Evaluation of the airborne quantum cascade laser spectrometer (QCLS) measurements of the carbon and greenhouse gas suite – CO2, CH4, N2O, and CO – during the CalNex and HIPPO campaigns

Abstract

effective optical path length (McManus et al., 1995).Laser beams from QCL2 and QCL3 are co-94 aligned through an anchor point before being directed into the sample multi-pass cell.The light 95 pulses from the 3 QCLs are detected using photovoltaic detectors housed and cooled in two 96 liquid nitrogen (LN 2 ) dewars: one for CO 2 /QCL1, and the other for both CH 4 /N 2 O/QCL2 and 97 CO/QCL3.The CO 2 optical table, QCL1, two 10-cm path length sampling cells, and a dewar 98 housing InSb detectors for the CO 2 portion of the QCLS are enclosed in a temperature-controlled 99 pressure vessel flushed with ultra-high-purity nitrogen to remove the effects of absorption 100 external to the sampling cells.QCL2 and QCL3, an astigmatic multi-pass sampling cell with an 101 effective 76 m path length, and a dewar housing the HgCdTe detectors are mounted on a second 102 optical table surrounded by a temperature-regulated enclosure.The pulses from QCL2 and QCL3 103 are temporally multiplexed on the same pair of detectors.104 The spectra acquired from the two optical tables are controlled and analyzed by the same 105 computer.TDL Wintel® software controls the laser temperature and overall output frequency, 106 the tuning ramp rate (the wavelength frequency range and rate over which the laser is tuned) and 107 the detector multiplexing for QCL2 and QCL3 which share a pair of common detectors.The 108 temperature regulation of the QCLs is achieved by means of Peltier modules coupled to a closed-109 circuit recirculating fluid kept at fixed temperature within 288.0 ± 0.1 K.With the exception of 110 the chiller fluid, electronics and computer, the CO 2 measurement (QCLS-CO2) can be 111 considered independent from the CH 4 , N 2 O, and CO measurements (QCLS-DUAL) and we refer 112 to those two sensors as such.113 The instrument is fully autonomous and sampling, calibration, temperature regulation, 114 and pressure regulation are controlled by a data-logger (CR10X, Campbell Scientific).It logs 115 control variables and periodically dumps them via a serial connection to the computer running 116 TDL Wintel® for storage on a solid state hard drive.Because the sampling and control strategy 117 is controlled by the data-logger and the spectral analysis is performed by the TDL Wintel® 118 software running on the computer, in-flight spectra are acquired using a fixed nominal cell 119 pressure and cell temperature.Raw spectra are later reanalyzed with the logged CR10 cell 120 pressure and cell temperature measurements to generate spectroscopically-calibrated mixing 121 ratios.Figure 1 shows the raw spectra and the Levenberg-Marquardt fits to the absorption lines 122 within the scan according to the HITRAN database (Rothman et al., 2007) for the three QCLs.123 The CO 2 spectrum appears inverted because this particular air sample has less CO 2 than the 124 calibration air flowing through the reference cell.125 Optical-based measurements are particularly sensitive to fluctuations in temperature and 126 pressure (Zahniser et al., 1995) and careful controls must be implemented, particularly during 127 flight where large dynamic ranges in both variables are observed (Fried et al., 2008).In-flight 128 calibrations at regular intervals from gas cylinders are used to track sensor drift.As long as the 129 inter-calibration time interval is shorter than the long-term drift, standard additions can offset 130 inaccuracies due to pressure and temperature fluctuations.The Allan variance, a measure of the 131 precision of a sensor as a function of averaging time (Werle et al.,1993), can be used to quantify 132 both the short-term (e.g.electronic noise) and long-term precision of a sensor as well as the drift.133 Figure 2 shows the in-flight Allan variance for the CO 2 , CH 4 , N 2 O, and CO measurements from 134 the QCLS with 1-second RMS precisions (Allan standard deviations) of 20, 0.5, 0.09, and 0.15 135 ppb, respectively.The measurements shown in Figure 2 were taken during a section of HIPPO 136 that sampled a relatively constant air mass above the remote Pacific Ocean.This is the same 137 section of data presented in the supplementary material section of Kort et al. (2011).The flow schematics for QCLS-CO2 are shown in Figure 3.The two schematics are very 147 similar.QCLS-CO2 and QCLS-DUAL have independent inlets.On the HIAPER-GV, both inlets 148 extend out from the QCLS rack to a dedicated NCAR HIAPER Modular Inlet (HIMIL) mounted 149 to the edge of the aircraft.The HIMIL extends the inlet 28 cm from the body of the aircraft 150 (NCAR, 2005) and the two QCLS inlets, both 316 stainless steel 0.25" OD sample from within 151 the center flow path, oriented away from the direction of flow (i.e.rear-facing).This orientation 152 minimizes large particle entrainment and protects the sampling system from liquid water and ice.153 For the NOAA P-3 aircraft, the inlets both consist of stainless steel 3/8" OD tubing bent at 90 154 degrees to be parallel to the aircraft and oriented at -135 degrees relative to the horizontal 155 direction of flight.Once the sample enters the body of both aircraft, the two sample lines consist 156 of ~1.5 m of Synflex type 1300 tubing (6.35 mm = ¼" OD for QCLS-CO2 and 9.525 mm = 3/8" 157 OD for QCLS-DUAL) and each sample stream reaches a 2 µm filter (47 mm OD Pall Zefluor 158 membrane) mounted in an aluminum filter holder (Gelman Sciences, Inc., Rossdorf, Germany).159 Calibration gases are added downstream of the filter using a combination of 2-way and 3-way 160 solenoid valves.When activated, the solenoid valves allow air from two sets of calibration gas 161 decks which each include 3 cylinders (1.1 L for QCLS-CO2 and 2.0 L for QCLS-DUAL) to 162 'over-blow' the inlet, with the excess flow exiting the aircraft through the HIMIL.The regulators 163 for the calibration cylinders are set on the ground to achieve an excess flow >100 sccm (QCLS-164 CO2) or >200 sccm (QCLS-DUAL) which flows via the filters and inlets out the aircraft.From 165 this point, the sample (or calibration) air travels through a 1-tube (QCLS-CO2, see Daube et al.,166 2002 for an explanation of this choice) or 50-tube (QCLS-DUAL) Nafion membrane dryer to 167 remove the bulk of the water vapor.Then the air passes through a Teflon dry-ice trap to further 168 reduce the dewpoint to below -70 °C.A stainless steel filter (Swagelok, SS-4FW-2, 2 µm 169 stainless steel mesh) at the outlet of the dry-ice trap ensures that particles cannot exit the trap, 170 thaw, evaporate, and contaminate the measurement cell mirrors.From the dry ice trap, air enters 171 the sample cells, the pressures of which are controlled both upstream and downstream of the cell 172 using a pressure controller and valve (MKS 722, 100 torr range).For QCLS-CO2, the 9.7 cm 3 173 sample cell is controlled to 70 ± 0.1 hPa using another MKS 722 and the reference cell pressure 174 is matched using a differential pressure controller and valve (MKS 223B, 100 torr absolute, 10 175 torr differential range).For QCLS-DUAL, the 0.5 L cell is controlled to 77 ± 0.1 hPa.After the 176 pressure control element downstream of the sample cells, the flows are routed back through the 177 outer tube enclosing the Nafion membrane tubes to create the necessary H 2 O gradient across the 178 membrane.The flows are then combined into a 4-stage diaphragm pump (KNF Neuberger, Inc. 179 UN726) fitted with Teflon-lined diaphragms.Two of the heads are connected in parallel and the 180 remaining two downstream pumps are connected in series to compensate throughput and power.

