A compact Incoherent Broadband Cavity Enhanced Absorption Spectrometer (IBBCEAS) for trace detection of nitrogen oxides, iodine oxide and glyoxal at sub-ppb levels for field application

We present a compact, affordable and robust instrument based on Incoherent Broadband Cavity Enhanced Absorption Spectroscopy (IBBCEAS) for simultaneous detection of NOx, IO, CHOCHO and O3 in the 400 – 475 nm wavelength region. The instrument relies on the injection of a high-power LED source in a high-finesse cavity (F∼ 36,100), with the transmission signal be detected by a compact spectrometer based on a high-order diffraction grating and a CCD camera. A minimum detectable absorption of 1.8 × 10−10 cm−1 was achieved within ∼ 22 minutes of total acquisition, corresponding to a figure 5 of merit of 7.5 × 10−11 cm−1 Hz−1/2 per spectral element. Due to the multiplexing broadband feature of the setup, multispecies detection can be performed with simultaneous detection of NO2, IO, CHOCHO, and O3 achieving ultimate detection limits of 9, 0.3, 8 ppt and 40 ppb (1σ) within 22 min of measurement, respectively (half of the time spent on the acquisition of the reference spectrum in absence of absorber, and the other half on the absorption spectrum). The implementation on the inlet gas line of a compact ozone generator based on electrolysis of water allows the measurement of NOx (NO + NO2) and 10 therefore an indirect detection of NO with detection limits for NOx and NO of 12 and 21 ppt (1σ), respectively. The device has been designed to fit in a 19”, 3U rack-mount case, weights 15 kg and has a total electrical power consumption < 300 W. The instrument can be employed to address different scientific objectives such as better constraint the oxidative capacity of the atmosphere, study the chemistry of highly reactive species in atmospheric chambers as well as in the field, and looking at the sources of glyoxal in the marine boundary layer to study possible implications on the formation of secondary aerosol particles. 15

. (left) A picture of the instrument mounted on a 19", 3U rack-mount case. (right) Schematic of the instrument. The light from the LED is collimated by lens L1 and injected into the cavity. The exiting light is then collimated with lens L2, and injected into the spectrometer. M1 and M2 are steering mirrors and F is an optical filter. The gas line is composed of a pump, a pressure sensor P, a flow meter FM, and a proportional valve PV. At the inlet, a 3-way 2-position valve in PTFE, V, is used to switch between the sample and zero-air. A manual PFA needle valve MV, is used to fix the flow rate. An ozonizer can be inserted in the inlet line for NOx measurements.
In IBBCEAS a broadband incoherent light source is coupled to a high-finesse optical cavity for trace gas detection. A picture of the spectrometer and a schematic diagram of the setup is shown in Figure 1. In the present study, the broadband light source consisted of a high-power LED Luminus SBT70 allowing ∼ 1 W of optical power to be injected into the resonator. A 85 thermoelectric (TEC) Peltier cooler (ET-161-12-08-E) and a fan/heatsink assembly were used to directly evacuate outside of the instrument up to ∼ 75 W of thermal heat from the LED. A temperature regulator (RKC RF100) with a PT100 thermistor was used to stabilize the LED temperature at ± 0.1°C. The LED spectrum was centered at 445 nm with 19 nm FWHM (Full Width at Half Maximum) which covers the main absorption features of NO 2 , IO, CHOCHO and O 3 . For better collimation of the LED spatially divergent emission (7 mm 2 surface), a dedicated optic (Ledil HEIDI RS) was used and coupled with a 25 mm focal 90 lens (L1, Thorlabs, LA1951-A). The high-finesse optical cavity was formed by two half-inch diameter high reflectivity mirrors (maximum reflectivity at 450 nm 99.990 ± 0.005 %, Layertec, 109281) separated by a 41.7 cm-long PFA tube (14 mm internal diameter, 1 mm thick) hold by an external stainless-steel tube. Both mirrors were pre-aligned and glued with Torr Seal epoxy glue on removable stainless-steel supports which were then screwed on the cavity holders. This enables the easy cleaning for the detection of the highly reactive IO radical. Behind the cavity, a Thorlabs FB450-40 filter was used in order to remove the broadband component of the radiation sitting outside the highly reflective curve of the cavity mirrors. The radiation is focused on an optical fiber (FCRL-7UV100-2-SMA-FC) using a 40 mm focal lens (L2, Thorlabs, LA1422-A). The optical fiber