A new in situ instrument for gas-phase formaldehyde
(HCHO), COmpact Formaldehyde FluorescencE Experiment (COFFEE), is presented.
COFFEE utilizes non-resonant laser-induced fluorescence (NR-LIF) to measure
HCHO, with 300 mW of 40 kHz 355 nm laser output exciting multiple HCHO
absorption features. The resulting HCHO fluorescence is collected at 5 ns
resolution, and the fluorescence time profile is fit to yield the ambient
HCHO mixing ratio. Typical 1
Formaldehyde (HCHO) is an abundant, photochemically influential trace
species in the Earth's atmosphere. Primary sources of HCHO include biomass
burning (Akagi et al., 2011; Andreae and Merlet, 2001) and fossil fuel combustion (Anderson
et al., 1996; Luecken et al., 2012; Olaguer et al., 2009), but these are
dwarfed by secondary production from the photochemical oxidation of volatile
organic compounds (VOCs). This secondary source is dominated by the locally
abundant VOC(s): CH
Atmospheric HCHO is measured using a variety of airborne instrumental methods, including mass spectrometry (Warneke et al., 2011), wet chemistry (Aiello and McLaren, 2009; Junkermann and Burger, 2006; Lazrus et al., 1988), absorption spectroscopy (Baidar et al., 2013; Catoire et al., 2012; Richter et al., 2015; Washenfelder et al., 2016; Weibring et al., 2006; Yokelson et al., 1999), and laser-induced fluorescence (LIF) (Cazorla et al., 2015; Hottle et al., 2009; Mohlmann, 1985). In addition to airborne observations, total column HCHO is measured by satellite (Chance et al., 2000; Steck et al., 2008), making HCHO one of the few VOCs observable from space. Numerous measurement technique reviews and instrument intercomparisons are available (Fried et al., 2008a; Hak et al., 2005; Kaiser et al., 2014; Zhu et al., 2016).
Traditionally, LIF measurements of HCHO have used a wavelength-tunable excitation laser to dither on and off the HCHO absorption feature, using the difference in signal to calculate the HCHO mixing ratio. The benefit of this approach is that the differential signal excludes any broadband background fluorescence from interfering with the HCHO measurement. The downside is that it requires either a large laser system unsuited for compact airborne instrumentation (Mohlmann, 1985) or a custom, high-cost fiber laser (Cazorla et al., 2015; Hottle et al., 2009). We present a new approach to the measurement of HCHO by non-resonant laser-induced fluorescence (NR-LIF), using a fixed-wavelength UV industrial laser at 355 nm to excite multiple HCHO absorption features simultaneously. Lacking the tunability and narrow linewidth necessary to dither on and off a single absorption feature, selectivity to HCHO is instead obtained using specialized fluorescence optical filters and by employing high-temporal-resolution data acquisition to uniquely identify HCHO by its characteristic fluorescence lifetime.
The new NR-LIF HCHO instrument, COmpact Formaldehyde FluorescencE Experiment (COFFEE), was designed specifically to join the payload of the Alpha Jet Atmospheric eXperiment (AJAX) out of the NASA Ames Research Center in Mountain View, CA. The robust optomechanical design of the COFFEE instrument, combined with its simple and reliable operation, makes the instrument ideal for long-term deployment to the NASA Ames Research Center with minimal maintenance. The routine, long-term nature of the AJAX project, with flights approximately every 2 weeks, makes the Alpha Jet a good platform for monitoring seasonal and long-terms trends, as well as for providing an extensive in situ data set for satellite validation.