181
For the HIAPER-GV, the exhaust is then dumped to a dedicated exhaust manifold in the aircraft.

182
For CalNex, the exhaust is dumped through a third stainless steel port downstream of the inlet.In-flight calibrations are done by replacing air in the sample cell with air from 194 compressed gas cylinders in two gas decks mounted on the QCLS flight rack (see Figure 3).to ~ 2100 psi and the flight cylinders in the gas decks are filled to as high a pressure as possible, 210 usually >1800 psi.Gas-deck cylinders are filled directly from the AL secondary cylinders using 211 1/8" stainless-steel tubing after 3 rounds of flushing and purging the regulator and the fill line.

212
The gas-deck cylinders are themselves conditioned by being purged, then filled and flushed 213 twice (to 300 psig then 500 psig) before being filled to maximum pressure.Gas decks are 214 sampled until the pressure drops to 500 psi, well before drifts in concentration become apparent 215 (Daube et al.,2002).Table 2a summarizes the two calibrations of the primary cylinders used in 216 HIPPO and CalNex and Table 2b summarizes the calibration obtained for the secondary 217 cylinders used to fill the QCLS-DUAL gas deck.

218
Figure 4 shows the calibration procedure used to calibrate a secondary cylinder for 219 QCLS-DUAL in the lab.We turn on the QCLS and allow it to equilibrate while it is sampling 220 zero-air from an AL cylinder for at least 2 hours.Three primary tanks and a secondary 'target' 221 tank are plumbed into an external bank of solenoid valves connected to the QCLS-DUAL via an 222 external port on the gas deck.The QCLS is operated in exactly the same mode as during in-flight 223 sample additions of calibrated air, where the calibration solenoid is actuated and excess 224 calibration air (>200 sccm for QCLS-DUAL) flows out through the QCLS inlet.The primary 225 and secondary tanks are plumbed into the external solenoid bank and the QCLS via 1/8" OD 226 stainless steel tubing.After equilibration, we sample zero-air for 5 minutes, then sequentially 227 flow air from three primaries in order of lowest to highest concentration for 3 minutes each.

228
After this sequence, we sample the target secondary tank, also for three minutes.We then repeat 229 this cycle an additional 3 times, as shown in Figure 4, not sampling the target secondary on the 230 last iteration.Figure 5 shows the data corresponding to the 270 pink points in Figure 4, 231 concatenated together for each of the QCLS-DUAL species during three independent sets of 232 calibrations in 2010-2012 for the same tank, CC89589 (Table 2b).We calculate linearly 233 interpolated (using the two closest) and quadratically interpolated (using all three) values that 234 correspond to the mean of the three 90-second sampling segments.The average of those values is 235 reported as the calibrated secondary values (Table 2b), where values more than 2σ from the 236 mean, if they exist, are excluded in the calculation.

