Tunable diode laser in-situ CH 4 measurements aboard the CARIBIC passenger aircraft: instrument performance assessment

. A laser spectrometer for automated monthly measurements of methane (CH 4 ) mixing ratios aboard the CARIBIC passenger aircraft is presented. The instrument is based on a commercial Fast Greenhouse Gas Analyser (FGGA, Los Gatos Res.), which was adapted to meet the requirements imposed by unattended airborne operation. It was characterised in the laboratory with respect to instrument stability, precision, cross sensitivity to H 2 O, and accuracy. For airborne operation, a calibration strategy is described that utilises CH 4 measurements obtained from ﬂask samples taken during the same ﬂights. The precision of airborne measurements is 2 ppb for 10 s averages. The accuracy at aircraft cruising altitude is 3.85 ppb. During aircraft ascent and descent, where no ﬂask samples were obtained, instrumental drifts can be less accurately determined and the uncertainty is estimated to be 12.4 ppb. A linear humidity bias correction was applied to the CH 4 measurements, which was most important in the lower troposphere. On average, the correction bias was around 6.5 ppb at an altitude of 2 km, and negligible at cruising ﬂight level. Observations from 103 long-distance ﬂights are presented that span a large part of the northern hemispheric upper troposphere and lowermost stratosphere (UT/LMS), with occasional crossing of the tropics on ﬂights to southern Africa. These accurate data mark the largest UT/LMS in-situ CH 4 dataset worldwide. An example of a tracer-tracer correlation study with ozone is given, highlight-ing the possibility for accurate cross-tropopause transport analyses.


Introduction
Atmospheric methane (CH 4 ) is the second-strongest longlived anthropogenically influenced greenhouse gas (GHG) after carbon dioxide (CO 2 ) (Solomon et al., 2007). It is most active via its radiative forcing in the upper troposphere and lowermost stratosphere (UT/LMS) (Riese et al., 2012). CH 4 mostly originates from biogenic sources, e.g. wetlands, rice agriculture, biomass burning and ruminant animals (Dlugokencky et al., 2009;Bloom et al., 2010). Anthropogenic CH 4 sources, which account for ∼ 60 % of the total source strength, include various industrial processes, e.g. fossil fuel mining and distribution. The emission rate is highly variable (Heimann, 2011) and in particular the future emission rates of wetlands, permafrost and oceanic methane hydrates are highly uncertain (Heimann, 2010;O'Connor et al., 2010).
Global measurements of CH 4 thus serve various purposes: (1) to better constrain the global CH 4 budget and atmospheric trend, (2) to better quantify the different CH 4 sources and assess their future evolution, but (3) also as a transport tracer of tropospheric air to study exchange processes in the UT/LMS. Satellite measurements of CH 4 have been performed with various satellite-borne instruments in recent years (Schneising et al., 2009;Payan et al., 2009;Xiong et al., 2010;Wecht et al., 2012;Worden et al., 2012). While these data are provided on a global scale, they cannot resolve small-scale variability of CH 4 in the UT/LMS. To study these processes, in-situ measurements with high spatial resolution are inevitable.
Published by Copernicus Publications on behalf of the European Geosciences Union. 744 C. Dyroff et al.: In-situ CH 4 measurements aboard CARIBIC CH 4 has been measured in-situ during various airborne field campaigns (see e.g. Collins et al., 1993;Spackman et al., 2007;Chen et al., 2010;O'Shea et al., 2013). Regular CH 4 in-situ measurements in the UT/LMS, however, have not been performed yet. Until today, only flask samples were regularly collected aboard aircraft and later analysed in the laboratory for their CH 4 mixing ratio in the framework of the CONTRAIL (Machida et al., 2008) and CARIBIC projects (Brenninkmeijer et al., 2007;Schuck et al., 2012, see also: www.caribic-atmospheric.com). In CARIBIC (Civil Aircraft for the Regular Investigation of the atmosphere Based on an Instrument Container), 15 such air samples per flight were obtained between 2004 and 2009. After a major modification and extension of the instrumentation of the CARIBIC container, altogether 116 air samples were collected during four consecutive long-distance flights per month. The CH 4 mixing ratio was determined by laboratory measurements with high accuracy after each flight sequence (Schuck et al., 2009).
The spatial resolution of these flask data is still modest, and continuous in-situ measurements of the CH 4 mixing ratio are highly valuable in order to better capture the atmospheric variability with a resolution of about a kilometre. Diode-laser absorption spectroscopy offers the capability for precise and accurate measurements at small instrument size, and several research instruments have been developed in the past. They have provided in-situ measurements in the laboratory (Weibring et al., 2010), field (Werle and Kormann, 2001;Nelson et al., 2004), and aboard balloon and aircraft platforms (Scott et al., 1999;Richard et al., 2002;Durry et al., 2002;Gurlit et al., 2005;Berman et al., 2012). In recent years, commercial instruments have become available for laboratory or ground-based field measurements (Crosson, 2008;Chen et al., 2010;Tuzson et al., 2010). However, for the fully automated application aboard civil aircraft, these instruments need to be strongly modified to fulfill the strict safety requirements for unattended operation and to reliably work under strong temperature variations of more than 20 K.
In this paper we present an airborne diode-laser spectrometer, which is based on a commercial Fast Greenhouse Gas Analyser (FGGA, Los Gatos Research). The modifications towards unattended employment aboard passenger aircraft are described. Laboratory tests were performed to determine the spectrometer precision, cross sensitivity to H 2 O, and accuracy. A calibration procedure based on the aforementioned flask sample measurements was developed and is presented in detail. The precision and accuracy of airborne in-situ CH 4 measurements during 103 intercontinental flights is analysed and cumulates in a total error estimate. Some illustrative examples of the observations are presented.

