Single-photon laser-induced fluorescence detection of nitric oxide at sub-parts per trillion mixing ratios

We describe a newly developed single-photon laser-induced fluorescence sensor for measurements of nitric oxide (NO) in the atmosphere. Rapid tuning of a narrow-band laser on and off of a rotationally resolved NO spectral feature and detection of the red-shifted fluorescence provides for interference-free direct measurements of NO with a detection limit of 1 pptv for 1 second of integration, or 0.3 pptv for 10 seconds of integration. The instrument was deployed on the NASA DC-8 aircraft during the NASA FIREX-AQ experiment during July September of 2019 and provided more than 140 hours of NO 5 measurements over 22 flights, demonstrating the ability of this instrument to operate routinely and autonomously. Comparisons with a seasoned chemiluminescence sensor during FIREX-AQ in a variety of chemical environments provides validation and confidence of the accuracy of this technique.


Introduction
Nitric oxide (NO) is central to radical chemistry in Earth's atmosphere. In the troposphere the catalytic reaction of NO with 10 the hydroperoxy and organic peroxy radicals is frequently the rate-limiting step for the production of tropospheric ozone (O 3 ), and is the reason why anthropogenic emissions of NO result in a buildup of O 3 pollution. Oxidation of NO ultimately also results in the formation of nitric acid, and consequently nitrate aerosols and nitrogen deposition. NO has an important control over the partitioning of atmospheric HO x et al., 2014). Autooxidation of organic compounds in low-NO environments is increasingly recognized as a key source of highly-oxidized/low-volatility organic compounds in the atmosphere (Crounse et al., 2013). Measurements of the ratio of NO 25 to nitrogen dioxide (NO 2 ) in the upper troposphere (UT) frequently cannot be reconciled with models (Cohen et al., 2000;Silvern et al., 2018). Zhao et al. (2019) show that differences in background NO at the single pptv level are responsible for differences in global modeled OH on the order of 50%. The ability to measure atmospheric NO at very low mixing ratios and with low uncertainty will be crucial to address these and other questions in atmospheric chemistry research for the foreseeable future.

30
Almost all of the research in the past two decades associated with direct detection of atmospheric NO has relied on the chemiluminescence (CL) detection technique (Ridley and Howlett, 1974). In this method, a sample of air is mixed with a high concentration (∼ 1%) of O 3 resulting in the formation of electronically excited nitrogen dioxide (NO * 2 ) which produces detectable luminescence in the near-infrared. This technique while being quite precise (∼ 5-10 pptv detection limit), has significant drawbacks. These include: potential positive or negative interferences from other species including a variety of 35 organic compounds (Drummond et al., 1985), precision limits on the order of 5-10 pptv, significant instrumental background levels (10-100 pptv equivalent) which might reduce accuracy, reliance on consumables including pure oxygen and dry ice, production of %-levels of ozone which is toxic and must be exhausted from the instrument.
An alternative technique which has been explored previously is laser-induced fluorescence. Due to the NO absorption cross sections in the deep ultraviolet region of greater than 10 −16 cm 2 molecule −1 and moderately high fluorescence quantum yield 40 (∼ 10%), it should be expected that very precise NO measurements could be made using such a technique. Atmospheric measurements using both single-photon and two-photon excitation schemes were demonstrated by Bradshaw et al. (Bradshaw et al., 1982 and Bloss et al. (Bloss et al., 2003). The single-photon excitation scheme employed by Bradshaw et al. used a dye-based laser system near 226 nm to excite the v = 0 manifold of the A 2 Σ electronic state and observe the red-shifted fluorescence emission from relaxation near 259 nm (v → v = 0 → 3). This system had a reported detection limit of 28 pptv 45 for 1 minute of integration. The two-photon scheme (Bradshaw et al., 1985) involved further excitation of NO that had been pumped into the A 2 Σ state using a second photon at 1.06 µm to promote NO into the D 2 Σ state. The highly excited NO would emit blue shifted fluorescence at 187 nm, which had the advantage of being detected on a near-zero background. The reported detection limits for this technique were 1 pptv given 5 minutes of integration, or 10 pptv for 30 seconds of integration (Bradshaw et al., 1985). To our knowledge, the two-photon scheme developed by Bradshaw et al. is the only NO-LIF measurement that 50 has been utilized extensively for atmospheric measurements including on aircraft, and this technique was last successfully used during the NASA TRACE-P experiment in 2001. Field comparisons of the two-photon excitation scheme with multiple CL instruments demonstrated similar performance for the two techniques (Hoell et al., 1987). Bloss et al. (2003) used a frequency quadrupled Ti:Sapphire laser system to excite NO near 226 nm and detected broadband red-shifted fluorescence at 240-390 nm. Although Bloss et al. did not state a detection limit for their prototype instrument, they calculated that a detection limit of 55 0.07 pptv for 60 s of signal integration might be achievable with improvements in their system. More recently Mitscherling et al. (Mitscherling et al., 2007, 2009) reported investigation of the use of single photon LIF detection of NO for human breath analysis by pumping the A 2 Σ ← X 2 Π transition both near 226 nm (v = 0) and near 215 nm (v = 1).
In this work we report on the recent development of a new single photon LIF sensor that pumps the A 2 Σ(v = 1) ← X 2 Π(v = 0) vibronic transition near 215 nm and observes the resulting red shifted fluorescence from ∼ 255 -267 nm.