The COFFEE instrument uses NR-LIF for the detection of HCHO. Previous
LIF-based instruments for atmospheric HCHO, such as the NASA In Situ
Airborne Formaldehyde (ISAF) instrument (Cazorla et al., 2015), have
used a narrow-bandwidth, state-specific tunable excitation laser to target a
specific absorption feature. COFFEE, in contrast, employs a
moderate-bandwidth (full width at half maximum, FWHM,
HCHO fluorescence occurs over the
Other LIF-based instruments for atmospheric HCHO (Cazorla et al., 2015; Hottle et al., 2009) collect fluorescence using a long-pass filter to exclude scatter and achieve measurement selectivity by alternately tuning the narrow-bandwidth laser on and off an HCHO absorption feature. The fixed-wavelength laser in COFFEE cannot provide on and off line measurements. Measurement specificity to HCHO is instead achieved by acquiring the time-resolved fluorescence signal, 5 ns bins for 500 ns, and leveraging the unique fluorescence lifetime of HCHO in data processing. The details of data acquisition and data processing are discussed in Sect. 3.4 and 3.5, respectively.
COFFEE instrument layout in the AJAX pod rack, including
the
A Spectra-Physics Explorer (EXPL-355–300-E, Fig. 2, item A) provides 300 mW
of 355 nm of pulsed radiation at 40 kHz (adjustable 20–60 kHz). The laser is
actively Q switched, with a Nd:YVO
The laser head requires proper thermal management for the laser to perform to specification. A total of 40 W of heat must be removed from the laser head at its maximum operating temperature of 308 K. Two thermoelectric cooler (TEC) devices (TE Technology) provide thermal control of the laser head. The laser side of the TECs is controlled to 303 K, and the other side of the TECs are in thermal contact with the optical plate and heat sinks mounted to the underside of the optical plate.
The optical plate layout is shown, with a cutaway of the
detection cell. The components include the
The optical layout of the instrument (Fig. 2, item B) is shown in more detail in Fig. 3. The entire optical system is contained on the optical plate in a single plane. The plate was machined out of 13 mm thick 6061 aluminum and is secured to the chassis at four points utilizing Sorbothane vibration isolation bushings. The plate is heated to 303 K.
The laser beam is directed by two antireflection (AR)-coated dielectric
mirrors (CVI Laser Optics) into the detection cell. A collimating lens (F
On two sides of the detection cell, aspheric lenses (NA
The fundamental design consideration for the instrument sample flow is to
minimize the potential for the adsorption/release of HCHO to/from exposed
surfaces (Cazorla et al., 2015; Wert et al., 2002). To that end, all surfaces that deliver gas to
the detection cell are either fluorocarbon (FEP, THV) or fluorocarbon coated
(FluoroPel, Cytonix). The current Alpha Jet inlet is a 9.5 mm OD (6 mm ID) rear-facing stainless
steel tube that extends 17 cm beyond the bottom of the
pod. A 9.5 mm OD (6.35 mm ID) THV fluoropolymer tubing connects the inlet to
the instrument chassis. The instrument is operated with an inline particle
filter (Balston 9922-05-DQ) to minimize related measurement artifacts from
high aerosol loading (see Sect. 4.5). The filter housing is Kynar
(polyvinylidene fluoride) and the filter element is a microfiber with a
fluorocarbon resin binder. The element retains 93 % of the particles with
a 0.01
Inside the instrument, 5 cm of 9.5 mm OD PFA tubing connects from the
chassis to a pressure controller, and 15 cm of 9.5 mm OD PFA tubing connects
from the pressure controller to the detection cell. The pressure controller
(Fig. 2, item C) is an actuator (iQ Valves) coupled with a custom valve
block and is heated to 308 K. The detection cell pressure is regulated to
10.7 kPa. The main flow passes directly down through the detection cell and
out of the chassis to the vacuum pump (Vacuubrand MD-1; Fig. 2, item D). A
small amount of air is pulled through the laser baffle arms to flush that
volume, and the flow is combined with the main flow (after the detection
cell) before exiting the chassis. In lab, the instrument sampling flow is
2.3 standard L min
Data acquisition and instrument control is conducted by a National Instruments (NI) CompactRIO system, hereafter RIO (Fig. 2, item E). The RIO consists of a main processor module (running a realtime operating system) and a backplane driven by a field programmable gate array (FPGA). Additional plugin modules add I/O. NI 9205 and NI 9264 modules provide analog input and output, respectively. Two channels of an NI 9402 high-speed digital I/O module are programmed as 5 ns resolution counters, with each PMT having its own counter. The counters are triggered by the OptoSync from the laser (30–100 ns after the laser pulse), which provides a digital logic pulse closer in coincidence with the laser light pulse than obtained from the “trigger out” logic pulse synchronous with the laser trigger. In order for the PMT signals to arrive after the counters are triggered, they are delayed by 50 ns with a passive delay circuit (Data Delay Devices, 1515 series).