237
We test for filling errors by filling the gas decks with secondary tanks and then 238 performing a similar calibration of the gas deck itself.For gas deck calibration, we sample the 239 low-span and high-span 'targets' one after the other and use all four primary cylinders.Because 240 use of the primary cylinders with QCLS required the instrument to be in a laboratory setting, we 241 were able to perform gas-deck calibrations only before or after a given deployment.When the 242 small cylinders in the gas decks reached 500 psi, they were flushed (3X) and filled with 243 calibration air from AL secondary cylinders.For HIPPO, the refills would take place in 244 Christchurch, NZ, using a different set of secondary cylinders than the secondary cylinders used 245 to fill the gas decks on the first half (southbound) set of the HIPPO flight circuit.We would 246 therefore calibrate the gas deck after filling it but before using in on the southbound flights, and 247 after both filling it and using it on the second half northbound set of the HIPPO flight circuit.

248
Because of these logistics, the calibration values calculated during gas-deck calibrations were 249 only used as a check against filling error, ensuring that the gas-deck values fell within 3σ of the 250 uncertainty attributed to the secondary tank calibration (Figure 5) from the NOAA primary 251 cylinders.For consistency, and because calibrations of the gas decks themselves required the use 252 of the primary cylinders, the calibration values assigned to the air in the gas decks were always 253 the values from the secondary cylinder calibrations shown in Figure 5. this effect is often unavoidable, but is important to consider in the context of measurement 291 traceability.We reached a compromise by sampling the zero-air for 2 minutes, sampling the LS 292 and HS before sampling the zero-air, and using a smaller sampling window to calculate the zero-293 air spectroscopically-calibrated mixing ratios of N 2 O, as seen in Figure 6.

294
Using a reference calibration cylinder (one with near-ambient atmospheric 295 concentrations, e.g.CC56519 in Table 2b) instead of a zero to track instrument stability would 296 minimize the effect of this problem.Because this tank is used so frequently to track drift, 297 however, it would have been impractical to use, particularly on HIPPO where opportunities to 298 ship calibration tanks and refill the gas decks are limited.We tested this assumption on one flight 299 during HIPPO V (RF14; Sept 9, 2011) and showed that using a 1-minute equilibration time for a 300 reference tank at ambient concentrations gave nearly equivalent results as using a 2-minute zero-301 air tank.

302
It should be noted that the boiling point of CO 2 (-57°C) is even higher than that of N 2 O, 303 so this effect is equally important for CO 2 and can be observed in Figure 6.However, it matters 304 to a much smaller extent as the Synflex is always in contact with air that is very close to ambient.

305
To distinguish sampling intervals from calibration intervals, we use an empirical 306 relationship that is a function of ambient pressure and tubing length.These differ for HIPPO and 307 CalNex because of the hardware configurations, notably the use of the HIMIL on HIPPO.For 308 HIPPO and QCLS-DUAL, we calculate experimental delay times from the HIMIL to the 309 calibration-addition point just downstream of the inlet filter (Figure 3) as a linear function of 310 ambient pressure in the HIMIL.We also calculate a time delay corresponding to the equilibration 311 time from that point to the measurement cell as a quadratic function of ambient pressure in the 312 HIMIL.These have the functional form: subtracted from the entire dataset.Using the zero-air Akima-spline-subtracted data for the 341 QCLS-DUAL species (for example,  !,!"#$ ), the mean values of each low-span and high-span 342 window are interpolated using the same Akima-spline to the measurement times. !,!"#$ is 343 then linearly interpolated to the low-span Akima-spline and high-span Akima-spline ( !,!_!"# 344 and  !,!_!"# , respectively) according to: 345 346 (3) 347

348
where  !,!! !"# and  !,!! !"# are the two constant values of the low-span and high-span 349 secondary AL calibration cylinders used to fill the gas-deck (Table 2b).The equations for N 2 O 350 and CO are equivalent and generate the calibrated sample dry-air mole-fractions ( !,!"# in 351 Equation 3). Figure 8 shows the different Akima-splines for an arbitrary flight during HIPPO 5, 352 along with the ambient pressure.The axes are all scaled such that the different tracers -zero-air, 353 low-span, high-span, and reference air -from the gas decks have equivalent ordinate ranges.The 354 CO 2 trace in Figure 8 is in units of ppb relative to the reference, meaning that a value of -17500 355 corresponds to the low-span that is 17.5 ppm lower than the near-ambient reference.Figure 8 is a 356 standard output product of the batch processing and is purposely scaled to emphasize the 357 fluctuations of calibration standards over the course of a given flight.Because of the linear 358 interpolation between the zero-subtracted low-span and high-span, the relative fluctuation of 359 those two standards has the largest effect on the effective calibrated measurements.360 The CO 2 calibration additions shown in Figure 8 are treated in a slightly different fashion 361 than the QCLS-DUAL species.Because QCLS-CO2 is a differential measurement and the range 362 of observations is the largest of any species (in terms of concentration changes measured over 363 the course of a flight), the CO 2 interpolation is not calculated linearly.Instead, we take the 364 median of the low-span, reference, and high-span values calculated over the course of any 365 particular flight and fit a quadratic function to those median values for that flight.The reference-366 subtracted measurements are then quadratically interpolated using this fixed function.We 367 experimented with different methods to calibrate the CO 2 measurements and found that using a 368 method similar to QCLS-DUAL resulted in spurious wave generation in the measurements that 369 was not physical.Because the reference/zero calibration is sampled at 2X the frequency of the 370 spans, the reference trace is able to best compensate for the measurement drift.Physically, we 371 expect that the response of QCLS-CO2 over the range of concentrations sampled should not 372 change dramatically, and this is confirmed in the flight-to-flight variability of the quadratic 373 interpolation function (see below).For this reason, we fix the quadratic function and make it 374 follow the more frequent reference calibration trace.

