A compact QCL spectrometer for mobile, high-precision methane sensing aboard drones

sensing aboard drones Béla Tuzson1, Manuel Graf1, Jonas Ravelid1, Philipp Scheidegger1, André Kupferschmid2, Herbert Looser1, Randulph Paulo Morales1, and Lukas Emmenegger1 1Laboratory for Air Pollution / Environmental Technology, Empa Swiss Federal Laboratory for Materials Science and Technology, Überlandstrasse 129, Dübendorf 8600, Switzerland 2Transport at Nanoscale Interfaces, Empa Swiss Federal Laboratory for Materials Science and Technology, Überlandstrasse 129, Dübendorf 8600, Switzerland Correspondence: Béla Tuzson (bela.tuzson@empa.ch)


Introduction
Global emissions of methane are continuing to increase, making CH 4 a relevant component for managing realistic pathways to mitigate climate change, especially with near-term benefits (Shindell et al., 2012). However, the accurate quantification 15 of the CH 4 budget and its variations still remain challenging, mainly due to uncertainties in CH 4 emissions from natural wetlands and fresh waters, and not sufficiently constrained partitioning of CH 4 emissions by region and processes (Kirschke et al.;Saunois et al., 2016). Therefore, there is a continuing need for technologies that enable the tracking and identification of methane sources at local scale and on a regular basis. In this context, lightweight and high-precision trace gas sensors aboard small unmanned aerial vehicles (UAVs) offer new opportunities for in-situ air pollution and emission monitoring (Golston 20 et al., 2017). Combined with fast response, they can provide valuable information about the spatial and temporal variability of emissions regardless of terrain complexity or pollution source characteristics. This significantly extends the scale and the 1 level of detail of measurements compared to those provided by traditional stationary monitoring networks. The potential of this approach is well reflected by the increasing number of reported commercial as well as research-grade sensors and related applications (e.g. Berman et al., 2012;Villa et al., 2016, and references therein). The majority of these solutions are triggered by 25 oil-and gas industry fugitive emission monitoring (Nathan et al., 2015;Yang et al., 2018;Golston et al., 2018;Fox et al., 2019;Martinez et al., 2020) and, correspondingly, are targeting elevated CH 4 concentrations near emission sources. It is, however, more challenging to develop a sensor for the investigation of natural emissions (e.g. O'Connor et al., 2010;DelSontro et al., 2010), which are typically more diffuse and much weaker. Here, we target these more demanding, but equally important application areas. Our aim is to drastically reduce the size of conventional lab-based laser spectrometers to an almost hand-30 held device, while maintaining the high precision required for atmospheric methane concentration monitoring. In the following sections, we describe our approach, present the hardware solution, and show the capabilities of the developed CH 4 sensor.

Spectral selection
For high sensitivity methane detection using infrared absorption spectroscopy, there are mainly two suitable spectral regions 35 corresponding to the strong fundamental absorption bands of C-H stretching (ν 3 at 3.3 µm) and bending (ν 4 at 7.7 µm) modes.
Although, the stretching mode contains transition lines of up to five times higher intensity compared to the bending band, we still opted for the latter spectral range. This was motivated by two practical factors: i) distributed feedback quantum cascade lasers (DFB-QCL) with high optical power can readily be used, and ii) the P -branch of the ν 4 -band is at the edge of the atmospheric window, i.e. interfering absorption due to water vapor is considerably weaker. This is particularly important in 40 the context of an open-path configuration, i.e. measurements at ambient pressure. Additionally, it is worthwhile to mention that the symmetric stretching vibration band (ν 1 ) of nitrous oxide (N 2 O) is also located in the vicinity and may be used at a later stage as a secondary tracer. Figure 1 shows the simulated transmittance of ambient air at atmospheric pressure for the spectral range covering the P -and Q-branch of the ν 4 -band of CH 4 . It might be tempting to select one of the strong absorption feature resulting from many overlapping transitions in the Q-branch at 7.6 µm. However, given the narrow tuning range of 45 a DFB laser source, this would make it very difficult to define the intensity baseline for a reliable spectral fitting. This is a consequence of the open-path configuration and the inherent atmospheric pressure broadening effect, which leads to large absorption features that eventually exceed the tuning range of the laser. In fact, the uncertainty of the baseline retrieval in the open-path system limits the performance in the Q-branch, especially when considering other influencing factors such as interferences from pressure-broadened, overlapping absorption features of other atmospheric species and changes in ambient 50 temperature and pressure.
A systematic survey of the 7.7 µm spectral region revealed an attractive window in the vicinity of a water absorption line at 1276.62 cm −1 that exhibits a distinctly narrow profile even at ambient pressure (see inset Fig. 1). According to the HI-TRAN2016 database (Gordon et al., 2017), this absorption is given by the completely overlapping high-J transitions, 15 0 15  Figure 1. Simulated transmission spectrum of ambient air at atmospheric pressure over an optical path length of 10 m. Only the contributing molecular species, e.g. H2O, CH4 and N2O, with their mean atmospheric abundance are considered. The inset shows the selected spectral window for methane measurements. The water absorption line is markedly narrow, due to a particularly small pressure broadening coefficient of this transition. ten times smaller than for transitions of low-J states. This can be explained by inefficient collisional relaxation due to the wide energy separation (∼300 cm −1 ) for high-J states of H 2 O (Eng et al., 1972;Giesen et al., 1992). Considering the typical energies of translation motion (kT ≈200 cm −1 ), it is obviously difficult to take up the transition energies from such rotational states. Therefore, the perturbation of these high rotational energy states by collision remains low, which results in low pressure broadening, so that the H 2 O absorption line at ambient pressure appears as narrow as it would be at 0.1 atm. This aspect has a 60 significant benefit for the spectral analysis: as water is omnipresent, there is always a clear absorption feature that we can use for exact frequency determination, providing a robust anchor for frequency locking by using its spectral position as a control parameter for an active feedback loop on the laser heat sink.