60
The present system is distinguished from previous efforts to use LIF to measure atmospheric NO primarily because we use a fiber-amplified laser system. This system has numerous advantages, including: 1) laser linewidth that is sufficiently narrow to resolve the Doppler broadened NO spectrum at room temperature and thereby achieve high signal levels and distinguish the NO isotopologues, 2) laser repitition rate high enough to enable single-photon counting of the fluorescence signal, 3) size, weight and environmental robustness allowing for practical and routine integration onto airborne research platforms. The current version of this system has a detection limit (2σ) of ∼ 1 pptv for 1 second of integration or ∼ 0.3 pptv for 10 seconds of integration. Uncertainty in the instrument zero is demonstrated to be less than 0.2 pptv. The instrument was integrated onto the NASA DC-8 aircraft during the NASA/NOAA FIREX-AQ experiment (Fire Influence on Regional to Global Environments Experiment -Air Quality) during 2019. Here we describe the instrument and its performance during this initial deployment.

Instrument Description
In this section we describe the physical components of the LIF instrument. Subsequent sections discuss details of the NO spectroscopy and instrument performance.
The laser and optical detection system used here are based on that originally described by Rollins et al. (2016) for measurements of sulfur dioxide. Subsequent to that work, important changes have been made to the design of the fiber laser system, 75 and therefore a complete description is provided here.
The laser wavelength is controlled by modulating the current of a distributed feedback laser (DFB). This fiber-coupled DFB laser can provide up to 50 mW CW optical power in a single-mode polarization-maintaining fiber and can be current tuned in the range of 1074 -1076 nm with a nominal linewidth of 10 MHz. The DFB output is chopped to make pulses of 2-3 ns in duration with a 320 kHz repetition rate using a fiber-coupled electro-optic modulator with 10 GHz of switching bandwidth and 80 an extinction ratio of 40 dB. The ∼ 40 pJ pulses are amplified in a multi-stage ytterbium-doped fiber amplifier to near 5 µJ and then exit the fiber-amplifier system through an end-capped fiber and the beam is collimated to a diameter of 350 µm.
The pulses pass through three nonlinear crystals to produce the fifth harmonic near 215 nm with a yield of ∼ 1%. The first crystal is a type-II phase matched KTP crystal (3 x 3 x 10 mm) producing the second harmonic at 537.5 nm. The second crystal is a type-I phase matched LBO crystal (3 x 3 x 10 mm) which mixes the 537.5 nm light with the residual 1075 nm light 85 to produce the third harmonic at 358.3 nm. A custom dual-wavelength waveplate (λ/2 @ 537.5 nm, λ @ 358.3 nm) rotates the residual second harmonic beam to be parallel with the third harmonic beam. A 40 mm focal length lens is then used to slightly focus the beams into a type-I phase matched BBO crystal (3 x 5 x 20 mm) producing the fifth harmonic near 215 nm.
A Pellin-Broca prism separates the harmonics and all of the light is trapped except for the 215 nm beam, which is steered into the LIF cell. 90 Figure 1 depicts the layout of the free-space portion of the optical system and the NO detection system. The 215 nm beam, which is typically near 1 mW, passes through the fluorescence sample cell a single time. The beam is then split about 10/90 with 10% of the power entering a solar-blind power monitoring phototube (Hamamatsu R6800U-01) and 90% passing through a reference fluorescence cell and into a second phototube. The reference cell typically has a constant flow of 500 ppbv NO flowing at 50 sccm and the exhaust of the cell is tied to the exhaust of the sample cell such that both cells are at pressures 95 within 0.5 hPa during measurements. Inside each cell, a fused silica lens with numerical aperture of 0.5 collects fluorescence light from the center of the cell, which then passes through a 260 ± 8 nm bandpass filter (Semrock FF01-260/16) and is then imaged onto the photocathode of a photomultiplier tube (PMT) module operated in single-photon counting mode (Hamamatsu H12386-113). In the reference cell, it is necessary to reduce the signal on the PMT to maintain a response that is linear with changes in laser power. Therefore a neutral density filter with a typical optical depth of 1.0 is placed between the bandpass 100 filter and the PMT.
The sample flow and pressure are controlled by two custom stepper-motor controlled butterfly valves that are designed to minimize pressure drop through the system and potential sampling artifacts (Gao et al., 1999). The inlet valve is machined out of PEEK material (polyether ether ketone) and the exhaust valve out of stainless steel. The inlet valve is servo controlled to a mass flow meter that measures the exhaust of the sample cell. The exhaust valve which is located after the point where 105 the sample and reference cell exhausts are tied together, is servo controlled to a pressure transducer measuring the pressure immediately downstream of the sample cell. During flight operation the pressure and flow are typically maintained to within a 1% range of their setpoints over the entire altitude range encountered. Flow through the reference cell is controlled using a pair of mass flow controllers to mix zero-air and gas from a NO standard gas cylinder.
Data collection and instrument control are performed using a National Instruments compact RIO data system. This system 110 incorporates a field-programmable gate array which is used to control the timing of the laser and photon detection gating with a precision of 5 ns. The instrument that was designed for operation on the NASA DC-8 aircraft houses the detection and reference cells, gas deck and data system in one enclosure (55 x 43 x 21 cm) and the fiber laser components in a second enclosure (43 x 43 x 5 cm). In total, these occupy 26 cm of vertical rack space in a standard instrumentation rack, and weigh 31 kg. Typically, a vacuum scroll pump (Agilent IDP-3) and a small calibration gas bottle are installed in the rack adjacent to the instrument.

