CAFE: a new, improved nonresonant laser-induced fluorescence instrument for airborne in situ measurement of formaldehyde

NASA Compact Airborne Formaldehyde Experiment (CAFE) is a nonresonant laser-induced fluorescence instrument for airborne in situ measurement of formaldehyde (HCHO). The instrument is described here with highlighted improvements from the predecessor instrument, COmpact Formaldehyde FluorescencE Experiment (COFFEE). CAFE uses a 480 mW, 80 kHz laser at 355 nm to excite HCHO and detects the resulting fluorescence in the 420–550 nm range. The fluorescence is acquired at 5 ns resolution for 500 ns and the unique time profile of the HCHO fluorescence provides measurement selectivity. CAFE achieves a 1σ precision of 160 pptv (1 s) and 90 pptv (10 s) for [HCHO]= 0 pptv. The accuracy of CAFE, using its curve-fitting data processing, is estimated as ±20 % of [HCHO]+ 100 pptv. CAFE has successfully flown on multiple aircraft platforms and is particularly well-suited to high-altitude research aircraft or small air quality research aircraft where high sensitivity is required but operator interaction and instrument payload is limited.


Introduction
Formaldehyde (HCHO) in Earth's atmosphere is primarily produced during the oxidation of volatile organic compounds 20 (VOCs) including methane and isoprene, with minor direct emissions from biomass burning and fossil fuel combustion (Fortems-Cheiney et al., 2012). The atmospheric lifetime of HCHO is a few hours against photolysis and reaction with OH.
HCHO is thus sensitive to local variability in both VOC emissions/transport and oxidation capacity. HCHO measurements are consequently valuable for deriving continental-scale isoprene emission fluxes (Palmer et al., 2003), remote-atmosphere maps of column OH , and regional/continental-scale maps of organic aerosol (Liao et al., 2019). Deep 25 convective transport of oxygenated VOCs (OVOCs) such as HCHO (e.g., Fried et al., 2016), or transport of OVOC precursors (e.g., Apel et al., 2012), can account for a significant portion of the upper tropospheric (UT) HOx budget (e.g., Ren et al., 2008) and thereby impact UT O3 levels (e.g., Crawford et al., 1999). HCHO measurements in the UT are necessary to close the HOx budget (Jaeglé et al., 2001) and can also be used in the UT and lower stratosphere (LS) as a qualitative tracer of recent convective activity. Monitoring HCHO is also important for public health: the EPA classifies 30 Atmos. Meas. Tech. Discuss., https://doi.org /10.5194/amt-2019-153 Manuscript under review for journal Atmos. Meas. Tech. Discussion started: 20 May 2019 c Author(s) 2019. CC BY 4.0 License.
HCHO as a hazardous air pollutant (HAP) and it is estimated to pose the highest cancer risk of all the HAP species (Strum and Scheffe, 2016).
A number of measurement techniques have been applied to in situ atmospheric measurements of HCHO and are reviewed elsewhere Hak et al., 2005;Kaiser et al., 2014;Shutter et al., 2019). The most common techniques are 5 absorption spectroscopy (e.g., Richter et al., 2015;Washenfelder et al., 2016), wet chemistry (e.g., Aiello and McLaren, 2009), and laser-induced fluorescence (LIF) (e.g., Cazorla et al., 2015;Hottle et al., 2009). The different techniques offer trade-offs between cost, instrument size, sampling time resolution, and sensitivity. A new approach, non-resonant LIF (NR-LIF), was recently developed to provide a more compact, lower cost instrument with simpler operation than typical HCHO LIF instruments (St. Clair et al., 2017). The NR-LIF approach trades the wavelength-tunable excitation laser traditionally 10 used in HCHO LIF for a fixed wavelength 355 nm laser that excites multiple HCHO absorption features. The broad, fixed wavelength laser cannot tune on and off a single absorption feature to uniquely detect HCHO. Measurement selectivity is provided by carefully selected fluorescence optical filters and rapid data acquisition that enables resolving the time profile of the HCHO fluorescence signal.