375
Figure 9 shows the variability of the quadratic function for QCLS-CO2 (during CalNex) 376 and the linearity of QCLS-DUAL (in lab using data from a secondary tank calibration).The 4 377 sets of panels show the 1:1 plot of the raw spectroscopically-calibrated QCLS mixing ratios 378 versus the NOAA-calibrated primary cylinder values.The linear fits to QCLS-DUAL are 379 calculated using a type II regression with prescribed errors in the abscissa and ordinate (York, 380 2004).For the x-axis, the uncertainties are prescribed by the NOAA calibrations and for the y-381 axis, errors are given by the standard deviation of the mean spectroscopically-calibrated QCLS 382 measurements.The bottom panel shows the residual values for the different tanks.For CO 2 , the 383 fit is not linear, as described above, and the residuals shown are flight-to-flight differences in the 384 quadratic fit function over the course of the CalNex mission, which showed greater variance in 385 the quadratic fit coefficients compared to HIPPO.The residual values shown for CO 2 correspond 386 to the standard deviation of the quadratic fit function over the mission, and can be considered an 387 estimate of the sensor accuracy as a function of concentration.To put these estimates of errors in 388 context, the histogram distributions of the HIPPO and CalNex CO 2 measurements are shown 389 along with their 10-90% quantile ranges (solid blue and red lines) to show that this is a very 390 minor error effect for the majority of the measurements.

391
The accuracy of the QCLS-CO2 measurements are determined by secondary cylinders 392 calibrated against NOAA standards using the Harvard Ground Support Equipment (GSE), 393 described in detail in Daube et  the exact corrections for the tanks.The NOAA primary tanks had near-atmospheric 13 C isotopic 440 composition of around -10 to -15‰, Scott Specialty tanks usually fell in the -45 to -50‰ range, 441 and Scott-Marrin usually fell in the -30 to -40‰ range.

442
The error for CH 4 due to differing isotopic composition between the atmosphere ( 13 C ≈ -443 47 ‰) and calibration cylinders ( 13 C ≈ -30‰) was calculated to be a ~0.3 ppb effect, smaller 444 than the 1 Hz precision.The effects for N 2 O and CO were proportionally smaller and these 445 effects are therefore ignored for QCLS-DUAL.446 447

Missions and Other Instrumentation 448
The QCLS was operated in the same configuration in both CalNex and HIPPO with only 449 minor changes due to the aircraft-specific issues already discussed.We now present comparisons 450 with other coincident instruments, synchronized in time using the STRATUM-1 aircraft data 451 system.