Instrument setup
The present instrument is based on a commercial Fast Greenhouse Gas Analyser (FGGA 1 , Los Gatos Research), which measures CH 4 , CO 2 , and H 2 O mixing ratios based on offaxis integrated cavity output spectroscopy (OA-ICOS) (Baer et al., 2002). We decided for this technique mainly because it provides a rather simple and very rugged optical system, which is an important aspect for unattended airborne employment at conditions of vibration and temperature changes.
In the FGGA, two lasers are used to separately measure CH 4 and H 2 O as well as CO 2 in an interleaved fashion. In the present paper, we focus on the CH 4 measurements, as in the CARIBIC project CO 2 is measured with superior precision and accuracy by another in-situ instrument (LI-COR 6252, Lincoln, Nebrasca, USA).
The OA-ICOS technique employed in this instrument is reviewed briefly. The beam of a fibre-pigtailed near infrared (NIR) tunable diode laser emitting around ν 0 = 6057.5 cm −1 (λ = 1651 nm) is collimated and guided into an optical cavity formed by two mirrors with reflectivity R ∼ 0.99992. The beam enters the cavity through the front mirror in an offaxis alignment. In contrast to on-axis alignment, where only the fundamental cavity mode is excited, this approach excites many cavity modes and yields a quasi-continuous transmission spectrum, i.e. the cavity is assumed non-resonant (Sayres et al., 2009). Residual cavity modes are minimised by dithering the distance between the cavity mirrors using piezo electric actuators connected to the front mirror.
Due to the high mirror reflectivity, the residence time τ RD of the photons traversing at the speed of light c inside the cavity, and thus the effective optical pathlength l eff is long compared to single or multipass absorption cells. In our particular case we obtain τ RD ∼ 10 µs on average, which corresponds to l eff = 3 km in a L = 25 cm long cavity, during which interaction of photons and CH 4 molecules can occur . (1) The laser radiation leaking out of the rear cavity mirror is collected and focused on a room-temperature InGaAs photodiode with a ∅ = 2 mm active area via a ∅ = 50 mm biconvex lens. Employing laser-wavelength scanning by modulating the laser-injection current, absorption spectra ( Fig. 2) are obtained in ∼ 5 ms and upon averaging are evaluated at 1 Hz with a duty cycle of 0.5. The spectral scan includes two isolated CH 4 absorption features and one H 2 O absorption line. The main CH 4 absorption feature (ν 0 ≈ 6057.1 cm −1 , S ij = 1.2 × 10 −21 cm molec. −1 ) is an unresolved superposition of four individual lines, and the weaker feature (ν 0 ≈ 6057.5 cm −1 , S ij = 4.5×10 −23 cm molec. −1 ) comprises two unresolved lines (Rothman et al., 2009). The H 2 O line (ν 0 ≈ 6057.8 cm −1 , S ij = 7.4 × 10 −26 cm molec. −1 ) is rather weak and allows for humidity measurements with relatively low precision. For our airborne instrument, we use the data processing unit provided with the commercial FGGA. This unit performs all necessary tasks, such as scanning the laser wavelength, acquiring the spectra, and performing a spectral fit to obtain the CH 4 mixing ratio. Fit results along with the cavity pressure, gas temperature, and the ringdown time τ RD are transferred via RS232 serial connection to the housekeeping computer for storage. Unfortunately, with this model version of the FGGA it is not possible to store the measured spectra for post processing or archiving. This important feature may be possible with later FGGA versions, though.
For the employment onboard aircraft, the original instrument was modified to be operated unattended and to fulfill the safety requirements imposed by civil aviation. All parts of the instrument are mounted inside a lightweight, aircraft certified 19-inch rack (enviscope, Germany). A schematic of the instrument is depicted in Fig. 1.
A custom developed power-supply unit comprising commercial DC/DC converters (Mini series DC/DC, Vicor) was implemented to reduce size and weight. For noise-critical components, such as the photodetector and laser driver unit, the supply voltages are filtered to suppress noise and ripple. Before and after instrument modification, the noise characteristics of the Indium Gallium Arsenide (InGaAs) photodiode and subsequent transimpedance amplifier (G = 3 × 10 6 V/W, f = 100 kHz) were studied. The detectordark noise was measured to be 1.74 × 10 −12 W (Hz) −1/2 , i.e. practically identical to the value achieved with the original power supply. This dark noise is the dominating noise contribution to the signal-to-noise ratio in OA-ICOS (Dyroff, 2011).
A control computer (V25, Max Planck Institute Mainz, Germany) was installed for housekeeping tasks such as pressure and temperature control.
In the CARIBIC Airbus 340-600 aircraft, an inlet system is permanently installed at the lower fuselage in front of the wing section (Brenninkmeijer et al., 2007). Air is sampled through a sideways facing PFA tube with 12 mm inner diameter at a flow rate of ∼ 80 vol-L min −1 (volumetric flow rate). The major part of this flow is bypassing the instruments for fast flushing of the inlet line, thus maximising the response to atmospheric variability (see Sect. 4.5).
A flow of 4 sLpm (sLpm = standard liters per minute) is picked off of the main flow and is guided through the FGGA using a piston pump (8006ZV DC, Gardner Denver Thomas GmbH, Germany) located downstream of the OA-ICOS cavity (Fig. 1). A 1 L buffer volume downstream of the cavity minimises pressure fluctuations in the cavity due to the pump strokes. Upstream of the cavity, a proportional valve (Fluid Automation Systems GmbH, model EQI-FIL) is used to establish a constant pressure of 180±0.05 hPa (FGGA standard setting) within the cavity. The flow is measured using a flow meter downstream of the cavity.
In contrast to the commercial FGGA device, the OA-ICOS cavity and laser coupling optics are thermally insulated by highly efficient insulation material (model va-Q-vip, va-Qtec, Germany) and are temperature controlled to 40 • C using resistive heaters in combination with software PID controllers of the V25 computer.
The instrument is remotely controlled via the CARIBIC container master computer, which sets it into measurement mode when the aircraft is above the 750 hPa pressure level (2 km standard altitude).

Laboratory performance
Prior to the implementation into the CARIBIC payload as well as periodically between flights, the instrument performance was checked in the laboratory. Tests included those for instrument precision, calibration with known CH 4 standards, and determination of the cross sensitivity of the CH 4     This procedure was repeated for 60 min. The measurement sequence was started with a calibration of the FGGA with the calibration routine of the commercial FGGA using standard # 1. For each 2-min sample measurement the average (CH 4 ) sample was calculated. The average mixing ratio derived for standard # 1 is 1982.9 ± 2.1 ppb, and for standard # 2 it is 1787.8 ± 2.1 ppb. Over the 60 min period we thus were able to measure the two standards with an uncertainty of 6.5 ppb (< 0.5%) using the single-point calibration provided by the FGGA software.