15
The NR-LIF HCHO instrument described in St. Clair et al., 2017, Compact Formaldehyde FluorescencE Experiment (COFFEE), was specifically designed for operation on the Alpha Jet stationed at NASA Ames Research Center and is not easily integrated onto other aircraft platforms. The Compact Airborne Formaldehyde Experiment (CAFE) instrument described here uses the same general measurement approach as COFFEE but with updated components and design changes for operation under the demanding conditions of the NASA ER-2 aircraft. The compact instrument design combined with 20 reliable, commercial subsystems makes CAFE well-suited for scientific projects on high altitude aircraft, where high sensitivity is required but size and weight are limited, as well as tropospheric aircraft platforms with limited payload. In this paper we present the instrumental details of CAFE, including modifications from the original COFFEE design that stem from an improved understanding of the HCHO NR-LIF technique.

Instrument description 25
The CAFE instrument is enclosed by a 42 cm x 53 cm x 30 cm (L x W x H) chassis ( Fig. S1) with all plumbing and electrical connections on one side of the instrument. The main subsystems of the instrument are shown in Fig. 1 with the chassis lid removed and are described in more detail in the following sections.

Laser
The light source in CAFE is a Spectra-Physics Explorer One 355-1W (Fig. 1, item A) that provides 480 mW of pulsed 355 30 nm light. The laser is operated at 80 kHz (adjustable 0-300 kHz) to reduce the peak power and minimize risk of mirror damage. Pulses are ~10 ns wide (FWHM) and the laser bandwidth is approximately 1 nm. Laser control is performed by a RIO data acquisition system (Section 2.5) through RS232 communication. Figure 2 illustrates the details of the CAFE optical plate (Fig. 1, item B). The laser beam is directed into the optical cell by three antireflection (AR)-coated dielectric (Acton Optics) steering mirrors (Fig. 2, item A) with a collimating lens (F=100 5 mm) and half wave plate (AR coated, Optisource) in between mirrors 2 and 3 (together Fig. 2, item B). AR-coated fused silica (CVI) cell windows (Fig. 2, item C) provide the vacuum seal for the optical entrance and exit of the optical cell.

Optical system
Optical baffles (Fig. 2, item D) along the entrance and exit 'baffle arms' of the optical cell significantly reduce the stray light reaching the detection volume. Baffle opening diameter increases from entrance to exit over the 7 baffles with three sizes: 2.5 mm, 3 mm, 3.5 mm. The baffles closest to the detection volume are designed in-house and commercially coated (Acktar, 10 vacuum black) while the other 5 baffles are commercially available spatial filter apertures (National Aperture). Cylindrical spacers between the baffles are coated (Acktar, magic black) though background counts are minimally sensitive to the spacer surface for spacers not adjacent to the detection volume. The inner surface of the detection cell is black anodized to reduce stray light. To achieve a more compact optical layout, the ends of the baffle arms are at 90° to the rest of the arm. The 90° turns are provided by two fixed steering mirrors (Fig. 2, item A). After exiting the optical cell, the laser beam passes through 15 a beam sampler (Thorlabs) (Fig. 2, item E) which directs a few percent of the beam to a power meter (Fig. 2, item F) comprised of a ground glass diffuser (Thorlabs DGUV 10-600), an absorption filter (Thorlabs FGUV11), and an amplified photodiode (OSI 555-UV). The majority of the laser beam passes through the beam sampler and into a beam dump.
Sample air flows through the detection cell perpendicular to the laser beam path and detection axes. Aspheric lenses 20 (NA=0.66, AR-coated, Edmund Optics) on opposite sides of the detection cell (Fig. 2, item G) image the intersection of the sample air flow and the laser path to collect the HCHO fluorescence (~355-550 nm). For each detection arm, the collected light is turned 90° by a mirror (right angle prism dielectric, Thorlabs) ( Fig. 2, item H) and directed toward the optical filters The primary motivation for the change in filters from St. Clair et al., 2017 was the realization that H2O Raman signal, at high [H2O], adversely affected data processing as discussed in the Supplement (Fig. S2). Replacing some of the long pass filters with red-shifted alternatives reduces the instrument sensitivity but improves accuracy at higher ambient water vapor. Axis 1, 30 proceeding toward the PMT, uses an AR-coated 370 nm long pass absorption filter (Hoya Candeo Optronics), a 405 nm long pass interference filter (Semrock), and a 420 nm long pass absorption filter (Thorlabs). Axis 2, proceeding toward the PMT, has an AR-coated 370 nm long pass absorption filter (Hoya Candeo Optronics), a custom 11-band band pass interference Atmos. Meas. Tech. Discuss., https://doi.org/10.5194/amt-2019-153 Manuscript under review for journal Atmos. Meas. Tech. Discussion started: 20 May 2019 c Author(s) 2019. CC BY 4.0 License. filter (Semrock), a 405 nm long pass interference filter (Semrock), and a 420 nm long pass absorption filter (Thorlabs). As of April 2018, COFFEE also uses these fluorescence filters.

