pptv NO3-equivalent, whereas the NO2 formed results in a positive offset of +0.4 pptv NO3-equivalent. This results in an overestimation of the zero signal by up to +0.2 pptv for which corrections must be applied to the [NO3] measurements. Even through the correction is small, when deploying the instrument in the field we therefore always make O3 measurements in parallel.

In the heated (NO3 + N2O5) channel, the rate coefficient for the reaction between NO and O3 increases (by a factor of about three) to k1(373 K) = 5.6 × 10⁻¹⁴ cm3molec.-1s-1. Accounting for the shorter residence time and lower air density, about 3.5 ppbv of NO2 are formed which corresponds to an overestimation of up to 0.35 pptv when assuming temperature-independent absorption cross-sections for NO2 and O3 at 662 nm.

Concluding, the major advantages of adding NO during zero measurements are rapid switching times, targeted removal of NO3 and the negligible dilution of the sampled air (7 in 14 000 = 0.05 %). Compared to other methods that sample/use clean air, we prevent changes in ambient concentration and therefore the absorption of other molecules like O3, NO2 and water vapour remains constant. Especially, a correction of [H2O] would introduce a high uncertainty since its absorption spectrum exhibits sharp features at 662 nm (Schuster et al., 2009; Coheur et al., 2002) and its concentration can vary rather quickly during field measurements and greatly between ambient and zero-air measurements.

Like most CRDS instruments, the major factor influencing the LOD is the random variation (white noise) of the ring-down time, which depends on the adjustment of the cavity mirrors and changes of ambient conditions like pressure and temperature. For both channels, the ideal LOD was determined from an Allan Deviation analysis over the respective zero signal. The Allan Deviation is obtained as the square root of the Allan Variance, which is calculated as the averaged squared difference between consecutive pairs of time-averaged (zero) measurements over a given integration time. Here, we use the Overlapping Allan Deviation (oadev by AllanTools.py version 2024.4) to evaluate about 6 h of nighttime data and under ideal laboratory conditions (stable temperatures, sampling particle-free clean air).

Figure 4 shows the 1σ Allan Deviation for both channels for integration times up to 2 h. For short integration times, both channels show the typical decrease related to the declining impact of white noise. Compared to the ideal white noise reduction (thin diagonal lines), the observed decrease is shallower indicating the presence of an additional noise source. With increasing integration time, the Allan Deviation begins to rise again, which is associated with drift effects. This results in a minimum, which can be interpreted as the best integration time for our instrument and is understood as the best achievable LOD of each channel.

The NO3 channel (blue) has a LOD of less than 1 pptv at 1 s integration time. For the (NO3 + N2O5) channel (orange), this increases to about 2 pptv (1 s). The lowest deviation is achieved at integration times of about 3 and 12 min, when the LOD of the NO3 channel is ∼ 0.1 pptv and that of the (NO3 + N2O5) channel < 0.2 pptv, respectively. Based on this, we determined the ideal time between zero measurements to be 3 min to minimize the influence of drifts while still being frequent enough to capture atmospheric variability.

During field measurements, we determine the LOD (for a certain period of time) as the maximum of either 2σ of all zero data points (“white noise”) or 2σ of all absolute differences between consecutive zero measurements (“drift”). This holds for 1 s resolution, but is divided by the n when resampling to a different resolution if the white noise is dominating. Here, n denotes the number of zero data point that are within the resampling interval on average (e.g., n = 8/3 when resampling to 1 min resolution since 8 zero data points are recorded every 3 min). Typical values range from 1–2 pptv for the NO3 channel and 3–5 pptv for the (NO3 + N2O5) channel and are dominated by white noise from zero measurements.

As previously mentioned, CRDS-based measurements of ambient NO3 are usually operated with inlet filters to prevent spurious signals due to particle scattering and to reduce contamination of the cavity mirrors. Several CRDS instruments that measure NO3 have used R2PJ047 PTFE membrane filters by Pall Corporation (2 µm pore size, 25 µm thickness and 47 mm in diameter). These filters have been shown to have good transmission for NO3 and previous studies have reported clean filter transmission factors in the ranges of 86 ± 5 % (Brown et al., 2002b), 90 ± 3 % (Crowley et al., 2010) and 93 ± 2 % (Dubé et al., 2006). As these filters are no longer produced, we have explored alternative products from Merck Millipore (PM2547050, 40 µm thickness) and Cytiva Whatman (7592-104, 40 µm thickness), where the PTFE membrane is supported by a polypropylene ring of similar width. Operating at normal conditions and a sample flow of 14 slm, Pall filters cause a pressure drop of about 10 mbar compared to an empty filter holder; those from Merck and Cytiva show a pressure drop of approximately 20 mbar.