300
Two observations were made when comparing the individual averages (CH 4 ) sample and their respective standard (1) Instrumental drift is the largest source of uncertainty for this measurement, as the FMA measurements drift to-305 wards lower mixing ratios [Figure 4 (b)]. (2) Standard # 2 is generally slightly more underestimated by the FGGA. This suggests that the calibration of the FGGA with a single gas standard leads to additional uncertainty at mixing ratios largely different to the calibration value. Thus a multiple-310 point calibration scheme is required for enhanced accuracy. measurements to the water vapour mixing ratio. The respective methods and results obtained are discussed below.

Precision
In order to determine the precision of the modified FGGA measurements of constant CH 4 mixing ratio from a certified gas standard (Basi Gase, Rastatt, Germany) were performed. This gas standard contained a dry synthetic air mixture with CH 4 = 1986.8 ± 1.16 ppb [NOAA04 Scale, 1 ppb = 1 nmol mol −1 , Dlugokencky et al. (2005); Schuck et al. (2009)]. At this mixing ratio, the fractional absorbance is around 11 % or 3.6 × 10 −7 cm −1 when normalised to the effective optical pathlength of 3 km inside the cavity. The CH 4 mixing ratio was confirmed using gas chromatography at the Max-Planck Institute in Mainz.
The gas standard was provided to the FGGA in a continuous flow, and the CH 4 mixing ratio was obtained at 1 Hz sampling frequency. From the corresponding CH 4 timeseries (Fig. 3a) the Allan variance σ 2 Allan (Fig. 3b) was obtained by calculating the variance of the differences between adjacent measurements of averaging time τ (Werle et al., 2004;Werle, 2011). With increasing averaging time τ , σ 2 Allan , which is an indicator of measurement precision, is reduced. For statistically independent measurements (constant noise spectral density, white noise) one expects a reduction of σ 2 Allan proportional to τ −1 as indicated by the theoretical dashed line. The measured Allan plot deviates from this line after 6 s, but it is still improved by averaging for up to τ opt ∼ 80 s. When averaging for longer than 80 s, low frequency disturbances (instrumental drift) become dominant, and σ 2 Allan is increasing again. Hence, an improvement of precision is not possible by averaging any longer than τ opt . For τ = 1 s averaging we obtain from the Allan cilloscope showing the CH 4 and H 2 O absorption lines probed. The shaded area indicates the time where the ringdown-time constant τ RD is determined.   This procedure was repeated for 60 min. The measurement sequence was started with a calibration of the FGGA with the calibration routine of the commercial FGGA using standard # 1. For each 2-min sample measurement the average (CH 4 ) sample was calculated.  plot a precision of σ Allan = 2.65 ppb, which corresponds to a fractional absorbance of 5.1 × 10 −10 cm −1 . The FGGA inflight data are reported as 10 s averages, where the precision in the laboratory is determined to be σ Allan = 0.96 ppb (absorbance 1.9 × 10 −10 cm −1 ). The best precision is found to be σ Allan = 0.56 ppb (absorbance 1.07 × 10 −10 cm −1 ) at τ opt ∼ 80 s.
The results obtained in the laboratory are very similar to what we have determined with the unmodified instrument, where the precision was around 2.2 ppb for 1-sec average data. Using an unmodified Fast Methane Analyser (FMA, Los Gatos Research), which relies on the same measurement principle as the FGGA, Tuzson et al. (2010) have determined a precision of ∼ 1 ppb for 1 s and ∼ 0.14 ppb for 10 s averaging. This is somewhat better than the precision of the present instrument, while the stability time of this FMA was limited to ∼ 30 s. We attribute the somewhat better precision found by Tuzson et al. (2010) partially to the fact that our instrument takes CH 4 measurements with a duty cycle of 0.5 as also CO 2 measurements are performed (yet not evaluated) using a second laser. This duty cycle reduction should limit the precision by a factor of √ 2×. The longer stability time of the present instrument may be attributed to a better thermal stability of our modified FGGA. Note that in contrast to the original FGGA, we maintain the OA-ICOS cavity at a constant temperature.

Accuracy
The instrument accuracy is affected by multiple factors. A nonlinear or biased instrument function can lead to errors at mixing ratios largely different to the calibration mixing ratio. Instrumental drift can reduce the accuracy of measurements taken long after the instrument stability time derived in an Allan-variance analysis.
We have determined the accuracy of the FGGA in the laboratory by taking measurements of two different calibration standards (standard # 1: 1986.8±1.16 ppb, standard # 2: 1794.3±0.81 ppb). Each of the standards was provided to the inlet of the instrument for 2 min, after which a valve was used to rapidly switch to the other standard (Fig. 4a). This procedure was repeated for 60 min. The measurement sequence was started with a calibration of the FGGA with the calibration routine of the commercial FGGA using standard # 1. For each 2-min sample measurement the average (CH 4 ) sample was calculated.
The average mixing ratio derived for standard # 1 is 1982.9 ± 2.1 ppb, and for standard # 2 it is 1787.8 ± 2.1 ppb. Over the 60 min period we thus were able to measure the two standards with an uncertainty of 6.5 ppb (< 0.5 %) using the single-point calibration provided by the FGGA software.
Two observations were made when comparing the individual averages (CH 4 ) sample and their respective standard ( (1) Instrumental drift is the largest source of uncertainty for this measurement, as the FMA measurements drift towards lower mixing ratios (Fig. 4b).
(2) Standard # 2 is generally slightly more underestimated by the FGGA. This suggests that the calibration of the FGGA with a single gas standard leads to additional uncertainty at mixing ratios largely different to the calibration value. Thus a multiple-point calibration scheme is required for enhanced accuracy.

Cross sensitivity to H 2 O
In the following, the cross sensitivity to water vapour (H 2 O) is investigated. H 2 O has strong absorbing bands even in the near infrared operation wavelength of the FGGA (λ = 1651 nm), and mixing ratios in the atmosphere that can vary by three orders of magnitude between the ground (up to 30 000 ppm, 1 ppm = 1 µmol mol −1 ) and the lower stratosphere (down to ∼ 3 ppm). Even if no direct spectroscopic interference would be present, molecules with large abundance can affect the spectral lineshape function of the target absorber by molecular collisions, which is being observed as a modified width of the lineshape function (Varghese and Hanson, 1984;Tuzson et al., 2010). The FGGA has a builtin humidity correction, and both wet and dry CH 4 data can be analysed. During initial testing of our FGGA model, we found that the built-in correction did not sufficiently account The average mixing ratio derived for standard # 1 is 1982.9 ± 2.1 ppb, and for standard # 2 it is 1787.8 ± 2.1 ppb. Over the 60 min period we thus were able to measure the two standards with an uncertainty of 6.5 ppb (< 0.5%) using the single-point calibration provided by the FGGA software.