Instrumental design
The instrumental design is strongly driven by the targeted application with special focus on compactness, ruggedness, low 65 power consumption and weight. These aspects support an open-path approach, where the absorption cell is directly exposed to the atmosphere at ambient pressure, eliminating the need for gas handling and a sampling pump. This considerably reduces weight and power consumption, leads to fast response, and assures low contamination. For high-precision trace gas detection, however, it is necessary to use a multipass cell (MPC) that folds the light path to achieve a sufficiently large signal-to-noise ratio (SNR). This key element often limits miniaturization of laser spectrometers, mainly because of its bulky design. Therefore, 70 we fundamentally reconsidered the MPC concept and recently developed a novel and versatile solution, which is unique in its combination of size, mechanical rigidity and optical stability. A full description of this design, called segmented circular Figure 2. Photograph of the CH4 laser spectrometer with and without the protecting cover (left). The main building blocks are the TO3packaged QCL installed on a three-axis stage with additional TEC stabilized heat sink, the SC-MPC with 10 m effective path length, the MCT detector, also with TEC stabilized heat sink, and five PCBs incorporating the hardware electronics: an icw laser driver, a DAQ unit with extension board, a power-supply module, the OEM TEC-board of the detector, and a custom TEC-board for laser and detector heat sinks.
The heat sinks of the laser and detector are, in addition, actively air cooled with fans. Picture of the CH4 sensor installed on the hexacopter (Matrice 600, DJI). multipass cell (SC-MPC), is given by Graf et al. (2018). Here, we only recall the main characteristics of the SC-MPC. It consists of a monolithic aluminum ring with a radius of 77.25 mm, containing sixty-five quadratic, spherically curved segments seamlessly shaped into the ring's inner surface. This geometry supports up to 10 m optical path length at a total mass of less 75 than 200 g. The SC-MPC has a confocal configuration, and it can thus directly accept the collimated laser beam emitted by the DFB-QCL (Alpes Lasers, Switzerland) encapsulated into a TO-3 housing with embedded collimating optics and thermoelectric cooler (TEC). Furthermore, the cell is largely insensitive to the aiming and coupling of the laser beam, which allows using a sturdy, custom-built three-axis linear stage for the laser source assembly. This module contains an additional TEC connected to a heat sink, which is cooled by forced convection using a fan. After undergoing 65 reflection, the light is focused onto the to the opposite side of the ring cell (see Fig. 2). The thermal management of the detector is implemented in the same fashion as for the laser, because of the elevated sensitivity of the detector signal to its heat sink temperature (see below). The thermal stability of the detector is crucial for low-drift operation. A 2-inch solid-Ge etalon, mounted on a periscope-like mechanical 85 support, can be placed into the MPC to determine the laser tuning-rate. The opto-mechanical setup is thus realized without any beam steering and shaping optics and has an overall weight of 750 g.
Of similar importance are the controlling and driving hardware as well as the data acquisition system. Here, we adopted our intermittent continuous wave (icw) laser driving concept (Fischer et al., 2014) and developed a current-source based laser driver optimized for a small footprint and a high temperature stability. The design relies on the concept recently published by 90 Liu et al. (2018), and features three main elements: i) an active current control, provided by an operational-amplifier based current regulator, ii) a high-precision analog pulse generator to generate current pulses with an adjustable amplitude, shape, and width; and iii) a capacitor that acts as a low-noise power supply, being disconnected from the external source during the operation of the QCL. The whole circuit was optimized for low noise and low temperature sensitivity.
The laser was driven by current pulses of 88 µs duration at 55 % duty cycle, corresponding to 6000 scans per second.