We measured the transmission of NO3 for different batches of unused filters using the NO + O3 flow tube source which produced about 500 ppbv of O3 and more than 100 pptv of NO3 after the dilution. The fractional transmission of NO3 was determined by measuring its mixing ratio after passage through an empty custom-made filter holder (FEP-coated aluminium, wire bail lid system with silicone O-ring seal) or through the same filter holder equipped with a filter. This type of filter holder has a more reliable closing mechanism than commercial PFA inline filter holders with a screw lock. Figure 5 shows the time dependent NO3 transmission through the Pall and Merck filters. The initial NO3 transmission of the filters is variable and ranges from 0.7 to 0.9, which is similar to measurements shown by Brown et al. (2002b) where the initial transmission ranged from 0.6 to 0.8. Continuous exposure to NO3 and O3 increases the filter transmission to its maximum within 5 to 10 min. The same behaviour was observed during the comparison between Pall and Cytiva filters, see Fig. S2. We approximate this exponentially increasing behaviour by Eq. (3) to determine the asymptotic maximum transmission factor Tmax⁡. 3T(t)=Tmax⁡-A×exp⁡(-kt) where t is the exposure time, k is a coefficient that defines the time-constant for “cleaning” the filter and A is used to adjust the amplitude of the exponential contribution. An overview of the comparison of unused filters is presented in Table S1 in the Supplement and shows that all tested filters achieve a maximum NO3 transmission of about 98 % with a 2σ variation of all determined Tmax⁡ values of ± 4 %. Additional details of the procedure and results for different filters are presented in Sect. S1 in the Supplement. Our experiments indicate that initially reactive PTFE filters can be made passive for NO3 by sufficient exposure to NO3 and O3. As the loss of NO3 to filters can be associated with the presence of unsaturated hydrocarbons adsorbed to the filter (Tang et al., 2010), we explored the potential to re-use filters that had become reactive towards NO3 during exposure to ambient air. As O3 is known to react with olefins (Cox et al., 2020), we determined the NO3 transmission of filters that had been exposed to high concentrations of O3 (100 sccm of more than 5 % O3 in O2) for a few minutes. For these experiments, O3 was generated using a commercial, electrical discharge ozone generator (COM-CD-HF4, Anseros) supplied with oxygen 5.0 at 1 bar. Using this method, we were able to fully restore the transmission of previously aged and reactive filters, see Fig. S3.

Our results indicate that fresh filters are more reactive to NO3 than those that have been treated with O3. Therefore, during operation in the field, the stack of filters (held in the automatic filter changer) is flushed continuously by 500 sccm of dry, clean air containing about 15 ppmv of O3 prior to use. Using this setup the NO3 transmission for fresh (or freshly passivated) filters from Pall and Merck is 98 ± 4 % while for untreated PTFE filters a value of 90 ± 5 % is appropriate and consistent with that reported by Brown et al. (2002b) and Crowley et al. (2010). Filters by Cytiva were not tested during field measurements, only in the laboratory.

Note that while such procedures can optimize the initial transmission of the filters for NO3, contamination and loss of transmission during sampling of ambient air will occur at different rates depending on, e.g. particle concentrations and composition. Under polluted conditions, this may imply that filter changes every 30 min are necessary, whereas every 2 h may be adequate for more remote locations. When using the automatic filter changer (see Sect. 2.2), an additional NO3 transmission factor must be considered. This was determined by the ratio of measured [NO3] when passed through the filter changer or through a PFA bypass with negligible NO3 loss (Schuster et al., 2009). Under normal operating conditions (14 slm sampled at about 920 mbar), six switching intervals (each about 5 min of data) were used to determine the average NO3 transmission (80 ± 5 %) through the filter changer. This value was subsequently verified by another measurement with ten intervals (each 6 min) more than a year later.

The fractional loss of NO3 to the cavity walls can be a major correction factor for CRDS instruments and depends on the residence time in the cavity and on the material from which the cavity is made. Common materials are PFA tubing (Wagner et al., 2011; Fuchs et al., 2008) or cavities made from glass or metal that are either coated with a halocarbon wax (Dubé et al., 2006; Brown et al., 2002b) or FEP (Sobanski et al., 2016a; Schuster et al., 2009).