input was composed of 7 cores in a round shape pattern on the collecting side, whereas, at the fiber end, on the spectrometer side, the cores were assembled in a line for better matching the 100 µm slit at the spectrometer. The spectrometer (Avantes, AvaSpec 100 ULS2048L) was composed of a diffraction grating (2,400 lines mm −1 ) and 2,048 pixels charge-coupled device (CCD). The resolution of the spectrometer was 0.54 ± 0.10 nm. All the optics including the cavity were mounted on a Z-shaped 8-mm thick aluminum board fixed on the rack using cylindrical dampers (Paulstra). On the board, four 5 W heating bands and one PT100 sensor were glued, and a second RKC module used to regulate its temperature. The board therefore acts as a large radiator inside the instrument, allowing to minimize internal thermal gradients and thermalize the instrument. Air circulation 105 from outside is ensured by an aperture at the front and a fan placed at the back wall of the instrument (Figure 1). The gas line system was composed of a manual PFA needle valve (MV) and a 3-way 2-position PTFE valve, V (NResearch, 360T032) at the entrance ; while a proportional valve PV (Burkert, 239083), a flowmeter F (Honeywell, HAFUHT0010L4AXT), a pressure sensor P (SLS ATM.ECO) and a diaphragm pump (KNF, N 816 AV.12DC-B) were placed after the cavity. The entire line was made of ¼" PFA tubing which was found to be least lossy for the transport of highly reactive species (Grilli et al., 2012). The 110 pump provided a constant flow that can reach 11 L min −1 at the end of the gas line while a constant pressure in the cavity was obtained by a PID regulator on the proportional valve. A data acquisition card (National Instruments, USB 6000) was interfaced to read the analogue signal from the pressure sensor, while a microcontroller (Arduino Due) drived the proportional valve. The manual valve at the entrance allowed to tune the flow rate. At the inlet, a 3-way 2-position PTFE valve allowed to switch between the gas sample and zero-air mixture for acquiring a reference spectra in the absence of absorption. Zero-air 115 was produced by flowing outdoor air through a filtering system (TEKRAN, 90-25360-00 Analyzer Zero Air Filter). 9 µm particle filters were also placed in the inlet lines (reference and sample) for preventing optical signal degradation due to Mie scattering as well as a degradation of the mirror reflectivity for long term deployment. The air flow was introduced at the center of the cavity and extracted at both ends of the cavity. The optimal cavity design was selected by running SolidWorks flow simulations at flow rates between 0.5 and 1 L min −1 (for more details see supplementary informations -SI). Cavity mirrors The absorption spectrum is calculated as the ratio between the spectrum of the light transmitted through the cavity without a sample, I 0 (λ), and with a sample in the cavity, I(λ). It is expressed as the absorption coefficient (in units of cm −1 ) by the following equation (Fiedler et al., 2003): where R(λ) is the wavelength dependent mirror reflectivity and d the length of the sample inside the cavity. Equation (1) is derived from the Beer-Lambert's law and applied to light in an optical resonator (Ruth et al., 2014). The light transmitted through the optical cavity is attenuated by different processes such as absorption, reflection and scattering of the mirror substrates and coating, as well as losses due to the medium inside the cavity. The losses of the cavity mirrors are assumed to be constant between the acquisition of the reference and the sample spectrum. Mie scattering is minimized with a particle filter 135 in the gas inlet, while Rayleigh scattering losses were calculated to be 2.55 × 10 −7 cm −1 at 445 nm at 25°C and 1 atm (Kovalev and Eichinger, 2004) and thus negligible with respect to the cavity losses normalized by the cavity lengh ( 1−R d = 2.09 × 10 −6 cm −1 ). Therefore, the light transmitted through the cavity is mainly affected by the absorption of the gas species, which leads to well-defined absorption spectral features, α i (λ), that are analyzed in real time by a linear multicomponent fit routine.
Experimental absorption spectra of the species i (i = NO 2 , IO, CHOCHO and O 3 ) have been compared with literature cross 140 section data accounted for the gas concentration, experimental conditions of temperature and pressure, and convoluted with the spectrometer instrumental function. Those experimental spectra are then used as reference spectra for the fit.