115
These components take another 21 cm of vertical rack space and weigh 19 kg. Figure 2 illustrates the relevant NO electronic potential energy surfaces and the LIF scheme used in this work. We pump the A 2 Σ(v = 1) ← X 2 Π(v = 0) transition near 215 nm, and observe the resulting red-shifted fluorescence from the A 2 Σ(v = 1) → X 2 Π(v = 4, 5) transitions.

120
The rovibronic spectrum of NO, especially in the 'gamma bands' (A 2 Σ ← X 2 Π), has been the subject of numerous previous studies (see Mitscherling (2009) and references therein). Figure 3 illustrates the absorption spectrum of NO. Line-by-line resolved spectra shown here have been calculated using the PGOPHER software package (Western, 2017). For the simulation of 14 N 16 O absorption spectra we use the spectroscopic data reported by Danielak et al. (1997) and Murphy et al. (1993). this excitation has multiple advantages over the A 2 Σ(v = 0) ← X 2 Π(v = 0) excitation scheme pumping at 226 nm. First is that the absorption cross section (σ) for NO is about twice as high in the 1 ← 0 transition compared to 0 ← 0. Second, the additional vibrational energy provides a more significant shift in the spectra of the various NO isotopologues, making them more easily distinguished spectroscopically. The origin of the and 70 cm −1 higher in energy than those for the 15 N 16 O and 14 N 18 O isotopologues. Third, 215 nm can be produced using 130 the fifth harmonic of a ytterbium-doped fiber amplifier system operating at 1075 nm, whereas 226 nm cannot currently be produced using such a system. In addition, excitation at 226 nm has the potential to produce spurious signal from fluorescence of SO 2 , while 215 nm is a minimum in the SO 2 absorption cross section, and the SO 2 fluorescence quantum yield here is less than 3% (Hui and Rice, 1973). Figure 4 shows the expected fluorescence emission spectrum based on the Franck-Condon factors (Scheingraber and Vidal,135 1985; Danielak et al., 1997). Excluding the emission at A 2 Σ(v = 1) → X 2 Π(v = 0), which cannot be distinguished from Rayleigh scatter, fluorescence from the A 2 Σ(v = 1) state is expected to peak at v = 4 (255 nm) or v = 5 (267 nm) although it should be possible to collect significant signal from any of v = 3 ∼ 8 (Scheingraber and Vidal, 1985). Multiple detection bandpass filters were tested to optimize the signal:noise for NO detection. The signals obtained with filters collecting the 1 → 4, 1 → 5, and 1 → 6 transitions scaled relative to each other as expected with the Franck-Condon factors for those transitions.