In order to reduce NO3 losses, our glass cavities are coated with FEP and operated at short residence times (i.e. high flow rates), as described in Sect. 2.2. As the loss of NO3 is a first-order process, we quantify the wall loss rate constant (kWall) by varying the residence time in each channel and observing the relative change in the NO3 signal. To avoid changes in either the NO3 concentration or its loss in inlet lines, the total flow (i.e. to both cavities) is kept constant during these experiments and the filter is bypassed. Figure 6 shows the expected exponential decrease in NO3 with increasing residence time. The residence times were calculated from the flow rates and are shorter in the heated inlet/cavity due to the increase in linear velocity at high temperatures. In the heated channel, measurements below flow rates of 5000 sccm were excluded due to increased noise in the ring-down signal, presumably caused by changing the temperature of the mirror mounts away from that experienced during the prior optical alignment. This is reproducible and a similar effect was already observed by Sobanski et al. (2016a).

Each data point in Fig. 6 represents an average of 3 min. The values of kWall obtained are 0.15 ± 0.06 s-1 for the NO3 channel and 0.45 ± 0.20 s-1 for the (NO3 + N2O5) channel. The more rapid loss of NO3 in the hot cavity may be related to increased collision rates of NO3 with the surface. Under normal operating conditions (7 slm each channel), the residence times are 0.43 s in the NO3 channel and 0.34 s in the (NO3 + N2O5) channel, which translate to wall loss factors of 0.94 ± 0.03 and 0.86 ± 0.06, respectively. These values are very similar to other instruments that reported losses in the range of 5 % to 15 % (Dubé et al., 2006; Schuster et al., 2009; Sobanski et al., 2016a; Brown et al., 2002b).

The true ambient NO3 mixing ratios can be derived from Eq. (4) which considers the transmission factors originating from the filter changer (TFC), filters (TFilter) and wall loss of NO3 (TNO3,unheated) for the unheated cavity. The complete overview of all cavity specific values is presented in Table 2. 4[NO3]=[NO3]measTFC⋅TFilter⋅TNO3,unheated Note that [NO3]meas is determined by Eq. (1). To derive the expression for true ambient [N2O5], we start with an expression for [NO3+N2O5]meas which is corrected by a minor factor due to the short reaction time of newly dissociated NO3 with NO (Cdis), see Sect. 3.3. 5Cdis⋅NO3+N2O5meas=TFC⋅TFilter⋅TNO3,heated⋅NO3+TN2O5⋅TNO3,heated*⋅N2O5 Here, we applied the respective transmission factors to [NO3] and [N2O5]. Please note that there is a different NO3 transmission factor for the heated cavity (TNO3,heated) and slightly higher transmission (TNO3,heated∗) for newly formed NO3 after the dissociation of N2O5, which was derived by the corresponding wall loss rate and the residence time purely within the absorption path of the cavity (about 0.24 s). The loss of N2O5 in transmission through filters and inert PFA tubing (TN2O5) has been shown to be negligible (Schuster et al., 2009; Brown et al., 2002b) but is listed for completeness. By rearranging Eq. (5) and inserting Eq. (4), we derive an expression for the true ambient N2O5 which depends on the measurements of both cavities and is similar to the one shown by Fuchs et al. (2008): 6N2O5=Cdis⋅NO3+N2O5meas-TNO3,heatedTNO3,unheated⋅NO3measTN2O5⋅TNO3,heated*

The total measurement uncertainty (TMU) includes contributions from the absorption cross-section of NO3, the ratio between mirror distance and absorption path length as well as the aforementioned NO3 transmission factors. The overview of all uncertainties is listed in Table 2. Based on this, we determine the TMU as propagated relative uncertainty and estimate 9.8 % for the measurements of [NO3] in the unheated cavity. Due to the subtraction within Eq. (5), this method becomes slightly more complicated for N2O5 and can be most easily derived by redefining the individual terms: I.

x≡[N2O5]

We derive the relative uncertainty (written in parenthesis) of y=a-b by: 7Δyy=Δa2+Δb2a-b=a2⋅Δaa2+b2⋅Δbb2a⋅1-ba=Δaa2+b2a2⋅Δbb21-ba=Δaa2+r2⋅Δbb21-r