300
Two observations were made when comparing the individual averages (CH 4 ) sample and their respective standard ( (1) Instrumental drift is the largest source of uncertainty for this measurement, as the FMA measurements drift to-305 wards lower mixing ratios [Figure 4 (b)].
(2) Standard # 2 is generally slightly more underestimated by the FGGA. This suggests that the calibration of the FGGA with a single gas standard leads to additional uncertainty at mixing ratios largely different to the calibration value. Thus a multiple-310 point calibration scheme is required for enhanced accuracy. for large humidity changes. In particular, we found a residual sensitivity of around 3 ppb CH 4 at a humidity of 10 000 ppm. While this is already very good, for best possible accuracy we decided to quantify the cross sensitivity in the laboratory in order to apply a correction to the wet airborne CH 4 measurements.
As performed in previous studies (Nara et al., 2012;Chen et al., 2013;Rella et al., 2013) we empirically determine a correction function by supplying a gas mixture of known CH 4 and varying H 2 O mixing ratios, where the latter spans the humidity range experienced in flight.
A main flow of a dry calibration-gas standard is provided to the inlet of the instrument. Additionally, a small flow of the same gas is humidified in a gas bubbler, and it is subsequently diluted into the main flow. The flow rate of the small flow is adjusted to stepwise achieve a number of humidity levels between dry gas and around 15 000 ppm.
The humidity is measured with relatively low precision by the FGGA probing a weak H 2 O absorption line in the vicinity of the CH 4 absorption line (Fig. 2). A dew-point hygrometer (Model DewMaster, EdgeTech, Marlborough, MA, USA) is operated in parallel and is used for humidity-calibration purposes.
In a first step of the analysis, the humidity measurements of the FGGA are calibrated using the dew-point hygrometer measurements by making use of a polynomial of second order (Fig. 5a). In the second step, the humidity dependence of the CH 4 measurements is quantified by a linear function, which is a sufficient approximation within the measurement uncertainty range (Fig. 5b).
The humidity-correction function was determined periodically in the laboratory. On average we found for this particular FGGA a correction function of The slope varied by ±10 %, which translates into an uncertainty of ±4.6 ppb (0.25 %) at CH 4 = 1900 ppb and a humidity of 10 000 ppm. Furthermore, the humidity measurements are associated with an uncertainty of ∼ 5 %, which translates in a 0.12 % uncertainty of the humidity corrected CH 4 data. At the aircraft cruising altitude (∼ 220 hPa, > 10 000 m), where around 93 % of the total measurement are obtained, the humidity is generally below 50 ppm. In this case the cross-sensitivity can be neglected, as the total uncertainty is defined by noise (precision) and general instrument accuracy. However, measurements are also obtained during aircraft ascent and descent, which allows the derivation of vertical profiles above the respective airports. In this case the correction function effectively compensates for the humidity bias.
While the FGGA provides H 2 O measurements itself, for the processing of our airborne data we use H 2 O measurements obtained by a dedicated instrument (Brenninkmeijer et al., 2007), which offers far superior precision and accuracy. In that respect, the FGGA H 2 O measurements serve as a redundant measurement in case the main H 2 O measurements may not be available.

Calibration
Instrument calibration during airborne operation is a prerequisite to achieve the necessary accuracy for valid data interpretation. The best strategy would be an in-flight calibration with multiple (2-3) calibration gas mixtures. These standards would have to be provided aboard the aircraft, which is impractical in CARIBIC due to space restrictions. Calibration in the laboratory is usually performed before each flight. However, due to necessary ground-tests and instrument/container transportation, several days may pass between instrument calibration and actual flight series. Generally, the last flight of a series takes place at least 5-6 days after the instrument leaves our laboratory. We have therefore developed a calibration routine utilising CH 4 flask-sample measurements obtained simultaneously during all CARIBIC flights.
During each flight 14-45 air samples are obtained with two different sampler systems. (1) TRAC: 28 2.7 L glass flasks are filled with compressed ambient air and later analysed for greenhouse gases (GHGs), non-methane carbon hydrates (NMCHs), and halocarbons. (2) HIRES: 88 1 L stainless steel flasks are filled with compressed ambient air and later analysed for GHGs and NMCHs. The analyses are performed with high accuracy in various laboratories (Schuck et al., 2009;Baker et al., 2010;O'Sullivan, 2007;Batenburg et al., 2012).
For the calibration, the FGGA CH 4 measurements are first averaged over the effective sampling times (0.5-4 min) during which the air samples were obtained. The CH 4 mixing ratios of each individual sample are then used to derive a piecewise linear calibration function for the FGGA CH 4 data between two consecutive flask samples. For FGGA measurements before the first and after the last air sample, the calibration is linearly extrapolated based on the first and last two air samples, respectively. Figure 6a shows as an example the CH 4 measurements of the flask samples and the respective FGGA data on a flight from Bogota to Frankfurt departing on 23 March 2011. Note that for this figure the FGGA in situ CH 4 data are corrected for slope and offset by an average calibration function of this particular flight (black dashed line), which has for this particular flight a slope of 0.9971 and an ordinate (bias) of 26.46 ppb. This depiction better illustrates the scattering of the CH 4 measurements with respect to the calibration function. It is also later being used for further data-quality analyses.
The FGGA was calibrated before the four flights in March 2011 in the laboratory. We attribute the relatively large offset to the fact that usually about one week passes between calibration and actual flight, including several instrument starts as part of ground tests. During the flight sequences we generally observed a flight-to-flight variability of the offset of 5-10 ppb, which occasionally reached 15 ppb. This changing offset further emphasises on the need for instrument calibration based on measurements taken during flight. Note that we do not consider zero-gas measurements, because the natural CH 4 variability (range) during every one flight generally provides a well enough defined calibration function. Figure 6b shows the raw CH 4 measurements (grey trace), the CH 4 measurements obtained from the flask samples (red crosses), as well as the calibrated in situ CH 4 measurements along the flight (blue trace). The altitude profile is also depicted (black trace).