140
Of the filters tested, the one found to produce the lowest detection limit is a filter centered at 260 nm with a full width of 16 nm (see Fig. 4). This has 63% transmission at the 1 → 4 transition (255 nm) and 69% transmission at 1 → 5 (267 nm) while completely rejecting laser Rayleigh and Raman scatter from N 2 and O 2 . While operating the detection cell near 80 hPa, typical background using this filter is 10 counts s −1 with 1 mW laser power. Of this background, about 1 count s −1 is a dark count from the detector. Using a filter to additionally collect the 1 → 6 emission increased the signal by 4.5 counts s −1 mW −1 pptv −1 , but 145 also increased the background to more than 350 counts s −1 mW −1 which would significantly degrade the detection limit. We expect that background levels would only further increase at longer collection wavelengths while collecting the fluorescence from v = 3 at 244 nm would likely increase the signal without substantial increases in the background.

LIF Signal
The anticipated LIF signal (S, counts s −1 ) is proportional to the product of the NO excitation rate E(ν) (s −1 ), the fluorescence 150 quantum yield φ, and the fluorescence collection efficiency of the detection system Ω.
In the optically thin regime, E(ν) can be approximated as the product of the convolution of the molecular absorption cross section (σ(ν) (cm 2 molecule −1 )) with the normalized laser spectral distribution (Λ(ν)), the concentration of NO in the sample cell (n, molecule cm −3 ), the volume within the sample cell that is illuminated and imaged onto the PMT (V , cm 3 ) and the 155 5 https://doi.org/10.5194/amt-2020-24 Preprint. Discussion started: 28 January 2020 c Author(s) 2020. CC BY 4.0 License.
Because the laser used here has a linewidth that is comparable to the Doppler broadened linewidth of NO at 300K (∆ν Doppler = 160 3 GHz) we approximate the convolution of the NO cross section with the laser lineshape as σ(ν)Λ(ν)dν ≈ σ = 1.5 × 10 −16 cm 2 molecule −1 . In reality, the non-negligible laser lineshape in the current system somewhat reduces the effective σ. A typical cell pressure that has been used in the laboratory and can be maintained during aircraft sampling up into the lower stratosphere is 42.5 hPa (1.05 × 10 18 molecules cm −3 ). Therefore, at a NO mixing ratio of 1 pptv the NO concentration in the cell would be n = 1.05 × 10 6 molecules cm −3 . We estimate that a cubic volume with an edge of about 5 mm is imaged onto the PMT.
The fluorescence quantum yield φ is determined by the competition between the natural fluorescence lifetime of NO * and collisional quenching by other molecules in the sample gas. The natural radiative lifetime of A 2 Σ(v = 1) is 200 ns (k r = 5 × 10 6 s −1 ) (Luque and Crosley, 2000). The primary quenchers in the atmosphere are expected to be N 2 , O 2 and 170 Ar. Nee et al. (2004) measured the quenching rate coefficients for NO A 2 Σ(v = 1) by N 2 , O 2 , and Ar to be 6.1 × 10 −13 , 1.48 × 10 −10 , and 3.3 × 10 −13 cm 3 molecules −1 s −1 . Therefore in dry air with 78% N 2 , 21% O 2 and 1% Ar, the quenching rate is k air = 3.2 × 10 −11 s −1 . Carbon dioxide is also a fast quencher of NO * , (k q = 3.8×10 −10 ) (Nee et al., 2004), and 400 ppm of CO 2 would increase the fluorescence quenching rate by 0.4%. For measurements of NO near very large CO 2 sources this additional quenching might need to be considered. Water vapor also has an important and variable effect on φ and this is 175 addressed in a subsequent section. In 42.5 hPa of dry air, we therefore expect the fluorescence quantum yield to be: Similarly, the e-folding lifetime of the fluorescence signal in these conditions is calculated to be 26 ns. By adjusting the photon counting gates in our system in 5 ns increments, we determined the signal lifetime at a range of pressures from 14.7 hPa to 103.5 hPa. A Stern-Volmer analysis of the observed lifetimes concluded that the natural radiative lifetime τ r is 180 ns, and the 180 quenching rate in dry zero air (no CO 2 ) is 3.6 × 10 −11 cm 3 molecules −1 s −1 , overall in good agreement with the literature values of 200 ns and 3.2 × 10 −11 cm 3 molecules −1 s −1 .
The fluorescence collection efficiency is determined by the product of the optical bandpass filter transmission with the geometric collection efficiency and the quantum efficiency of the PMT. The Franck-Condon factor for the A 2 Σ(v = 1) → X 2 Π(v = 4) transition is 0.13 (Danielak et al., 1997) and the bandpass filter transmission at 255 nm is 0.63, while the Franck-