Data for each PMT channel are acquired in two ways: (1) integrated every 0.1 s with ungated (continuous) and gated data streams, which are used primarily for diagnostic purposes; and (2) integrated every 1 s and time resolved to 100 discrete time bins, each 5 ns wide, that cover the 500 ns immediately following the counter trigger. The 5 ns time-resolved data are the key to the data processing approach necessary to minimize measurement artifacts with the NR-LIF approach, as will be discussed in Sect. 3.5, and are used to produce the HCHO mixing ratio data product. Diagnostic data (laser power, pressures, temperatures, etc.) are also recorded every 1 s.
HCHO mixing ratios are obtained using the 5 ns bin time-resolved profiles from the two detection axes. The data processing consists of three steps, each done independently for the two detection axes: (1) subtraction of the minor “long-lived” component from the time profile; (2) two-parameter nonlinear least squares fit of the data using profiles (hereafter referred to as exemplars) that represent the HCHO and non-HCHO (chamber scatter, Raman and Rayleigh scatter, fluorescence of optics, etc.) contributions to the observed profile; (3) one-parameter nonlinear least squares fit with the non-HCHO contribution fixed from the previous two-parameter fit and only the HCHO contribution allowed to vary. The second pass fit with only one parameter improves the precision of the measurement. In addition to the fitting-based data processing, HCHO mixing ratios can also be obtained from gated count data, as discussed in Sect. 3.5.3.
The fluorescence signal at the end of the bin-resolved data (> 400 ns) is small but nonzero, and changes in this long-lived signal do
not scale with changes in the non-HCHO “air exemplar”, necessitating a
separate treatment. The long-lived signal has a longer fluorescence lifetime
than HCHO, which permits fitting and removal of the long-lived signal
without interference from ambient HCHO. For detection axis 1, an empirical
profile determined from a laboratory run is scaled to fit the observed 1 Hz
data using a single parameter least squares fit to the observed profile from
bin 75 to bin 100 (370–500 ns), and the scaled profile is subtracted from
the observed data before performing the exemplar fits. For detection axis 2,
the long-lived signal is smaller than for axis 1 by a factor of
Exemplar time profiles are obtained in the laboratory. The air exemplar is created by averaging the profile with no added HCHO (blue dashed line), and the HCHO exemplar (red dashed line) is obtained by subtracting the air exemplar from data with high (29 ppbv) HCHO (cyan circles).
The representative time profiles, or exemplars, are determined from laboratory calibration runs where the instrument samples clean, dry air (typically ultra-high purity, UHP, dry air) with varied amounts of HCHO added. The air exemplar, which represents all non-HCHO contributions to the observed profile, is obtained by time averaging the observed profile when no HCHO is added to the dry air. Figure 4 shows the profiles involved in creating the “HCHO exemplar”. The HCHO exemplar (Fig. 4, red dashed) is obtained by time averaging the observed profile (Fig. 4, cyan circles) during the calibration period of maximum HCHO (typically 25–30 ppbv) and subtracting the air exemplar (Fig. 4, blue dashed) from the time-averaged profile. The highest HCHO period is used so that HCHO dominates the shape of the observed profile. Once the air and HCHO exemplars are obtained, they are used to fit laboratory calibration data with multiple HCHO concentrations using the two-step fit described below. The calibration factor unique to this HCHO exemplar is obtained from the linear regression of the HCHO added by the calibration system and the HCHO exemplar scaling factor, which is the output of the fit.