Air sampling
The details of the inlet and sample tubing external to the instrument are dependent on the aircraft platform for a given 5 experiment. CAFE has used previously existing inlets for various aircraft platforms, e.g. the inlet described in Cazorla et al., 2015 for the NASA DC-8. A new inlet was designed in 2016 for use on the high-altitude NASA ER-2 aircraft. The inlet body consists of a nacelle mounted to the end of an air foil that attaches to the ER-2 superpod midbody via an adapter plate ( Fig. 4). The center of the nacelle is held 10.1 cm from the adapter plate to extend sampling beyond the boundary layer of the superpod. Ram pressure is generated from the tapered shape of the nacelle and allows a higher instrument operating 10 pressure than possible with ambient pressure sampling. Sample air is pulled into the instrument through a heated 6.35 mm OD Silcosteel tube that protrudes into the nacelle perpendicular to the ambient air flow. The tube is chamfered 15° away from the direction of flight to reject particles larger than ~ 1 µm. 28 W of heat is provided by a cartridge heater (SunRod) mounted in a copper block that clamps around the sample tube, and is controlled to 30° C by a proportional controller (Minco CT335) with protection from thermal run-away using a bimetallic hermetic thermostat.

Thermal management
Temperature control, to varying degrees, is necessary for some of the instrument subsystems. Adequate heat removal is required for proper laser performance: two thermoelectric cooler (TEC) elements (TE Technology) provide unidirectional thermal control to cool the laser mounting plate and maintain temperature to £ 303 K. TECs are thermally attached to the optical plate with additional heat dissipation provided by heat sinks mounted to the underside of the optical plate. Air is 5 drawn up through the bottom of the instrument chassis by two fans (22 L s -1 each) that blow directly onto the heat sinks. The fans are controlled by a 30°/35° C off/on controller (NuWave Technologies) to avoid cooling at low temperatures. The pump is a significant heat source and so the four fans that exhaust air out of the chassis (visible in Fig. S1) pull air across the pump to prevent its heat from adversely affecting the laser and other subsystems. The exhaust fans are also controlled to turn off below 30° C. 10 The pump operates best when the ambient temperature is 10-40° C. To maintain pump function in the unheated pods of highaltitude aircraft, 31 W of cartridge heaters are installed on the pump. Cartridge heaters are also used to heat the pressure controller (24 W) and detection cell (24 W) to 30° C. Polyimide foil heaters provide 70 W of heat to the optical plate, also maintained at 30° C. All heaters are controlled by proportional controllers (Minco CT335) and protected from thermal run-15 away using thermostats.