185
Condon factor for A 2 Σ(v = 1) → X 2 Π(v = 5) is also 0.13 (Danielak et al., 1997) 190 Taking the product of the excitation rate with the fluorescence quantum yield and fluorescence collection efficiency, we estimate that the anticipated signal rate is approximately 38 counts s −1 pptv −1 mW −1 . The anticipated signal level is comparable to the fluorescence sensitivity that has been measured for this instrument (11.3 counts s −1 pptv −1 mW −1 see Figure 10). The comparison between the theoretical sensitivity and measured sensitivity is quite reasonable given uncertainties in a number of the parameters discussed above. The absorption cross-section is proportional to the rotational populations in the ground states that are being probed (J = 1.5; 2.5; 3.5) and, therefore, a temperature dependence to the signal is anticipated. Near 300K, populations in all of the probed rotational states will decrease with increasing temperatures. Figure 5 shows the calculated temperature dependence of the Doppler broadened absorption cross-section relative to 300K at 214.880 nm. In this region, a relative decrease in sensitivity of 200 0.34% K −1 is calculated.
If changes in the sample temperature were identical to changes in the temperature of the gas in the reference cell, any sensitivity changes would be exactly accounted for during data reduction. If significant differences arise in the gas temperatures between the sample and reference cells, an artifact would arise. For ground based measurements, the temperatures of the sample cell, reference cell and sample gas flow will usually be very similar. For aircraft measurements where gas from the cold 205 atmosphere (as low as -90K) is rapidly drawn into a warmer analysis region, care must be taken to ensure the probed sample gas is well thermalized with the measurement cell and that the reference cell is close in temperature.
The pressure dependence of the signal arises due to changes in both the excitation rate E(ν) and the fluorescence quantum yield φ. The excitation rate is directly proportional to the NO concentration in the cell and will increase linearly with pressure at a constant NO mixing ratio. In the low pressure limit, φ is independent of pressure, and in the high pressure limit, φ is 210 inversely proportional to the pressure. Thus, at low pressures the LIF signal increases with pressure and eventually the signal becomes independent of pressure. Figure 6 shows the calculated pressure dependence of the LIF signal based on the previously cited fluorescence and quenching rates. We measured the LIF signal between 20 and 100 hPa and show that it generally follows the anticipated pressure dependance. At 100 hPa, we observed ≈ 10% more signal than we do at 30 hPa. Significantly higher pressures reduce the 215 instrumental response time and cannot be maintained on aircraft in the upper troposphere with sufficient flow.
Quenching of NO * due to water vapor is fast and must be considered in humid environments. While a quenching rate coefficient (k H2O ) of the A 2 Σ(v = 1) state has not been reported, Paul et al. (1996) reported for quenching of the A 2 Σ(v = 0) state k H2O = 8.97 × 10 −10 cm 3 molecules −1 s −1 . Based on the very small differences between quenching rates for the A 2 Σ(v = 1) and A 2 Σ(v = 0) states by N 2 and O 2 (Nee et al., 2004), it was expected that quenching by H 2 O of A 2 Σ(v = 1) 220 is reasonably close to the value measured for A 2 Σ(v = 0) and therefore would be important for high humidities.
Quenching by H 2 O will decrease the sensitivity of the instrument to NO by introducing an additional term in the fluorescence quantum efficiency: Defining φ 0 as the fluorescence quantum efficiency at zero H 2 O concentration, it follows that:  s of data acquisition. The observed fluorescence spectrum is plotted on top of a theoretical absorption spectrum that has been calculated using the PGOPHER program at a temperature of 300K. The wide wings of the observed spectral lines are believed 255 to be due to spectral broadening in the fiber laser system, most likely due to self-phase modulation. In the future it should be possible to reduce this broadening which would both increase the online signal somewhat and reduce the online/offline tuning separation required.
Here we use the feature near 214.88 nm as the NO 'online' signal and a minimum in the fluorescence near 214.89 nm as the NO 'offline.' For ambient measurements, we typically tune the laser online for 80 ms, followed by a 20 ms measurement of the 260 offline signal. Figure 9 shows a typical five second segment of ambient measurement data. A small amount of hysteresis when tuning online/offline is sometimes apparent due to the finite impedance of the seed laser diode/driver system. However, these transients are also captured in the reference cell signal and therefore do not contribute to increased uncertainty in the calculated NO mixing ratio.
The reference cell is used primarily to maintain the laser wavelength near the peak of the NO online feature. This is ac-265 complished by continuously walking the laser wavelength around the peak (typically with a period of ∼ 10 s) to maintain a local maximum signal in the reference cell. A fixed differential seed laser current is maintained between the online and offline positions. For data reduction, a running average of the offline signal (counts mW −1 ) is subtracted from the online signal in both the measurement and reference cells. Then, the online/offline difference in the measurement cell is normalized to the online/offline difference in the reference cell. This normalization accounts for the known small changes in instrument sensitivity 270 when walking off the side of the NO peak, as well as any unintentional changes such as pressure fluctuations, or small changes in the laser linewidth.