Airborne uncertainty assessment
During unattended instrument deployment aboard aircraft, rather large changes in environmental conditions occur. For instance, the cabin pressure aboard the CARIBIC aircraft varies from 1000 hPa on the ground to around 815 hPa at cruising altitude, which may lead to subtle changes in the optical alignment of, for example, the ICOS cavity. Furthermore, the temperature in the container can be as low as 15 the calibration is linearly extrapolated based on the first and last two air samples, respectively. 26.46 ppb. This depiction better illustrates the scattering of the CH 4 measurements with respect to the calibration function. It is also later being used for further data-quality analyses.
The FGGA was calibrated before the four flights in March 425 2011 in the laboratory. We attribute the relatively large offset to the fact that usually about one week passes between calibration and actual flight, including several instrument starts as part of ground tests. During the flight sequences we generally observed a flight-to-flight variability of the offset of 430 5-10 ppb, which occasionally reached 15 ppb. This changing offset further emphasizes on the need for instrument calibration based on measurements taken during flight. Note that we do not consider zero-gas measurements, because the natural CH 4 variability (range) during every one flight generally 435 provides a well enough defined calibration function.
Figure 6 (b) shows the raw CH 4 measurements (grey trace), the CH 4 measurements obtained from the flask samples (red crosses), as well as the calibrated in situ CH 4 measurements along the flight (blue trace). The altitude profile is 440 also depicted (black trace).

Airborne uncertainty assessment
During unattended instrument deployment aboard aircraft, rather large changes in environmental conditions occur. For instance, the cabin pressure aboard the CARIBIC aircraft 445 varies from 1000 hPa on the ground to around 815 hPa at cruising altitude, which may lead to subtle changes in the optical alignment of e.g. the ICOS cavity. Furthermore, the temperature in the container can be as low as 15 • C and, in extreme cases, rises up to 40 • C. It is thus important to note 450 that the instrument performance may suffer under airborne conditions. Therefore we have performed a detailed uncertainty assessment under airborne conditions based on a statistical analysis of all FGGA flight data. 103 flights were performed 455 between October 2010 and July 2013. During this time 2780 CH 4 measurements were obtained from the simultaneously collected flask samples.

Replication precision
In the laboratory we have analyzed the instrument precision 460 by means of the Allan variance method. During airborne operation, this technique is not applicable due to additional, usually dominating atmospheric variability of CH 4 . We have thus investigated the short term precision of in-flight measurements by calculating the standard deviation of 10-sec av-465 erages. Figure 7 shows the corresponding probability density function (PDF) of the calculated standard deviations.
It is found that, on average, the airborne 1-σ precision of the instrument for 10-sec data averaging is 2 ppb. This is in good agreement (factor of two) with the Allan variance 470 depicted in Figure 3. About 90 % of our measurements show a precision of better than 4.5 ppb. Note however, that natural variability is likely to increase the standard deviation, even   Probability   Fig. 7. PDF of the standard deviation of CH4 measurements obtained airborne (10 s averages). The precision during our airborne measurements is better than 2 ppb for more than 50 % of the time, and better than 4.5 ppb for more than 90 % of the time.
The instrumental drift δ(CH 4 ) is approximated as the difference in the deviation d(CH 4 ) of consecutive measurements normalized by the elapsed time t We note that this method assumes linear drift during the time interval [t n−1 t n ]. This may not always be valid during flight, but is the only method applicable given the limited number of flask samples.

480
The observed deviation between FGGA and flask from one calibration measurement to the next is partly determined by the relative uncertainty (precision) of the measurements. However, it is mostly defined by instrumental drift between calibration measurements (15-45 min), which is longer than 485 the stability time of the instrument (80 s in the laboratory). and, in extreme cases, rises up to 40 • C. It is thus important to note that the instrument performance may suffer under airborne conditions. Therefore, we have performed a detailed uncertainty assessment under airborne conditions based on a statistical analysis of all FGGA flight data. The 103 flights were performed between October 2010 and July 2013. During this time 2780 CH 4 measurements were obtained from the simultaneously collected flask samples.

Replication precision
In the laboratory, we have analysed the instrument precision by means of the Allan variance method. During airborne operation, this technique is not applicable due to additional, usually dominating atmospheric variability of CH 4 . We have thus investigated the short term precision of in-flight measurements by calculating the standard deviation of 10 s averages. Figure 7 shows the corresponding probability density function (PDF) of the calculated standard deviations.
It is found that, on average, the airborne 1-σ precision of the instrument for 10 s data averaging is 2 ppb. This is in good agreement (factor of two) with the Allan variance depicted in Fig. 3. About 90 % of our measurements show a precision of better than 4.5 ppb. Note however, that natural variability is likely to increase the standard deviation, even on a 10 s time scale. This manifests itself in the somewhat wider tail of the PDF towards higher σ (CH 4 ). flight we find a mean calibration function with a slope 0.9971 and ordinate of 26.46 ppb. (b) Calibrated FGGA CH 4 measurements (solid blue trace, 10-sec averages) and simultaneously obtained flask sample measurements (red +) on flight LH336. The grey trace depicts the uncalibrated FGGA measurements. The black trace shows the altitude profile for this flight.  Probability   Fig. 7. PDF of the standard deviation of CH 4 measurements obtained airborne (10 s averages). The precision during our airborne measurements is better than 2 ppb for more than 50 % of the time, and better than 4.5 ppb for more than 90 % of the time. on a 10-sec time scale. This manifests itself in the somewhat wider tail of the PDF towards higher σ(CH 4 ).

Drift analysis
A statistical analysis of instrumental drift in flight conditions was performed. First the difference d(CH 4 ) between FGGA and flask is calculated for every coincident calibration mea-surement n d(CH 4 ) n = CH 4 (FGGA) n − CH 4 (flask) n . (4) The instrumental drift δ(CH 4 ) is approximated as the difference in the deviation d(CH 4 ) of consecutive measurements normalized by the elapsed time t We note that this method assumes linear drift during the time interval [t n−1 t n ]. This may not always be valid during flight, but is the only method applicable given the limited number of flask samples.