Each 1 s data profile (cyan circles) is fit using a linear combination of the air (dashed blue line) and HCHO (dashed red line). The fit profile, over the time window used for the least squares fit, is shown in black. The HCHO mixing ratio is 8 ppbv.
An example two-parameter exemplar fit is shown in Fig. 5. The observed profile (Fig. 5, cyan circles), with the long-lived component removed, is fit with a linear combination of the air exemplar and the HCHO exemplar. The fit parameters are the scalar multipliers applied to the exemplars: the scaled air exemplar (Fig. 5, blue dashed) and scaled HCHO exemplar (Fig. 5, red dashed). The least squares optimization is performed on the data from bin 13 to bin 60, with the fit window chosen to maximize data precision and fit quality, as determined by visual inspection of fit residuals. The optimized fit for the bin 13–60 window is shown in black.
The first step of the one-parameter fit applies a 21 s median filter to the
vector of air exemplar fit scalars from the two-parameter fit. The smoothed
vector is then used in a one-parameter fit where the air exemplar
contribution is fixed to the air exemplar scaled by the smoothed vector, and
the HCHO exemplar scaling factor is allowed to vary. The result is a higher
precision fit and is possible because the phenomena that comprise the air
exemplar contribution to the observed profile (chamber scatter, Raman and
Rayleigh scatter, fluorescence of optics, etc.) do not change rapidly. The
output of the one-parameter fit, the HCHO exemplar scalar, is directly
proportional to HCHO mixing ratio. HCHO data in pptv are obtained by
applying a calibration factor, which is unique to the HCHO exemplar used, to the fit
output. The final HCHO mixing ratio data product is the arithmetic mean of
the data from the two detection axes. Data from the two axes generally
agree well – Fig. S4 shows the cross plot of 60 s data from the two axes,
along with a linear fit (slope
In addition to data processing with exemplar profiles, time-gated 1 Hz data derived from the time-resolved profiles can be used to obtain HCHO, with higher measurement precision than is achieved with the exemplar fits. For example, a calibration experiment yields, for 1 s data and 0 pptv added HCHO, standard deviations of 150 pptv for the one-parameter exemplar fit (175 pptv for the two-parameter fit only) and 130 pptv for the gated count data (167 pptv for ungated count data). The time-gated data exclude much of the prompt signal from scatter by summing counts from bin 24 to bin 100 (115 to 500 ns). Using the same laboratory calibration experiment as an example, with no HCHO added, the gate excluded 89 % (450 nm filter detection axis) and 95 % (multi-band-pass axis) of the total signal in the first 500 ns from the trigger. More of the HCHO signal is retained due to its fluorescence lifetime: 73 % of the HCHO signal is excluded by the gate. Figure S5 shows the time profile from Fig. 5 with the gate window shaded. Gated count 10 Hz data, as well as ungated count 10 Hz data, can be used to obtain HCHO mixing ratios. The 10 Hz data are used only for diagnostic purposes, e.g., the instrument flush time experiment in Sect. 4.4.
The count signal is converted to the HCHO mixing ratio using a linear
relationship determined from laboratory calibrations, with the slope being
the instrument sensitivity to HCHO (discussed in Sect. 4.1) and the
intercept being the signal at HCHO
The sensitivity of each detection axis to a given amount of HCHO is a function of a number of instrument parameters: laser power, collection optics efficiency, fluorescence optical filter transmission, and PMT response. As for ISAF, none of the instrument parameters that affect instrument sensitivity are expected to degrade on a timescale shorter than years. The HCHO calibration of the instrument has been measured 2–3 times per year and will be measured at least once a year in the future to track any changes in sensitivity.