Linearity
The dynamic range of the instrument is limited by the pulse-pair resolution of the photon counting system (20 ns) and the repetition rate of the laser (320 kHz). At 85 hPa cell pressure, the lifetime of the fluorescence signal is primarily controlled 275 by quenching and is 14 ns. Since the pulse-pair resolution is greater than the signal lifetime the system will at most count one fluorescent photon per laser shot and therefore at very high signal levels the observed count rate will start to deviate from a linear response to the rate at which photons strike the photocathode. However, as discussed previously (Wennberg et al., 1994;Rollins et al., 2016) under the conditions of a known maximum count rate (320 kHz) the observed count rate can be corrected exactly to match the true signal rate and thereby significantly increase the dynamic range without loss of accuracy. At the 280 present typical signal rates (∼ 10 counts s −1 pptv −1 ), errors associated with saturation would not be encountered for NO less than 100 ppbv. Figure 10 shows a typical calibration slope measured during a flight by dynamic dilution of an NO standard into the instrument inlet.
Photolysis of other species to produce NO within the LIF cell by the probe laser could in principle pose an interference. Species 285 known to photolyze at 215 nm producing NO include NO 2 (σ = 5.0 × 10 −19 cm 2 molecule −1 ), HONO (σ = 1.9 × 10 −18 cm 2 molecule −1 ), and ClNO (σ = 1.6 × 10 −17 cm 2 molecule −1 ). Using the estimated photon flux of 4.4 × 10 15 photons s −1 cm −2 we calculate for the species with the largest absorption cross-section (ClNO) the photolysis rate in the probe volume of the LIF cell would be approximately 0.07 s −1 . We estimate that the residence time in this volume is less than 0.01 s and therefore that less than 0.07% of any sampled ClNO could be photolytically converted into NO. Photolysis conversion for HONO and NO 2 290 would be respectively 10-100 times smaller. Such interferences are therefore negligible considering typical concentrations of these other species relative to NO in the atmosphere. A lack of significant photolytic interferences is confirmed by some of the nighttime FIREX-AQ observations when the LIF measurements show NO < 0.1% of the simultaneously measured NO 2 on the aircraft.