452
For HIPPO, two additional fast-response (>1 Hz) CO 2 sensors were available for 453 comparison: the OMS sensor (Daube et al.,2002), and the NCAR Airborne Oxygen Instrument 454 (AO2), which includes a single-cell Licor-820 sensor.Figure 10 shows the 1 Hz measurement 455 difference distribution for QCLS against OMS and AO2 for all HIPPO flights.QCLS-CO2 and 456 OMS agree to better than 0.05 ppm, with a standard deviation of the difference of 0.37, owing in 457 part to the slower cell response time of OMS.Assuming the sensors have no covariance, the 1Hz 458 OMS precision of 0.1 ppm and the 1 Hz QCLS precision of 0.02 ppm would sum in quadrature 459 for an expected precision of 0.1 ppm.The actual distribution is 0.37 ppm, roughly a factor of 4 460 higher.The AO2 instrument has a 1-sigma, 1-second precision of ~0.6 ppm.QCLS-CO2 and 461 AO2 agree to within 0.15 ppm and have an even larger variance on the distribution of the 462 measurement differences.It is important to note that many unresolved biases spanning one hour 463 or an entire flight exist among the CO 2 sensors on HIPPO and tend to average out as presented in 464 Figure 10.shown (bottom).Table 1 summarizes these data and also provides corresponding values for 696 sampling from a calibration cylinder in the laboratory.All concentrations are reported in units of 697 ppb, including CO 2 , which is the concentration relative to the reference concentration.In this 698 plot, the CO 2 concentration is ~1.5 ppm above the reference gas concentrations of ~390 ppm.(Table 2b) using 3 primary cylinders (Table 2a).Zero air (light blue) is sampled for 5 minutes, 708 the primary cylinders are then each sampled for 3 minutes (light green) in order of increasing 709 concentration and then the target secondary cylinder (pink) is sampled for 3 minutes.We use the 710 last 90 seconds of a given 5-minute zero-air sample (light blue) to calculate 5 zero-air values 711 (blue squares).These 5 values are linearly interpolated (Interp.) to the sampling times and that 712 time series (blue trace) is then subtracted from to the raw mixing ratios (gray trace) to yield the 713 black trace (which appears nearly indistinguishable from the gray trace except in the case of 714 CO).The last 90 seconds (light green) of the 'zero-subtracted' data (black trace = gray trace -715 blue trace) are then averaged to generate a value for each of the 4 primary sampling intervals 716 (dark green squares).For each primary, the primary value at the QCLS sampling times is 717 linearly interpolated to those 4 values (green lines).The last 90 seconds of the target sampling 718 window (pink) are then interpolated (both linearly and quadratically) to the green lines (either the 719 2 closest for linear interpolation or the closest 3 for quadratic interpolation) which bracket the 720 secondary concentration of interest, as shown in Figure 5. typically contributes no more than 0.1 ppm (i.e.black line intersects blue and red lines within 0.1 769 ppm).For QCLS-DUAL (B,C,D), the 1:1 correspondence of the spectroscopically-calibrated 770 QCLS mixing ratio is plotted against 4 known primary cylinders and regressions are calculated 771 using the error uncertainties from the primary cylinders shown in Figure 5 also shows that, 254 within uncertainty, there is no evidence of drift in the secondary cylinders from 2010 to 2012.255In-flight data are then tied to the NOAA scale by periodic sample replacement with air 256 from the gas decks.The sampling structure is shown in Figure6.Within a given 60 minutes, the 257 calibration sequences is as follows: minutes 7-9, 22-24, 37-39, 52-54 sampled zero/reference, 258 minutes 9-10 and 39-40 sampled low-span, minutes 10-11 and 40-41 sampled high-span, and 259 minutes 41-42 sampled a check-span.Because of the different equilibration times for the 260 different species, we changed the order of the LS and HS additions to occur before the zero-air 261 additions (see below).The zero was sampled most frequently at 15-minute intervals to track 262 QCLS drift.The low and high-spans were sampled at 30-minute intervals, and the reference-span 263 was sampled every hour for one minute.For a given hour of flight, the effective sampling duty 264 cycle was therefore ~78% (47 minutes of sampling per hour).The calibrations for QCLS-DUAL 265 and QCLS-CO2 occur on the same interval.Instead of sampling zero-air like QCLS-DUAL, 266 however, the QCLS-CO2 samples the reference gas in both the sample and reference cell in 267 order to obtain a relatively flat spectrum.The zero/reference is sampled for 2 minutes for two 268 main reasons: 1) the zero/reference is the most frequently sampled calibration standard and 269 therefore tracks the environmental temperature and pressure variability which cause drift, and 2) 270 the equilibration of the N 2 O trace is slower than the other species.Because the gas deck 271 reference/zero-air additions are used to track drift and to interpolate the measurement to standard 272 values, equilibration of the gas-deck standard additions is essential.273 We ran a number of tests to characterize the slow equilibration in N 2 O observed in the 274 zero-air additions.Figure 7 shows a concatenated time series of various sampling intervals in 275 which we repeatedly switched between a zero-air cylinder and a span cylinder for 3-minute 276 intervals.The different colors indicate different combinations of elements upstream of the 277 sampling cell that came into contact with the sample.The tests included instances in which the 278 air went straight from the cylinders to the sampling cell (through a nominal 0.5 m of Synflex that 279 was unavoidable).Various other upstream elements were added between the cylinder and the 280 sampling cell, including different lengths of Synflex, stainless steel tubing, PFA, and Nafion™ 281 tubing.Figure 7 shows this data superimposed upon one another (with the zero-air value 282 assigned from the mean value of 145-165 seconds of the 180 second sampling window) and the 283 y-axis range normalized by the secondary cylinder calibrated value (N 2 O = 319.3ppb, CH 4 = 284 1919.6 ppb, CO = 223.4ppb) and multiplied by an arbitrary constant (here 500) to zoom in on 285 the transition to the zero-air sampling.Both CH 4 and CO are largely unaffected by the different 286 sampling materials, likely because of their lower boiling points (-164°C and -192°C, 287 respectively) relative to N 2 O (-88°C).Stainless steel was the only sampling material that was not 288 affected by absorption/desorption for N 2 O.The importance of this effect scaled with the surface 289 area of the Synflex or PFA encountered.Using stainless steel is impractical in many instances, so 290

Figure 1 : 693 Figure 2 :
Absorption spectra for the 3 quantum cascade lasers.QCL1 (a) is a differential 688 measurement of 12 CO 2 and therefore appears inverted because this sample has a lower 689 concentration of CO 2 than the reference gas.QCL2 (b) shows the spectrum for CH 4 and N 2 O and 690 QCL3 (c) shows the spectrum for CO. 691 692 Time series for the 4 QCLS species during 20 minutes of in-flight sampling over the 694 Pacific during HIPPO II (top) and the Allan variance as a function of averaging time for the data 695