In the last step, we defined r=ba (with a>0) which represents the ratio between (b) the proportion of NO3 detected in the heated cavity over (a) the corrected sum of NO3 + N2O5. Furthermore, we derive the relative uncertainty of [N2O5] = x=y/c which can be written as: 8Δxx=Δyy2+Δcc2=Δaa2+r2⋅Δbb21-r2+Δcc2

This depends only on the relative uncertainties of a, b and c which depend on the presented (and propagated) uncertainties of this instrument as well as r. A more intuitive understanding of r can be gained by its relation to ambient [NO2] and the equilibrium constant (Keq) that together with the NO2 concentration and the temperature defines the NO3/N2O5 ratio (IUPAC, 2026; Burkholder et al., 2020; Osthoff et al., 2007). Since r is defined by NO3 and N2O5 measured (and corrected) in the heated cavity, we can rewrite: 9r=NO3heatedNO3heated+[N2O5]heated=[NO3]⋅d[NO3]⋅d+[N2O5]⋅c=1+[N2O5][NO3]⋅cd-1=1+Keq⋅[NO2]⋅cd-1

With c and d=TFC⋅TFilter⋅TNO3,heated describing the combined transmission factors of the respective molecules within the heated channel. Concluding, we can estimate the TMU of the heated channel based on an additional ambient [NO2] measurement and well-known constants for the thermal equilibrium as well as the transmission factors of this instrument.

The lowest uncertainty of x = [N2O5] is reported when r→0 (e.g. ambient temperature < 0 °C and [NO2] > 2 ppbv results in r < 3 %) and yields 11.5 % while we estimate about 22.3 % for r=0.5 (i.e. [NO3] = [N2O5]). A graphic of this estimate is shown in Fig. S4 and highlights the case for r→1 (i.e. very low [N2O5]) when the subtraction introduces uncertainties greater than 100 %. Please note that the (atmospheric) variability of [NO3]meas and [NO3+N2O5]meas does not contribute to the TMU estimation but uncertainties of σ and RL for both cavities are included.

Another potential uncertainty is the thermal dissociation of N2O5 to NO3 during transit through the filter changer and cavities, and becomes important when sampling cold air into a warm container/instrument (Schuster et al., 2009). Assuming an equilibrium of NO3 and N2O5 at 1 ppbv of NO2 and no additional loss or production, an immediate increase from 0 to 25 °C causes the dissociation of about 1 % of N2O5 in 0.25 s, which is the residence time from the filter changer entrance to the cavity entrance. This increases to about 2 % until the exit of the cavity (after 0.55 s) and has to be corrected when there is a significant temperature difference compared to ambient conditions.

During the last few decades, several instruments using cavity-enhanced absorption spectroscopy have been developed for ambient measurement of NO3 (and N2O5). Table 2 lists those instruments that have been described in the literature which were designed or used for field-deployment. The table includes instrumental parameters such as LOD (for both NO3 and N2O5) and the footprint, where this information is available.

While some of the first instruments, e.g. Brown et al. (2001), used large pulsed lasers and were mounted on optical tables, the availability of laser diodes has reduced the cost and size requirements. The most commonly deployed modern detection methods are BBCEAS and CRDS, the best of which (Brown et al., 2001, 2002a; Dubé et al., 2006; Wagner et al., 2011; Sobanski et al., 2016a) achieve < 1 pptv limits of detection within 1 s of integration time, as does the present instrument. Note that all of these are CRDS instruments, with BBCEAS-based methods having slightly poorer LODs. The two features of the present instrument that set it aside from most others is the small footprint and ease of mirror-adjustment/cleaning and optical stability.

The vertical alignment of the cavities in our system result in the smallest footprint among those instruments that achieve sub-pptv performance at 1 s integration time. Compared to our present instrument (Sobanski et al., 2016a), and previous versions (Schuster et al., 2009) we have greatly reduced the footprint (factor > 2) and weight (factor 2), which together allow easier transport and enhances the possibility of measurements in confined spaces such as tower-elevators. While it is not possible to compare the “ease of optical adjustment” and optimization of ring-down times with instruments that do not stem from our group, the mirror-tilting mechanism in our latest instrument and its (lack of) response to e.g. mechanical and thermal stress is superior to the use of commercial optical mounts (Sobanski et al., 2016a) as is the ease of mirror removal, cleaning and replacement. These attributes make deployment of the present instrument in confined spaces and non-laboratory environments easier to achieve than our previous setups.