Data acquisition
CAFE utilizes a National Instruments CompactRIO (RIO) data acquisition system (Fig. 1, item E) that runs a real-time operating system. Signal from each PMT is counted with 5 ns resolution by two channels of a NI 9402 high speed digital I/O module, which is triggered by a TTL pulse from the laser. The signals from the PMTs are delayed by 50 ns with a 20 passive delay circuit (Data Delay Devices, 1515 series) so that the PMT signals arrive after the laser TTL pulse. The digital I/O module, along with analog NI 9205 (input) and NI 9264 (output) modules, interfaces with a main processor module and a backplane with a field programmable gate array (FPGA). Data are recorded at 10 Hz for non-gated (continuous) and timegated counts, and at 1 Hz for all other data including instrument diagnostics and time-resolved counts. 10 Hz data are used mainly as an alignment metric. The 1 Hz time-resolved count data consists of a 500 ns time profile of counts, resolved to 25 100 discrete 5 ns bins, immediately following the laser TTL pulse. HCHO mixing ratio data is generated from the 1 Hz time-resolved count data.

Data processing
The data processing approach for CAFE is almost identical to the methods described in St. Clair et al., 2017 for COFFEE with minor changes related to laser pulse timing unique to each instrument. Briefly, data from each detection axis is 30 processed by performing three sequential fits to the 1 Hz, 5 ns binned data (0-500 ns from trigger). The first fit is a single Data can also be processed using delayed-gate counts at 1 Hz or 10 Hz to give higher precision HCHO data (see Sect. 3.2) that is potentially more susceptible to measurement artifacts. Delayed-gate count data sums the PMT counts over the same time window as the exemplar fit (bin 11 to bin 56, 50-280 ns), keeping the background signal low by excluding the prompt signal from scatter. HCHO data in ppbv is generated by subtracting the background counts determined from a preflight zero 15 run and applying the calibration factor. More details on the fits, the 'exemplar' time profiles, and processing with gated counts are available in St. Clair et al., 2017. 3 Performance

Calibration and sensitivity
CAFE calibration is performed in the same manner as for the ISAF and COFFEE instruments. Flow (0-50 standard cm 3 min -20 1 , sccm) from a ~500 ppbv HCHO in N2 gas cylinder is added to a main flow (~4 standard L min -1 , sLm) of ultrahigh purity air to give HCHO typically ~ 0-10 ppbv. The ~4 sLm flow is greater than the instrument sample flow (~2.3 sLm) to speed up the time response of the calibration set-up, with the excess flow exhausting to the room through a 'T' at the instrument chassis. The HCHO mixing ratio is stepped through 3-5 different flows with >45 minutes at each level to provide confidence that the calibration set-up has equilibrated. The mixing ratio of the HCHO calibration cylinders are verified 25 before and after each aircraft mission using IR absorption as described in Cazorla et al., 2015. Uncertainty in the gas standard combines with uncertainty in the calibration system flow measurements to give a calibration uncertainty of ±10%.

Precision and measurement uncertainty
The main source of noise for CAFE HCHO is Raman and Raleigh scattering of the laser beam. Chamber scatter also significantly contributes to the noise and varies in magnitude with the alignment of the laser through the optical cell. Figure   6 shows the standard deviation of HCHO as a function of HCHO mixing ratio for data processed using exemplar fitting (red symbols) and gated counts (blue symbols) from multiple laboratory experiments. At [HCHO] = 0 ppbv, the average standard 5 deviation for the exemplar-fitted data was ±152 pptv for 1 s and ±87 pptv for 10 s data. For gated count data, the standard deviation was lower: ±100 pptv for 1 s and ±37 pptv for 10 s data. These values are somewhat higher than reported in St. Clair et al., 2017 for COFFEE despite CAFE's higher laser power due to the changes in optical filters and the increased delay from the laser pulse to the start of the fit window in CAFE data processing. The precision can be improved by shortening the delay between the laser pulse and the start of the fit window, thereby increasing the amount of fluorescence 10 data used in the fit. With the H2O Raman signal effectively blocked, shortening the delay may now be appropriate but further investigation is necessary. For UT/LS measurements of HCHO, the requisite precision exceeds what is possible using exemplar fitting with the current fluorescence filters. Blocking signal from H2O Raman, however, is not necessary at the low specific humidity of the UT/LS and the water-blocking long pass filters will not be used for high altitude projects, thereby increasing the fluorescence signal. The gated count data also provides an improvement in precision over the 15 exemplar fits, as shown in Fig. 6, and will be available for higher signal-to-noise measurements in the UT/LS. Figure 7 shows the Allan-Werle deviation as a function of averaging time for two-parameter fit data with [HCHO] = 6 ppbv to illustrate how precision improves with increased data averaging. HCHO data precision improves with averaging up to ~200 s, after which additional time averaging has no benefit. A linear fit to the data with t £ 256 s gives a slope (-0.58) 20 similar to the value expected from white noise (0.5), suggesting that white noise dominates on those time scales. The overall HCHO measurement uncertainty is a combination of the calibration uncertainty (±10%) and the less easily quantified uncertainties from unidentified measurement interferences and fit biases. To account for both the known and unknown uncertainties, the overall measurement uncertainty is conservatively estimated to be ± (20% of [HCHO] + 100 pptv).