A fourth order polynomial function, p(λ) = a 0 + a 1 λ + a 2 λ 2 + a 3 λ 3 + a 4 λ 4 , is added to the absorption coefficient equation (2) to adjust the spectral baseline and account for small changes between the reference and the sample spectra. The transmitted 145 light intensity, as well as the optical absorption path, will be modulated by the shape of the mirror reflectivity curve. Therefore, the later should be defined in order to retrieve the correct absorption spectrum recorded at the cavity output.
4 Calibration, performance and multi-species detection

Calibration
Washenfelder and coworkers (Washenfelder et al., 2008) described a procedure for retrieving the mirror reflectivity curve by 150 taking advantage of a different Rayleigh scattering contribution to the cavity losses while the measuring cell was filled with different bulk gases (eg. Helium versus air or nitrogen). In this work we propose an easier approach consisting of using a trace gas at a known concentration (in this case NO 2 , since O 3 spectrum is less structured, IO is highly reactive and CHOCHO is not easy to produce at a known concentration) and its literature cross-sections (Vandaele et al., 1998) for retrieving the wavelength dependent reflectivity curve. The shape of the reflectivity curve is first approximated with a fourth order polynomial function, (1−R) ) of ∼ 36,100. While the shape of the mirror reflectivity curve is determined once and for all, its offset is slightly adjusted after each mirror cleaning, by flushing in the cavity a known concentration of NO 2 . The spectral emission of the LED centered at 445 nm is well suited also for the detection of IO, CHOCHO and O 3 , which are other key species in atmospheric chemistry. For the field measurements of NO 2 and NO x , two twin instruments named IBBCEAS-NO 2 and IBBCEAS-NO x are deployed, with the later equipped with an ozone generator on the gas inlet line. At this wavelength 165 region water vapor also absorbs and is accounted in the spectral fit analysis. However, the absorption of oxygen dimer is not required in the fit routine since the absorption feature will be present in the reference (I 0 ) as well as in the a absorption (I) spectra. In Figure 2(b) simultaneous detection of species NO 2 , IO and O 3 is reported. Ozone, at 26.5 ppm, was produced by water electrolysis as described in Section 4.4, 175.6 ppb of NO 2 were provided by a permeation tube, and 389.7 ppt of IO were generated by photochemical reaction of sublimated iodine crystals and ozone in the presence of radiation inside the cavity. For

Instrumental inter-comparison and calibration
As standard gas, a NO 2 bottle from Air Liquide (NO 2 in N 2 announced at 1.00 ± 0.05 ppm (2σ)) was used to calibrate 175 the IBBCEAS instruments. To confirm the right amount of NO 2 in the bottle, the later was first calibrate against a CLD instrument (ThermoFisher™, 42iTL trace analyzer calibrated to NIST traceable standards by the manufacturer just before the experiments). The NO 2 concentration in the bottle was measured at 577.4 ± 2.3 ppb. The large discrepancy with respect to the value provided by the manufacturer probably comes from the losses due to the presence of the gas regulator. This gas bottle was then used as local standard for the calibration of the IBBCEAS systems. To confirm the calibration process as well as the 180 stability of the instrument within a greater range of concentrations, two inter-comparisons of the IBBCEAS with two different CLD instruments (ThermoFisher™, 42i NO x analyser and ThermoFisher™, 42iTL NO x trace analyser) were performed in outdoor air over 39 and 12 hours, respectively. Results are reported in Figure 3. The experiments took place at the Institute of Geosciences of the Environment (IGE) in Saint Martin d'Hères, France. The IGE is located in the university campus, ∼ 1 km from the city center of Grenoble and ∼ 300 m from a highway. Ambient air was pumped simultaneously from the same 185 gas line by the instruments at flow rates of 1.0 and 0.5 L min-1 for the IBBCEAS and the CLD instrument (ThermoFisher™, 42i NO x analyser), respectively. The measurements were conducted from 6 pm on Saturday 29 th of September until 9 am on Monday 1 st of October 2018 (Figure 3(a)). On Saturday 29 th of September evening the NO 2 peak occurs at slightly later time than normally expected (from 8 pm to midnight). This may be due to the fact that during Saturday night, urban traffic can be significant until late, but also due to severe weather conditions prevailing at this time, with a storm and lightnings known to be 190 a major natural source of NO x (Atkinson, 2000). For the second experiment shown in Figure 3(b), ambient air was pumped at flow rates of 1.0 and 0.8 L min −1 for the IBBCEAS and the CLD trace instrument (ThermoFisher™, 42iTL NO x trace analyser), respectively. The measurements were conducted from 8 pm on Thursday 18 th of July until 8 am on Friday 19 th of July 2019. Both instruments showed the expected variability from an urban environment with an increase of NO 2 in the evening and morning due to photochemical processes and anthropogenic activities (i.e mainly urban traffic).