699 701 Figure 3 :Figure 4 :
Figure 4: The sampling sequence used to calibrate a secondary cylinder for QCLS-DUAL 707(Table2b) using 3 primary cylinders (Table2a).Zero air (light blue) is sampled for 5 minutes, 708 the primary cylinders are then each sampled for 3 minutes (light green) in order of increasing 709 concentration and then the target secondary cylinder (pink) is sampled for 3 minutes.We use the 710 last 90 seconds of a given 5-minute zero-air sample (light blue) to calculate 5 zero-air values 711 (blue squares).These 5 values are linearly interpolated (Interp.) to the sampling times and that 712 time series (blue trace) is then subtracted from to the raw mixing ratios (gray trace) to yield the 713 black trace (which appears nearly indistinguishable from the gray trace except in the case of 714 CO).The last 90 seconds (light green) of the 'zero-subtracted' data (black trace = gray trace -715 blue trace) are then averaged to generate a value for each of the 4 primary sampling intervals 716 (dark green squares).For each primary, the primary value at the QCLS sampling times is 717 linearly interpolated to those 4 values (green lines).The last 90 seconds of the target sampling 718 window (pink) are then interpolated (both linearly and quadratically) to the green lines (either the 719 2 closest for linear interpolation or the closest 3 for quadratic interpolation) which bracket the 720 secondary concentration of interest, as shown in Figure5.721 722

721 722 723 Figure 5 :Figure 6 :Figure 8 :Figure 9 :
Figure 4: The sampling sequence used to calibrate a secondary cylinder for QCLS-DUAL 707(Table2b) using 3 primary cylinders (Table2a).Zero air (light blue) is sampled for 5 minutes, 708 the primary cylinders are then each sampled for 3 minutes (light green) in order of increasing 709 concentration and then the target secondary cylinder (pink) is sampled for 3 minutes.We use the 710 last 90 seconds of a given 5-minute zero-air sample (light blue) to calculate 5 zero-air values 711 (blue squares).These 5 values are linearly interpolated (Interp.) to the sampling times and that 712 time series (blue trace) is then subtracted from to the raw mixing ratios (gray trace) to yield the 713 black trace (which appears nearly indistinguishable from the gray trace except in the case of 714 CO).The last 90 seconds (light green) of the 'zero-subtracted' data (black trace = gray trace -715 blue trace) are then averaged to generate a value for each of the 4 primary sampling intervals 716 (dark green squares).For each primary, the primary value at the QCLS sampling times is 717 linearly interpolated to those 4 values (green lines).The last 90 seconds of the target sampling 718 window (pink) are then interpolated (both linearly and quadratically) to the green lines (either the 719 2 closest for linear interpolation or the closest 3 for quadratic interpolation) which bracket the 720 secondary concentration of interest, as shown in Figure5.721 722

Figure 10 :Figure 11 :Figure 12 :
The 1 Hz HIPPO I-V data comparison for QCLS-CO2 with OMS (left) and AO2 778 (middle) as well as the QCLS-DUAL CO comparison with the RAF VUV-CO.QCLS-DUAL comparisons to the onboard gas chromatographs PANTHER (A,B,C) 785 and UCATS (D,E,F) for CH 4 (A,D), N 2 O (B,E), and CO (C,F).The UCATS instrument had 786 issues with the chromatography during HIPPO 2 and is not shown.The HIPPO 4 measurements 787 of N 2 O from UCATS were also excluded because of non-linear instrument response during 788 QCLS comparisons to NOAA flask data during HIPPO I-V for CO 2 (A), CH 4 (B), 792 N 2 O (C), and CO (D).With the exception of N 2 O which has a much tighter correlation with the 793 flask measurements, the axes are all scaled to the same ranges as Figure 11.The biases for each 794 fit are reported in