480
The observed deviation between FGGA and flask from one calibration measurement to the next is partly determined by the relative uncertainty (precision) of the measurements. However, it is mostly defined by instrumental drift between calibration measurements (15-45 min), which is longer than 485 the stability time of the instrument (80 s in the laboratory). Figure 8 shows the corresponding PDF of δCH 4 for all CARIBIC flights considered here. The average drift is zero, and the PDF is symmetric, which indicates a random effect. An instrumental drift of about 0.24 ppb/min (1σ) is derived 490 from the PDF.
To bring this finding into perspective, we can compare δCH 4 with the results obtained by the Allan-variance analysis. δCH 4 = 0.24ppb/min suggests that the instrument will have drifted by 2.4 ppb after 600 s assuming linear drift.

495
Comparing to the Allan plot (Figure 3), we deduce in the Fig. 7. PDF of the standard deviation of CH 4 measurements obtained airborne (10 s averages). The precision during our airborne measurements is better than 2 ppb for more than 50 % of the time, and better than 4.5 ppb for more than 90 % of the time.

Drift analysis
A statistical analysis of instrumental drift in flight conditions was performed. First the difference d(CH 4 ) between FGGA and flask is calculated for every coincident calibration measurement n d(CH 4 ) n = CH 4 (FGGA) n − CH 4 (flask) n . (4) The instrumental drift δ(CH 4 ) is approximated as the difference in the deviation d(CH 4 ) of consecutive measurements normalised by the elapsed time t δ(CH 4 ) i = d(CH 4 ) n − d(CH 4 ) n−1 t n − t n−1 .
We note that this method assumes linear drift during the time interval [t n−1 t n ]. This may not always be valid during flight, but is the only method applicable given the limited number of flask samples. The observed deviation between FGGA and flask from one calibration measurement to the next is partly determined by the relative uncertainty (precision) of the measurements. However, it is mostly defined by instrumental drift between calibration measurements (15-45 min), which is longer than the stability time of the instrument (80 s in the laboratory). Figure 8 shows the corresponding PDF of δCH 4 for all CARIBIC flights considered here. The average drift is zero, and the PDF is symmetric, which indicates a random effect. An instrumental drift of about 0.24 ppb min −1 (1σ ) is derived from the PDF.
To bring this finding into perspective, we can compare δCH 4 with the results obtained by the Allan-variance analysis. δCH 4 = 0.24 ppb min −1 suggests that the instrument will have drifted by 2.4 ppb after 600 s assuming linear drift. Comparing to the Allan plot (Fig. 3) are taken, the cabin pressure is rather constant, and we may rule out any pressure-induced drift. The average time between flask samples during all CARIBIC flights considered here was ∼ 16 ± 5 min, during which the instrument may have drifted by 4 ± 1.25 ppb. The 510 samples are collected according to a pre-determined time schedule and therefore variations in the flight schedule cannot be compensated. Generally no air samples are obtained during aircraft descent. The average measurement time before a flask sample was obtained after aircraft take off was 515 16.5 ± 3 min. During aircraft descent, measurements were taken for 47±18 min after the last flask sample was obtained. Again assuming linear drift we estimate the uncertainty of our CH 4 measurements at the beginning and the end of a flight to be 4.2 ± 0.75 ppb and 11.75 ± 4.5 ppb, respectively.

Accuracy
For the analysis of the instrument accuracy it is not possible to directly compare the flask-sample measurements and the corresponding corrected FGGA measurements. By definition these measurements are identical to the flask measure-525 ments after applying the calibration procedure described in Section 3.4. We have therefore used a slightly different approach to estimate the instrument accuracy. laboratory a drift of ∼ 1.2 ppb after the same time. This suggests, that under airborne conditions, the instrumental drift is ∼ 2 × larger than in the laboratory.
There are some occasions where the instrument was not yet properly warmed up when we started the measurement in flight. Then we sometimes observed what might be a direct correlation of temperature and d(CH 4 ). However, it is difficult to fully disentangle the processes that lead to changes in d(CH 4 ). At the cruising altitude, where most measurements are taken, the cabin pressure is rather constant, and we may rule out any pressure-induced drift.
The average time between flask samples during all CARIBIC flights considered here was ∼ 16 ± 5 min, during which the instrument may have drifted by 4 ± 1.25 ppb. The samples are collected according to a pre-determined time schedule and therefore variations in the flight schedule cannot be compensated. Generally no air samples are obtained during aircraft descent. The average measurement time before a flask sample was obtained after aircraft take off was 16.5 ± 3 min. During aircraft descent, measurements were taken for 47±18 min after the last flask sample was obtained. Again assuming linear drift we estimate the uncertainty of our CH 4 measurements at the beginning and the end of a flight to be 4.2 ± 0.75 ppb and 11.75 ± 4.5 ppb, respectively.

Accuracy
For the analysis of the instrument accuracy it is not possible to directly compare the flask-sample measurements and the corresponding corrected FGGA measurements. By definition these measurements are identical to the flask measurements after applying the calibration procedure described in Sect. 3.4. We have therefore used a slightly different approach to estimate the instrument accuracy.
Atmos. Meas. Tech., 7, 743-755  First we calculate a single linear calibration function for every flight i [ Figure 6 (a)]. This function is then applied to the FGGA measurements. The average slope of the cali-530 bration function is 0.947 ± 0.122 for all 103 flights. The accuracy of our measurements ∆(CH 4 ) is then expressed as the deviation of the linearly corrected individual FGGA measurements to their respective flask sample measurements The PDF of ∆(CH 4 ) of all 2780 calibration measurements is shown in Figure 9. On average (103 flights) we obtain a distribution that reveals an uncertainty of ∼ 3.4 ppb (1σ) at the aircraft cruising altitude where flask samples are obtained. 535 We note that this method of obtaining an accuracy estimate most likely overestimates the measurement uncertainty, since using a piece-wise calibration function for the reported flight data reduces deviations by linear interpolation between flask measurements.