1 9 / 0 7 / 1 8 -2 0 : 0 0 1 9 / 0 7 / 1 8 -2 1 : 0 0 1 9 / 0 7 / 1 8 -2 2 : 0 0 1 9 / 0 7 / 1 8 -2 3 : 0 0 1 9 / 0 7 / 1 9 -0 0 : 0 0 1 9 / 0 7 / 1 9 -0 1 : 0 0 1 9 / 0 7 / 1 9 -0 2 : 0 0 1 9 / 0 7 / 1 9 -0 3 : 0 0 1 9 / 0 7 / 1 9 -0 4 : 0 0 1 9 / 0 7 / 1 9 -0 5 : 0 0 1 9 / 0 7 / 1 9 -0 6 : 0 0 1 9 / 0 7 / 1 9 -0 7 : 0 0 The correlation plot, based on data of all instruments, (Figure 4(a)), shows good linearity with a slope of 1.064 ± 0.118 and a correlation coefficient R 2 = 0.960 with measurements averaged over 5 minutes. In order to perform linearity tests, the previous NO 2 bottle from Air Liquide was used and diluted with a zero-air line to produce NO 2 at concentrations of 0, 18.2, 80.8 and 139.7 ppb. Figure 4(b) shows the good linearity of the IBBCEAS instrument with a slope of 0.968 ± 0.019 and a correlation factor of R 2 = 0.996. While the system measures NO 2 directly, the CLD technique applies an indirect measurement of NO x 200 from the oxidation of NO through a catalyzer, then in CLD, the NO 2 mixing ratio is obtained by the subtraction of the NO signal to the total NO x signal. The discrepancies observed at low concentrations (< 5 ppb) between the two techniques maybe due to the fact that the measurement from the CLD could actually corresponds to NO y , leading to an overestimation of the  For the same field time series an Allan-Werle (AW) statistical method on the measured concentrations was employed (Werle et al., 1993). In this case, spectra were averaged in block of ten and analysed by the fit routine. The results of the fit are reported on the top graph of Figure 6. For an acquisition time of 2.5 s, corresponding to 10 averaged spectra, the AW standard deviation σ AW−SD was 230, 6.7, 195 ppt and 800 ppb for NO 2 , IO , CHOCHO and O 3 , respectively. By increasing the integration time,    Long-term stability of the instrument was further studied by taking regular reference spectra within the optimum white noise 240 time of the instrument while continuously flushing the instrument with zero-air mixture. In this case ∼ 5 min and ∼ 3 min intervals were chosen, corresponding to 1000 and 580 averages (for 300 ms integration time) and a precision on the NO 2 concentrations of ∼ 20 and ∼ 23 ppt (1σ), respectively. The results are reported in figure 7. Test 1 (1000 averages) was run for 9h and test 2 (580 averages) for 15h. These tests highlights the reliability of the measurement protocol, with the long term measurements well distributed within the 3σ of the measurement precision (60 and 70 ppt respectively). A Box-plot is also 245 reported representing the average values (green triangles) and medians, quartiles, minimum and maximum values. Table 1 hereafter shows a comparison between our instrument presented in this work and other recently developed IBBCEAS systems. Measuring NO and NO 2 simultaneously is important to study the NO x budget in the atmosphere. In the selected blue visible region, there are no NO absorption features for direct optical measurements, and optical absorption detection of NO is typically done in the infrared region (Richard et al., 2018). However, its detection can be performed by indirectly measuring NO 2 after chemical conversion of NO to NO 2 in a controlled O 3 excess environment. This will lead to the measurement of NO x , which, The experimental results were in good agreement with the simulations as reported in Figure SI-6. In addition, the instrument was found to have a linear response regarding the detection of the produced O 3 . The detection limit for the NO x measurement was found to be similar to the one of NO 2 (12 ppt (1σ) in 22 min of integration time) while for NO, retrieved as the difference 280 between the NO x and the NO 2 concentrations, the detection limit estimated from the error propagation corresponds to 21 ppt.