295
Calibration is accomplished periodically during operation by adding typically 2-10 sccm of a 5 ppmv NO in N 2 mixture to the instrument sample flow of 2500 sccm. This provides calibration mixing ratios of 4-20 ppbv. Typically, the analytical accuracy of the NIST traceable NO standard is ±1%. The flow controller that delivers that NO to the inlet and the mass flow meter which measures total inlet flow are routinely checked against Bios Drycals with ±1% uncertainty each. Addition of these uncertainties in quadrature suggests that the uncertainty in the mixing ratio of NO delivered to the inlet for a single 300 calibration point is ±2%. For airborne measurements where water vapor mixing ratios may change rapidly and high accuracy water vapor measurements are available to correct for the NO fluorescence quantum yield, the instrument is calibrated in zero air by additionally overflowing the inlet with dry zero air (<10 ppmv water vapor).
During FIREX-AQ, the NO-LIF instrument logged more than 140 hours of airborne operation over 22 flights. Typically, calibrations were performed once per hour during flights. Throughout the mission, a range of calibration coefficients were 305 measured spanning a ±5% range from the mean calibration coefficient, while the precision of each measurement can explain less than ±1% of this variation. These variations in the measured calibration factors were more than was typically observed during laboratory operation, and although the calibration coefficients do not systematically vary with the environmental temperature, it is believed that the instability in sensitivity is related to the temperature of the DC-8 cabin, which varied from roughly 25 -40 • C. This range may be due to apparent sensitivity changes related to temperature effects on the flow/calibration 310 system, or real changes in sensitivity due to perhaps an optical effect (e.g., etalon). The cause for the range in calibration coefficients measured in flight will be a focus of future investigation. For now, we conservatively add in quadrature this ±5% uncertainty to the ±2% uncertainty in the mixing ratio delivered during a calibration to arrive at a ±6% uncertainty in the sensitivity of the instrument in dry air.
For humidity corrections, we need to consider both the uncertainty in the water vapor measurement and in the model used 315 to calculate the correction (i.e., Fig. 7). Typically, water vapor measurements from aircraft are known to ±10%. By applying a 10% uncertainty to the relationship shown in the top panel of Fig. 7 Fig. 7) would add an addi-320 tional 2-3% uncertainty. This uncertainty can likely be reduced in the future by improvements in the model used to reproduce the observed effect of H 2 O on the NO-LIF signal.

Detection Limit
The precision with very low mixing ratios of NO in the system was measured in the laboratory to test for any zero artifacts and to determine the instrumental detection limit. For these tests, the observed data were analyzed assuming that no artifact of any 325 kind exists (i.e., the sampled NO mixing ratio is proportional to the difference between online and offline signals at all mixing ratios), and the instrumental response as determined by additions of NO standards is linear down to zero concentration. Doing this, we typically find that flowing air directly from zero air cylinders (Praxair) into the instrument results in a measurement of 1-2 pptv NO. The observed NO was reduced to less than 0.2 pptv by flowing zero air through a potassium permanganate trap (KMnO 4 ).
330 Figure 11 shows the distribution of 1 Hz NO mixing ratios measured when sampling KMnO 4 scrubbed zero air for 1.4 hours in the laboratory. During this period, the mean NO measured was 0.19 pptv and the noise was normally distributed with a 2σ width of 0.76 pptv. For this period, the laser power was 0.9 mW, and the NO sensitivity was 10 counts s −1 mW −1 . The average count rate was 12 counts s −1 , and thus the calculated background count rate is 10.3 counts s −1 . The width of the observed mixing ratio distribution is what would be expected from a Poisson limited distribution of the photon counts (σ = 335 (12 counts)/(10 counts mW −1 pptv −1 × 0.9 mW) = 0.385 pptv). This suggests that no sources other than photon counting statistics contribute significantly to the precision near the detection limit. The calculated 2σ detection limit for a 1 second integration is therefore 0.97 pptv, and for ten seconds is 0.25 pptv. No evidence exists to suggest that the 0.19 pptv observed in the scrubbed zero air is due to anything other than NO remaining in that sample. Figure 12 shows an Allan deviation analysis of the scrubbed zero-air sampling. During data reduction, a choice must made 340 about what duration to use for averaging of the offline signal. Sufficiently long averaging effectively eliminates the offline signal as a source of noise, while shorter averaging assures that any changes in offline signal are completely resolved. To illustrate this two analyses are shown, one where a 1 hour average is used for the offline and another using a 1 second average offline.
At integration times of less than 100 seconds, the analysis using a 1 hr average of the offline signal shows that the precision is limited only by the counting statistics associated with the online signal. Instabilities in the offline signal with time constant on 345 the order of 100-1000 seconds leads to lower σ using the 1 s offline average for integrations exceeding ∼ 10 minutes.
7 In-situ results FIREX-AQ provided an excellent opportunity for comparing the LIF instrument to a state-of-the-art CL instrument. The CL instrument was located at the front of the cabin, and sampled from a probe on the port side of the aircraft. The LIF instrument was located mid-cabin, with a probe extending from the starboard side of the aircraft. The LIF instrument shared the probe 350 described by Cazorla et al. (2015) with four other instruments, each of which sampled about 2 slpm from the total flow of more than 20 slpm. Table 1 compares key performance and physical characteristics for the LIF and CL instruments.
In Figure 13, we show a time series of the 1 Hz measurements from a 3.4 hour flight on July 24, 2019. This flight shows typical results from the mission. Generally, agreement between the LIF and CL instruments is excellent and we have no evidence of detectable interferences for either instrument. Small differences were sometimes observed when leaving large 355 plumes where the NO mixing ratio would decrease by more than one order of magnitude over the period of one second. These are believed to be due to a volume in the CL instrument sample line, which is designed to match the NO 2 photolysis volume in a paired channel. The lower noise of the LIF instrument is apparent primarily at mixing ratios lower than 10 pptv. Figure 14 shows LGR to demonstrate that once the data are adjusted for the measured water vapor measurements, systematic differences due to differences in water vapor are not apparent. For all data shown in Figure 14, the regression fit slope is 0.993, indicating that the LIF measured NO was on average 0.7% lower than CL -a difference easily attributable to calibration uncertainties for either 365 instrument.