Total uncertainty
The total uncertainty of the CH 4 measurements is determined by the relative uncertainty (precision, σ(CH 4 )), and the uncertainty of the absolute value (accuracy, ∆(CH 4 )). The former is dependent on the averaging time. We report in situ 545 data as 10 s averages, where the precision was determined to be σ(CH 4 ) = 2 ppb. The accuracy at flight level was determined to be ∆(CH 4 ) = 3.4 ppb, which includes the relative uncertainty and instrumental drift terms. During ascent and descent of the aircraft, no flask samples are obtained. 550 Fig. 9. PDF of the airborne accuracy expressed as deviation of the individual airborne calibration measurements from the linear calibration function for each flight. On average the accuracy is found to be CH 4 ∼ 3.4 ppb (1σ ) at the aircraft cruising altitude.
First we calculate a single linear calibration function for every flight i (Fig. 6a). This function is then applied to the FGGA measurements. The average slope of the calibration function is 0.947 ± 0.122 for all 103 flights. The accuracy of our measurements (CH 4 ) is then expressed as the deviation of the linearly corrected individual FGGA measurements to their respective flask sample measurements (CH 4 ) n = [a i CH 4 (FGGA) n + b i ] − CH 4 (flask) n .
The PDF of (CH 4 ) of all 2780 calibration measurements is shown in Fig. 9. On average (103 flights) we obtain a distribution that reveals an uncertainty of ∼ 3.4 ppb (1σ ) at the aircraft cruising altitude where flask samples are obtained. We note that this method of obtaining an accuracy estimate most likely overestimates the measurement uncertainty, since using a piece-wise calibration function for the reported flight data reduces deviations by linear interpolation between flask measurements.

Total uncertainty
The total uncertainty of the CH 4 measurements is determined by the relative uncertainty (precision, σ (CH 4 )), and the uncertainty of the absolute value (accuracy, (CH 4 )). The former is dependent on the averaging time. We report in-situ data as 10 s averages, where the precision was determined to be σ (CH 4 ) = 2 ppb. The accuracy at flight level was determined to be (CH 4 ) = 3.4 ppb, which includes the relative uncertainty and instrumental drift terms. During ascent and descent of the aircraft, no flask samples are obtained. Depending on the time to the closest calibration event, instrumental drift is estimated to reduce the accuracy by up to 10 C. Dyroff et al.: In-situ CH 4 measurements aboard CARIBIC Depending on the time to the closest calibration event, instrumental drift is estimated to reduce the accuracy by up to 11.75 ppb during the aircraft descent, i.e., 47 min after the last flask sample was obtained. Furthermore, the analysis of the flask samples is stated with an uncertainty of 1.8 ppb 555 (Schuck et al., 2009). From the individual uncertainties we can derive the total uncertainty of our in situ CH 4 measurements as the sum of the individual uncertainty terms. At flight level we achieve measurements with an uncertainty of 3.85 ppb (0.2 % for an 560 average CH 4 ∼ 1800 ppb). During aircraft ascent and descent, where instrumental drift becomes more relevant, we estimate the uncertainty to be 12.4 ppb (0.7 %) for measurements obtained at the lowest sampling altitude of 2 km.

565
The temporal resolution of the FGGA is inferred by comparing fast fluctuations in CH 4 with data recorded by the fast chemiluminescence ozone detector described in Zahn et al. (2012). At the tropopause and lowermost stratosphere (LMS) both species are highly anticorrelated as ozone in-570 creases but methane decreases with height above the midlatitude tropopause (see Sec. 5.2). Figure 10 shows a short-time cross-section of CH 4 and O 3 observed just above the chemical tropopause of O 3 ∼ 75 ppbv (see Zahn and Brenninkmeijer, 2003;Thouret et al.,  The CH 4 measurements are calibrated using the flask sample data as described in Section 3.4. The cross sensitivity of the CH 4 measurements to H 2 O is corrected by a linear correction function employing the humidity measurements also measured in the CARIBIC container. 605 Figure 11 shows the measurements obtained during the first 2.5 years of operation along flight routes between Frankfurt (Germany) to ten destinations in North/South America, South Africa, and Asia. The color code represents the CH 4 mixing ratio (60 s averages).

610
Methane is a medium-lived greenhouse gas with an atmospheric lifetime of ∼ 9 years (Dlugokencky et al., 2011) and thus shows modest variations by up to ∼ 20 % in the upper troposphere (UT), as seen in Figure 11. These variations are mainly due to the differing origin and nature of the air masses 615 encountered in the UT.
Presently, the globally averaged CH 4 mixing ratio is around 1800 ppb (Dlugokencky et al., 2011). The elevated CH 4 mixing ratios of up to 1950 ppb are indicative of polluted air masses and are thus often encountered downwind 620 of the continents, e.g. imbedded in warm conveyor belt systems starting e.g. in the gulf of Mexico and reaching the upper troposphere over the mid-Atlantic (Eckhardt et al., 2004). The weak seasonal variation of CH 4 of ∼ 30 ppb in the midlatitude UT (not shown) is not resolved in Figure 11 and leads 625 to some further variability.
The most drastic feature in Figure 11 is the detection of stratospheric air north of ∼ 35 • N where the CH 4 mixing ratio drops down to 1600 ppb in spring when CH 4 -poor air descends from the higher stratosphere within the downward 630 branch of the Brewer Dobson circulation.
Note the sometimes very sharp gradients in the CH 4 mixing ratio, especially when crossing the local tropopause, 11.75 ppb during the aircraft descent, i.e. 47 min after the last flask sample was obtained. Furthermore, the analysis of the flask samples is stated with an uncertainty of 1.8 ppb (Schuck et al., 2009).
From the individual uncertainties we can derive the total uncertainty of our in situ CH 4 measurements as the sum of the individual uncertainty terms. At flight level we achieve measurements with an uncertainty of 3.85 ppb (0.2 % for an average CH 4 ∼ 1800 ppb). During aircraft ascent and descent, where instrumental drift becomes more relevant, we estimate the uncertainty to be 12.4 ppb (0.7 %) for measurements obtained at the lowest sampling altitude of 2 km.