Possible chemical and spectral interferences
Further possible interferences on NO 2 detection in the presence of high levels of O 3 were also studied, since a large excess of O 3 could trigger the following reactions with rate constants that are few orders of magnitude lower than k 1 (from the NIST Kinetics Database): To study those possible interferences, 100 ppb of NO 2 produced by a permeation tube were pumped through the ozonizer and the spectrometer at a flow rate of 1 L min −1 while varying the concentration of O 3 from 0 to 10 ppm. NO 2 concentration was stable at low ozone concentrations, while a drop of 14 % was observed at high levels of O 3 ( 8 ppm). Kinetics simulations showed that the NO 2 consumption in favor of the NO 3 production (NO 2 + O 3 → NO 3 + O 2 ) was kinetically possible under those conditions. The consumption of NO 2 is strongly dependent on the reaction time and the concentration of O 3 . The later 295 should be selected according to the reaction time imposed by the volume of the inlet line and the flow rate, therefore making this interference negligible. Other chemicals reactions could led to an overestimation of NO 2 mixing ratios: HONO → NO + HO k 5 = 3.9 x 10 −21 cm 3 molec −1 s −1 at 25°C (R5) HO 2 NO 2 → NO 2 + HO 2 k 6 = 1.3 x 10 −20 cm 3 molec −1 s −1 at 25°C (R6) 300 Couach et al. estimated the background levels of HONO and HO 2 NO 2 in Grenoble to be 4 and 2 ppq (or 10 −15 mol mol −1 ), respectively (Couach et al., 2002). With such low concentrations and kinetic constant rates, interferences due to reactions (R5) and (R6) can be neglected in an urban envirionment. However, in remote areas such as the East Antarctic Plateau, HO 2 NO 2 levels were estimated by indirect measurements to be around 25 ppt (Legrand et al., 2014). Because the lifetime of HO 2 NO 2 decreases with temperature (τ HO2NO2 = 8.6 h at -30°C and 645 mbar), its measurement using an instrument stabilized at 305 higher temperature would lead to an overestimation of the NO 2 due to the thermal degradation of the HO 2 NO 2 . However, this interference can be minimized by working at low temperatures : at 10°C and 1 L min −1 flow in our IBBCEAS instrument, the NO 2 signal would be overestimated by only 1 ppt, which is below the detection limit of the sensor.The instruments were therefore designed for working at low temperature (up to few degrees Celsius). Last reaction, (R7), may also lead to possible interferences on the NO 2 detection:

310
HONO + OH → NO 2 + H 2 O k 7 = 4.89 x 10 −12 cm 3 molec −1 s −1 at 25°C (R7) In urban environments and remote regions, one can observed up to 4 x 10 6 OH radicals cm −3 (Heard, 2004;Mauldin et al., 2001). With background levels of HONO such as 4 ppq in the city of Grenoble and around 30 ppt in Dôme C, Antarctica (Legrand et al., 2014), very low mixing ratios of NO 2 (< few ppq) would be produced by (R7) in less than 8 s (residence time of the molecules in the instrument at 1 L min −1 ). Therefore, contribution from this interference can be neglected. Previous 315 works also highlighted possible artifacts through the heterogenous reaction of NO 2 and H 2 O occurring in thin films on surfaces: the approximate rate production of HONO plus NO calculated in their study was reported to be between 4 x 10 −2 and 8 x 10 −2 ppb min −1 per ppm of NO 2 (Finlayson-Pitts et al., 2003). Assuming linearity between production rates and concentrations, this would represent a range of 8 to 16 ppq for 200 ppt of NO 2 in remote area such as the East Antarctic Plateau. The losses that may occur on the thin films on surfaces through the heterogeneous reaction of NO 2 and H 2 O are therefore negligible.

320
Finally, detection of NO 2 , CHOCHO and IO may be affected by spectral interferences. For instance, water vapour also shows an absorption signature at this wavelength region which was included in the fit routine. Its spectral fit is important particularly for the measurement of NO x , where the inlet sampling line gets saturated in water vapor while passing through the water reservoir of the ozone generator. In addition, artifacts on the signal and the spectral fit were studied by varying the O 3 , NO 2 or NO mixing ratios in cavity. Small imperfections of the fit could lead to large effects on the NO 2 retrieved mixing ratio, 325 particularly at sub-ppb concentrations and in presence of large amounts of ozone. However, no appreciable effects of possible artifacts were observed while O 3 concentrations up to 8 ppm were used. These performance studies and the simplicity of the ozone generator, compact and fully controllable, make it suitable for field applications.