Time response 25
For airborne in situ observations, the effective spatial resolution is limited both by the data acquisition rate and the time response of the instrument. To understand the time response of CAFE, pulses of 1.3 ppmv HCHO gas were introduced into a high flow of ultrahigh purity air by a rapidly switching valve (The Lee Company, IEP series) and CAFE subsampled this flow. The decay of the 10 Hz HCHO signal after the pulses was fit using an exponential function, as shown in Fig. 8, and the measured instrument time response is 158±10 (1-s) ms. An additional set of measurements was conducted for CAFE 30 with the particle filter in-line with the sample flow to verify that the filter does not adversely affect the instrument time response. With the filter in-line, the e-fold time (161 ms) was nearly identical to the response without the filter.  (Fig. S1) and internal pump make CAFE a simple instrument to integrate. Fasteners attach CAFE to structural angles from below, using mounting 5 holes with blind locking nut plates on the bottom of the instrument. For tropospheric aircraft projects, the angles are riveted to non-pivoting cabinet slides (General Devices) to facilitate instrument access. The slides are mounted to the aircraft rack and the slide assembly is secured to the aircraft rack with custom retaining brackets and fasteners. For high altitude aircraft, the angles mount directly to the aircraft rack-photos of the CAFE installation in the left superpod midbody of the NASA ER-2 are shown in Fig. S3. 10

Data Comparison with ISAF
During the boreal fall deployment (2017) of the NASA Atmospheric TOMography (ATom) campaign (https://espo.nasa.gov/atom/content/ATom), CAFE joined ISAF on the NASA DC-8 aircraft and sampled from the same inlet. While most ATom flights were in the remote atmosphere where [HCHO] is low, the first test flight on Sept. 14, 2017 out of Palmdale, CA provided good dynamic range with [HCHO] ranging from 100 pptv to over 3 ppbv as shown in Fig. 9. 15 CAFE was operated for ~3 hrs of the 4.3 hr flight to allow the instrument operator to focus on ISAF, the primary instrument for the campaign. The data in Fig. 9 is plotted as both (a) a time series along with altitude and (b) a cross plot with a linear regression. The time series clearly demonstrates the superior precision of the ISAF measurement. The cross plot and linear regression (York et al., 2004) show a good agreement between the two instruments, with a slope of 1.04 ± 0.02 and a yintercept of 24 ± 7 pptv. 20 The agreement was improved through the application of an ozone-dependent correction to the CAFE data, particularly for the time period of 22.5-23 hrs where the DC-8 passed through stratospherically influenced air. Earlier ATom circuits revealed a strong positive correlation between O3 and ISAF HCHO in stratospheric air masses, with 0.08 pptv of HCHO produced from exposure to every 1 ppbv of O3. Laboratory tests revealed that prolonged exposure to O3 in lab eliminated 25 the interference in subsequent flights: no O3 correction was needed for ISAF during ATom-3. CAFE was not conditioned in a similar manner before ATom-3 and a correction of 1.2 pptv of HCHO per 1 ppbv of O3 was necessary for ATom-3 data.
Possible sources of the higher HCHO yield in CAFE include newer instrument components and the use of THV fluoropolymer tubing, which is no longer used in CAFE for sampling.