Table 1
138 summarizes the Allan precisions at 1, 10, and 100 seconds for the 4 species.During flight 139 sampling, the Allan precision between 1 and 10 seconds improves for all species, but only 140 continues to improve between 10 and 100 seconds for N 2 O.This is largely because atmospheric 141 variability in CO 2 , CH 4 , and CO is larger relative to N 2 O as the atmospheric lifetime of N 2 O is 142 ~118 years (Hsu and Prather, 2010) and the sources are more spatially uniform than for the other 143 species.Because of this, Table 1 also includes the Allan precision from laboratory tests that 144 sampled air continuously flowing from calibration cylinders with near-ambient atmospheric 145 concentrations.146 within the different sampling volumes are second order effects, but are minimized by using 6.35 186 mm OD and 9.525 mm OD Synflex for QCLS-CO2 and QCLS-DUAL respectively.The larger 187 diameter tubing is needed for QCLS-DUAL because of the larger sample cell volume.The flow 188 rates through QCLS-CO2 and QCLS-DUAL are 0.1 and 1.0 slpm, respectively, which 189 correspond to cell flushing times on the order of 1 sec for both sensors, assuming plug flow.
183 Overall instrument response time is largely controlled by the sample cell pressure and 184 volume, the flow rate, and the inlet pressure and volume.Additional lags associated with mixing 185 2) 315where time and pressure have units of sec and mbar.The dynamic range of ambient pressure is 316 much smaller in CalNex and does not include a HIMIL, which affects the pressure at the inlet, so 317 the equilibration time for CalNex is treated as a constant value derived from plume comparisons 318 between QCLS and a fast-response black-carbon measurement (Schwarz et al., 2010) that was 319 available for both HIPPO and CalNex.Equations for QCLS-CO2 have different coefficients but 320 the same form.The equations were calculated empirically during several test flights on each 321 campaign and then held constant throughout each campaign.322 The HIMIL port, designed to slow air-flow, complicated the instrument equilibration 323 time but dampened the input pressure variability of the sample.For CalNex, however, the 324 variability in the sample pressure was occasionally not adequately controlled by the pressure 325 control elements.Certain fluctuations in pressure were able to propagate to the QCLS-DUAL 326 sample cell and affect the measurements.The effect of this cell 'ringing' was most apparent in 327 the N 2 O measurement which occasionally showed high-frequency (1 Hz) positive and negative 328 excursions of >1-2 ppb for N 2 O, a trace that should only see negative excursions in stratospheric 329 air.We apply a filter which removed measurements in which the 1Hz rate change of pressure is 330 greater than 3 standard deviations of the mean (σ =0.16 hPa/sec).This resulted in an effective 331 duty cycle that was 3% lower than without the pressure filter, but removed spurious spikes in the 332 data.333 Calibration time intervals were determined using these functions and the solenoid valve 334 actuation time, and a mean mixing ratio for each sample addition was calculated in a given 335 window.The zero-air values measured every 15 minutes were then fit using a penalized Akima 336 spline interpolation technique (Akima, 1970) to evaluate the drift of the instrumentation.Other 337 filters, such as loess, splines, interpolators occasionally cause severe curvature in the 338 interpolation, particularly near the beginning of flight where sensors may not be fully 339 equilibrated.This zero-air Akima-spline is evaluated at all the 1 Hz sampling times and 340 al. (2002).The GSE is a Licor model 6251 NDIR analyzer, which 394 measures molecular absorption of CO 2 in a sample stream relative to a reference stream of air.395 Because it is a nondispersive analyzer, the measurement is sensitive to different parts of the 396 molecular absorption band of CO 2 .Tohjima et al. (2009) characterized the sensitivity of 3 Licors 397 (two 6252 and one 6262) to each of the isotopologues of CO 2 .They use a Relative Molar 398 Response (RMR) value for each isotopologue to calculate the effective change in concentration 399 determined for each isotopologue (see their Table 4).Given a hypothetical CO 2 mixing ratio of 400 400 ppm, the isotopic abundances in HITRAN (Rothman et al.,2007) can be used to approximate 401 the individual mixing ratios of the three dominant isotopologue -16 O 12 C 16 O, 16 O 13 C 16 O, and 402 16 O 12 C 18 O -as 393.68160, 4.42296 and 1.57883 ppm, respectively.The sum of these three 403 concentrations is less than 400 (399.68339) as other minor isotopes contribute to the total 404 concentration.Atmospheric CO 2 has an approximate isotopic composition of δ 13 C = -10 ‰ and 405 δ 18 O = 40 ‰, where these quantities are calculated according to: 406 the ratio of 13 C to 12 C in a sample of CO 2 or in the standard Vienna Pee 409 Dee Belemnite (vpdb = 0.011180) and R18 represents the ratio of 18 O to 16 O in CO 2 or in Standard Mean Ocean Water (smow = 0.0020052).Using these equations, we can calculate 411 atmospheric values for R13 and R18 of 0.0110682 and 0.002085408, respectively.The 412 abundance of the dominant isotopologue ( 12 C 16 O 16 O) must therefore be 1 minus the R13 and 413 twice the R18 abundances, or 0.984761, which corresponds to a concentration of 393.592606.