Conclusions
A new instrument has been described for performing direct measurements of atmospheric NO using single-photon laserinduced fluorescence. The demonstrated detection limit for 10 s of integration is 0.3 pptv, and to our knowledge this is the lowest detection limit at this time resolution that has been demonstrated for an airborne atmospheric NO sensor. Besides 370 having excellent precision, the instrument has significant practical advantages as compared to CL instruments. Consumables such as dry ice and pure oxygen are not required. CL instruments have background levels on the order of 10-100 pptv equivalent and the background typically decreases for a number of hours after instrument operation begins. CL background also increases significantly at high altitudes and latitudes due to the effect of cosmic rays on the large infrared-sensitive PMTs. These issues mean that at higher altitudes CL instrumental precision will be degraded. This can clearly be observed in Figure 13 where the 375 CL precision degraded significantly relative to LIF as the DC-8 climbed from 2 km to 10 km altitude. The variable background in CL also means that for accurate measurements on the order of 10 pptv to be made, frequent zero determinations must be performed, and running the instrument for a number of hours before measurements are made is desirable. For this reason, the LIF instrument requires less effort to operate and has the potential to be more accurate at low mixing ratios for typical aircraft experiments where continuous running prior to flights adds an additional experimental burden.

380
The LIF instrument performed well without failure and without a dedicated operator for the 22 science flights during the NASA/NOAA FIREX-AQ mission. Precision in flight was typically not as good as demonstrated in the laboratory, and this was due to reductions in laser power associated with the wide range of cabin air temperatures (∼ 25 -40 • C) experienced on the DC-8. Future improvements in the thermal management of the laser system are expected to improve this issue. In addition, a number of possibilities exist to further improve the signal level. These include: laser linewidth reduction, improved bandpass 385 filter to collect the v = 3 emission and increase the transmission at v = 4 and 5, increased laser power, and increases in the geometric fluorescence collection efficiency. Altogether, we estimate that increases in signal by as much as a factor of 10 are possible.
The one notable additional challenge associated with this LIF technique is that the signal is significantly reduced in the presence of high water vapor mixing ratios. Therefore, a fast and accurate water vapor measurement must be deployed with 390 the NO-LIF instrument to obtain accurate NO measurements in the planetary boundary layer from aircraft. For ground-based operations where water vapor changes much more slowly, it may be acceptable to periodically calibrate the sensitivity in ambient air. For measurements in the upper troposphere and stratosphere where H 2 O mixing ratios are generally less than 1000 ppmv, this effect is negligable. We showed that use of the LGR measurements on the DC-8 allow us to remove any observable difference between the CL and LIF techniques associated with variable water vapor. A potential alternative strategy in the 395 future is to use a mixture of ambient air doped with high mixing ratios of NO in the reference cell, instead of the zero air / NO reference mixture.
The new sensor has the potential to provide high confidence in future measurements of atmospheric NO at mixing ratios of less than 10 pptv which are characteristic of much of the global remote marine boundary layer (Singh et al., 1996;Bradshaw et al., 2000). The technique can be extended to perform measurements of NO 2 using selective photolytic conversion to NO 400 (Pollack et al., 2010), or total reactive nitrogen using catalytic conversion (Kliner et al., 1997;Ryerson et al., 1999). In either case, the LIF instrument could be operated with a significantly reduced flow rate to enable the use of a smaller converter than what is typically required for use with a CL based NO detector. In addition, we have demonstrated the ability to make an isotopologue specific measurement, and have observed in the laboratory signals also from the 15 N 16 O isotopologue. Future