Calibration is performed using measured flows from two cylinders, one
containing ultra-high purity air further purified with a Drierite
and molecular sieve scrubber and the other a
For calibration, flow of the HCHO standard is sequentially set to 3–5
different flows in the range 0–50 standard cm
Instrument sensitivity to HCHO differs for the two detection axes primarily
due to their respective optical filter transmission, with axis 1 being more
sensitive than axis 2. The gated count sensitivities for axis 1 and axis 2
are 0.29 and 0.13 counts s
The standard deviation as a function of HCHO is shown to demonstrate the precision of the HCHO measurement for 1 and 10 s averaging.
Measurement precision is the dominant component of overall measurement
uncertainty at low (< 700 pptv) mixing ratios. The standard
deviation using data from two laboratory calibration experiments is shown in
Fig. 6. At [HCHO]
The normalized Allan–Werle deviation as a function of averaging time
(
Precision should improve as data are time averaged. In practice, the benefit
of additional time averaging ceases when the data variability is no longer
dominated by random noise. The Allan–Werle deviation plot shown in Fig. 7
demonstrates this point for COFFEE HCHO data from a laboratory calibration
with
The overall measurement uncertainty for COFFEE HCHO is estimated to be
Instrument time response to a pulse of HCHO is fit with an
exponential decay, giving an empirical
Instrument time response directly affects the ability to resolve fine
structure in atmospheric HCHO and can affect measurement accuracy in
regions of high HCHO contrast such as biomass burning plumes. Understanding
the instrument time response is critical to properly interpreting the in
situ data. Assuming a volume of 60 cm
COFFEE (blue) and ISAF (red) data sampling from the roof of
Building 33 at GSFC in June 2015. Shaded sections indicate COFFEE sampling
without a particle filter. HCHO above 8 ppbv was from sampling indoor air,
Mie scattering from the presence of aerosol increases the prompt signal
(
The COFFEE instrument, installed in its AJAX pod rack, is mounted into the mid-body of the inboard left pod.
COFFEE was designed specifically for integration onto the Alpha Jet (H211,
LLC) stationed at the NASA Ames Research Center Moffett Field to participate in the Alpha Jet
Atmospheric eXperiment (Hamill et al., 2016). The
Alpha Jet carries four wing pods, with the outboard pods containing fuel and
the inboard pods available for instrumentation. Each instrument wing pod has
a usable volume of
Map of AJAX flight tracks with COFFEE in payload through March 2017.
The first flight of COFFEE on the Alpha Jet was on 15 December 2015. Since then, COFFEE has operated on 27 AJAX flights through March 2017, with data coverage predominately in the Bay Area and Central Valley of California. Figure 11 shows a map with overlaid flight tracks for all the AJAX flights with COFFEE, and Table S1 lists the dates and objectives for each of the 27 flights. During this period COFFEE returned to GSFC just three times for maintenance, typically timed to coincide with aircraft maintenance.
Flight data for AJAX flight 185 on 19 April 2006.
Data from AJAX flight 185 on 19 April 2016 are shown in Fig. 12 as an
example of COFFEE HCHO performance. The flight included two spiral profiles,
one over the San Joaquin Valley near Merced, CA (37.38
The NR-LIF technique utilized in COFFEE has proven to be a viable,
operationally robust approach to measuring gas-phase in situ HCHO. While not
achieving the sensitivity of a state-selective LIF instrument such as ISAF
(Cazorla et al., 2015), the NR-LIF technique provides adequate precision (1
AJAX data are available upon request (Laura Iraci, laura.t.iraci@nasa.gov).
The authors declare that they have no conflict of interest.
Funding was provided by the Goddard Internal Research and Development (IRAD) program and NASA (NNH16ZDA001N-UACO). J. E. Marrero gratefully acknowledges funding from the NASA Postdoctoral Program. Edited by: Piero Di Carlo Reviewed by: Alan Fried and one anonymous referee