Response time
The temporal resolution of the FGGA is inferred by comparing fast fluctuations in CH 4 with data recorded by the fast chemiluminescence ozone detector described in Zahn et al. (2012). At the tropopause and lowermost stratosphere (LMS) both species are highly anticorrelated as ozone increases but methane decreases with height above the midlatitude tropopause (see Sect. 5.2). Figure 10 shows a short-time cross-section of CH 4 and O 3 observed just above the chemical tropopause of O 3 ∼ 75 ppbv (see Zahn and Brenninkmeijer, 2003;Thouret et al., 2006) on a flight from Caracas to Frankfurt. The 5 Hz ozone data indicate fast fluctuations in the 0.5-10 s range, most likely generated by gravity waves. The CH 4 data (blue) is recorded with 1 Hz, i.e. each point indicates 1 s or 250 m along the flight path. The compact anticorrelation of both species is nicely visible and little smoothing of the CH 4 data is resolved. Indeed, the volume sample flow of ∼ 22 vol-L min −1 (see Sect. 2) together with the volume of the cavity of ∼ 0.5 L suggests a time resolution of 1.4 s.

Near-global CH 4 measurements
The instrument described in the present paper has been deployed for four consecutive long-distance flights per month aboard the CARIBIC passenger aircraft since October 2010. Flights originate from Frankfurt, Germany (50 • N, 8 • E) to various destinations worldwide (Fig. 11). In total, CH 4 was measured in situ during 103 flights between October 2010 and July 2013. The flights took place mostly in the northern hemispheric upper troposphere and lowermost stratosphere (UT/LMS), and occasionally lead across the tropics to southern Africa. Note that the available flight destinations of the CARIBIC aircraft are determined by the semi-annual Lufthansa flight schedule for this particular aircraft, which limits the number of possible flight routes. The instrument operation will continue until at least 2014.
The CH 4 measurements are calibrated using the flask sample data as described in Sect. 3.4. The cross sensitivity of the CH 4 measurements to H 2 O is corrected by a linear correction function employing the humidity measurements also measured in the CARIBIC container. Figure 11 shows the measurements obtained during the first 2.5 yr of operation along flight routes between Frankfurt (Germany) to ten destinations in North/South America, South Africa, and Asia. The colour code represents the CH 4 mixing ratio (60 s averages).
Methane is a medium-lived greenhouse gas with an atmospheric lifetime of ∼ 9 yr (Dlugokencky et al., 2011) and thus shows modest variations by up to ∼ 20 % in the upper troposphere (UT), as seen in Fig. 11. These variations are mainly due to the differing origin and nature of the air masses encountered in the UT.
Presently, the globally averaged CH 4 mixing ratio is around 1800 ppb (Dlugokencky et al., 2011). The elevated CH 4 mixing ratios of up to 1950 ppb are indicative of polluted air masses and are thus often encountered downwind of the continents, e.g. imbedded in warm conveyor belt systems starting, for example, in the gulf of Mexico and reaching the upper troposphere over the mid-Atlantic (Eckhardt et al., 2004). The weak seasonal variation of CH 4 of ∼ 30 ppb in the mid-latitude UT (not shown) is not resolved in Fig. 11 and leads to some further variability.
The most drastic feature in Fig. 11 is the detection of stratospheric air north of ∼ 35 • N where the CH 4 mixing ratio drops down to 1600 ppb in spring when CH 4 -poor air descends from the higher stratosphere within the downward branch of the Brewer Dobson circulation.
Note the sometimes very sharp gradients in the CH 4 mixing ratio, especially when crossing the local tropopause, which cannot be captured by the flask samples taken on average each 225 km at cruising altitude.

Tracer-tracer correlation
One powerful test to assess the uncertainty of the CH 4 data during a flight is the correlation with another tracer that can very accurately be measured and that is controlled by the same atmospheric process. In CARIBIC, ozone (O 3 )  that is mainly determined by the uncertainty in spectroscopic data), but with a reproducibility of better than 0.5 %. In spring when O 3 -rich and CH 4 -poor high stratospheric air descends in mid-and high-latitudes, both O 3 and CH 4 constitute very long-lived (basically inert) transport tracers in the lower stratosphere and are virtually unaffected by variations in the troposphere. And indeed, as shown in Fig. 12, O 3 and CH 4 are compactly correlated above the tropopause (above O 3 ∼ 120 ppb) during this flight. There is a weak influence due to the mixing with upper tropospheric air within the first 1.0-1.5 km above the tropopause (up to O 3 = 300 ppb). Above this level and up to 5 km above the tropopause (O 3 = 1000 ppb) the correlation is very tight. These stratospheric data have been recorded over a time of ∼ 4.5 h or ∼ 3500 km.

Conclusions
The setup, characterisation, and airborne deployment of a laser spectrometer to measure CH 4 mixing ratios in situ aboard a passenger aircraft have been described. The airborne instrument is based on a commercial fast methane analyser (FGGA, Los Gatos Research), which has been largely modified to meet aircraft requirements and to be operated autonomously.
Laboratory characterisation by means of the Allan variance method has revealed a precision of ∼ 0.96 ppb and 0.6 ppb for an averaging time of 10 s and 80 s, respectively. The CH 4 data can be averaged for as long as 80 s after which instrumental drift starts to become relevant, and the precision is not further improved. The in-flight precision is found to be 2 ppb (1 σ ), which is in agreement with the laboratory results within a factor of 2.
Using the standard single-point calibration suggested for the FGGA, an uncertainty of ∼ 6.5 ppb has been determined in the laboratory using known calibration gas standards. Besides instrumental drift, the instrument accuracy is affected by a cross-sensitivity of the CH 4 measurements to the H 2 O mixing ratio. This cross sensitivity has been quantified in the laboratory to be around 4.6 ppb when the humidity reaches lower tropospheric mixing ratios (10 000 ppm). A linear correction function is applied to reduce this humidity bias.
As a means to calibrate the in situ CH 4 measurements, a calibration method has been developed that utilises CH 4 measurements from flask samples obtained during the same flights. Furthermore, we make use of in situ H 2 O measurements in order to apply a humidity-bias correction. With this calibration method we achieved a total uncertainty of 3.85 ppb (1σ ) for measurements obtained at the aircraft cruising altitude. During ascent and descent of the aircraft, where no flask samples are obtained, instrumental drift affects the measurements stronger. Then the total uncertainty at lower altitudes (∼ 2 km) is estimated to be ∼ 12.4 ppb (1σ ).
Airborne measurements in a large part of the northern hemispheric UT/LMS, and to some extend in the Southern Hemisphere, were obtained during the first two and a half years of monthly operation. The data quality and the spatial resolution are well sufficient to study e.g. small-scale air mass transport across the tropopause.