429To account for the combined effect on the QCLS-CO2 calibration, the -0.059 ppm and 430 the 0.1598 values must be added to the retrieved sample mixing ratio.The -0.059 ppm puts the 431 calibration cylinder values calculated using the GSE onto the same isotopic scale as the NOAA 432 primaries (i.e.atmospheric isotopic composition).The 0.1598 value accounts for the fact that 433 QCLS-CO2 derives a total mixing ratio using the absorption spectrum of the dominant 12 C 16 O 2 434 isotopologue and the HITRAN abundance, which differs from the atmospheric abundance as 435 shown above.These effects partially offset, but result in a ~0.1 ppm bias term, which is 436 important considering that atmospheric concentration gradients are often not much larger than 437 this.The particular isotopic values of the calibration cylinders (Table 2b) were measured by the 438 Stable Isotope Ratio Facility for Environmental Research (SIRFER) and were used to calculate 439 (Gerbig et al, 1999)n Facility (RAF) vacuum ultraviolet (VUV) CO sensor is the 465 only other fast-response instrument measuring one of the QCLS species(Gerbig et al, 1999).466Thatcomparison,alsoshown in Figure10, shows a bias of 1.8 ppb over the HIPPO mission.467 Two onboard gas chromatographs-the Unmanned Aircraft Systems (UAS) 468 Chromatograph for Atmospheric Trace Species (UCATS, Moore et al., 2003; Fahey et al., 2006; 469 Wofsy et al., 2011) and the PAN and other Trace Hydrohalocarbon ExpeRiment (PANTHER; 470 Elkins et al., 2002; Wofsy et al., 2011)-measured a variety of chemical species including CH 4 , 471 N 2 O, and CO. Figure 11 shows the one-to-one comparison of the QCLS to PANTHER (top) and 472 UCATS (bottom) after applying the averaging kernel of each GC to the 1 Hz QCLS data.In 473 addition to the in situ data, sparser flask measurements from the NOAA Whole Air Sampler 474 (NWAS) are compared in Figure 12.The axis ranges on Figures 11 and 12 are the same, with the 475 exception of N 2 O, which has large variability from the GC-based measurements.Table 4 476 summarizes the median differences with NOAA for each of the QCLS species at the mean 477 concentration measured on each of the 5 HIPPO transects.Mean biases calculated over the 478 course of HIPPO are -112, 0.85, 1.07, and -1.94 ppb for the 4 species.Only N 2 O falls outside of 479 the estimated uncertainties in the measurements.This is in part due to the recalibration of 480 primary cylinder 4 (Table 2a) that deviated from the original value by more than 4 times the 1σ 481 NOAA calibration uncertainty.This cylinder falls on the high range of the NOAA N 2 O uncertainty, with a standard deviation of 5.1 ppb.The cause of the CH 4 500 measurement discrepancy has remained a mystery despite extensive efforts to explain the 501 difference.These biases correspond to errors of 0.01% and 0.25 % for CO 2 and CH 4 , 502 respectively, using background concentrations of 390 ppm and 1800 ppb.The bias between the 503 independent CO sensors was 1.1 ppb during CalNex.It should be noted that the NOAA VUV 504 CO sensor did not dry the ambient air during measurement and reported wet mole fractions.505 Dilution therefore accounts for some of the bias.The in-flight NOAA VUV CO measurements 506 were calibrated by means of standard additions traceable to NIST with backgrounds determined 507 by catalytically scrubbing CO from the ambient air sample.508 To minimize data gaps in the 1 Hz flight data over the missions, we fit a loess curve with 509 a 1000 second span window to calculate the time-evolution of the QCLS minus 510 OMS/CRDS/VUV concentration bias.The QCLS data is used as the primary data, and 511 calibration gaps are filled using the sum of the OMS/CRDS/VUV data and the loess bias curve 512 (CO2.X in HIPPO, CO2.X and CH4.X in CalNex, and CO.X in both HIPPO and CalNex).This 513 resulted in an overall mission data retrieval duty cycle of over 95% for HIPPO and 97% for 514 CalNex, a significant improvement over the ~78% duty cycle from QCLS alone.These merge 515 products are denoted CO2.X, CH4.X, and CO.X.A merge product for N 2 O was not created 516 because no other fast-response N 2 O sensors were available for either mission.CH 4 , N 2 O, and CO relative to background concentrations of 390 ppm, 1850 ppb, 522 325 ppb, and 100 ppb, respectively, by adequately regulating pressure and temperature and by 523 using a robust in-flight calibration procedure that improves upon spectroscopically-calibrated 524 measurements.We report long-term compatibility for CO 2 , CH 4 , N 2 O, and CO from nearly 450 525 flight hours of 100, 1, 1.1, and 2 ppb, respectively.The datasets generated using the QCLS for 526 HIPPO and CalNex have provided extensive global (HIPPO) and regional (CalNex) coverage 527 and have been useful in many studies to date (Graven et al.,2013; Wunch et al., 2010; Kort et al., by NSF as a core instrument on NCAR's Gulfstream V aircraft.We would like 535 to thank all the pilots, aircraft technicians, and support staff of the NCAR HIAPER-GV and 536 NOAA P-3 as well as the many NOAA and NCAR collaborators who made the CalNex and 537 HIPPO measurements possible.This work was supported by the following grants to Harvard

Table 4
The 1 Hz CalNex data comparison for QCLS with the NOAA/Picarro CRDS for CO 2 802 (left) and CH 4 (middle) as well as the comparison with the NOAA VUV sensor for CO (right).

Table 1 :
Allan precision as a function of averaging time for the 4 QCLS species measured 806 during the in-flight sampling of a relatively constant air mass on HIPPO II, October 22 nd , 2009 807 ('flight') and during laboratory testing sampling continuously from a secondary calibration 808 cylinder ('lab').Accuracy estimates are based on the accuracy of the NOAA primary cylinders, 809 where accuracy in this context is an estimate of how well the scale can be transferred to different 810 instruments or laboratories at near-ambient mole fractions.

Table 2a :
Summary of the primary calibration cylinders used during the CalNex and HIPPO 815 campaigns for QCLS-DUAL.The primary cylinders were filled and calibrated at NOAA in 816 2005, then recalibrated again after CalNex and before HIPPO IV in 2011.The difference 817 between the two calibrations is shown for each tank and each species.

Table 3 :
Summary of the HIPPO and CalNex flight dates, duration, and locations.829

Table 5 :
Biases between QCLS and MEDUSA flask measurements at the reported mean 837 concentrations of CO 2 for the five HIPPO campaigns.838 839