95
Hardware control and DAQ functions are provided by an integrated System-on-Chip (SoC, STEMlab 125-14, Red Pitaya, Slovenia) open-source platform featuring an integrated dual-core ARM Cortex A9 processor with Xilinx 7-series FPGA logic.
The FPGA firmware (VHDL-Code) and the Linux service routine (C-Code) were custom developed. The SoC GPIOs provide digital triggers required for operating the QCL driver. Further functionalities are realized by a custom-developed extension board with additional ADCs (16-and 24-bits) and DACs. These are used to set and read several temperature values for spectral 100 and diagnostic purposes. The hardware internal communication is based on the Inter-Integrated Circuit (I 2 C) protocol. The temperature control of the QCL relies on a high precision thermoelectric cooler (TEC) controller (WTC3243, Wavelength Electronics, USA), whilst the additional heat sinks (laser and detector) are stabilized by a custom-developed PWM based controller. The FPGA contains a state-machine that is clocked eight times slower than the sample clock (125 MS/s), which makes routing less critical, i.e. larger data path delays are tolerated. For higher flexibility in data acquisition, several user 105 defined time-windows within a spectral scan are supported. By means of decimation (up to 8), the total acquisition time can be increased, while the surplus samples can be dropped or averaged. The summation of consecutive spectra is implemented as DSP-Adders and Dual-Port-BRAMs. Port A of the BRAMs is for storing the new sum of the spectra, while port B is for reading the last sum. The spectra are then transferred from the programming logic (PL) via AXI-FIFO, DMA and DDR-RAM to the processing system (PS). The DMA is configured from a thread started by the Linux service routine. This thread polls 110 the IRQ-bit of the DMA status and retriggers the DMA after each interrupt to copy the spectrum to a new destination address within a ring buffer. Whenever a client requests a spectrum, the current data is read from the ring buffer and sent to the client via TCP/IP. Two independent instances of the Linux service routine are started with their own TCP/IP port to assure that the writing and reading of additional signals (e.g. analog and digital inputs) do not interfere with the fast query of spectra.
The detector signal is digitized by the RP's on-board ADC with 14-bit resolution and a sampling rate of up to 125 MS/s, 115 while the FPGA performs the real-time and on-board averaging of the data-stream as described above. These data, together with environmental and system log-data, are continuously saved on a USB flash-memory and, optionally, a copy is sent to a host-PC via TCP/IP using wired or wireless connection. Environmental parameters, such as barometric pressure, relative humidity, and ambient temperature are provided by an integrated PHT combination sensor (MS8607, Measurement Specialties Inc., USA) situated at the lower edge of the multipass cell (see Fig. 2). Finally, a GPS module (Adafruit Industries, USA) for 120 altitude and position retrieval rounds-up the fully autonomous sensor, capable of delivering all the necessary parameters for mobile laser spectroscopy applications. The modular elements are fixed on a mechanically rugged 3D-printed polycarbonate platform with an embedded battery case. A lightweight cover assures protection and insulation of the device (Fig. 2). The overall instrument weights 2.1 kg including the LiPo battery (18 V) and has an outdoor average electrical power consumption of 18 W. For the flight deployments, the device is mounted beneath a hexacopter (Matrice 600, DJI, China) on a bay plate using a gimbal frame with anti-vibration rubber dampers (see Fig. 2). The maximum flight time with the 2 kg payload is about 20 min.
During this period, a wireless bi-directional data link (SkyHopper PRO, Israel) assures real-time access to the raw spectral data and all hardware-related parameters. This allows for real-time spectral fitting and logging, and the user has full control over the entire hardware settings and can continuously monitor status of the instrument.

130
The spectral analysis takes place on the averaged data of 6000 spectra corresponding to 1 Hz time resolution. A typical transmission spectrum with the associated fitted curves using Voigt profiles is shown in Fig. 3. The spectral line intensity and the broadening parameters are taken from the HITRAN2016 database (Gordon et al., 2017), whereas the gas pressure and temperature are measured.

Characterisation and Validation
The controlled characterization of any open-path spectrometer under representative flight conditions is highly challenging.
Therefore, we custom-built a small (60 l volume) climate-chamber to have full control over the parameters of interest, such as ambient pressure, temperature and methane mole fraction (given in units of parts-per-million or billion (ppm or ppb) throughout this paper, representing the number of moles of methane per moles of air). Equipped with a high-power TEC-assembly (with  The precision and stability of the spectrometer was assessed within the chamber under continuous purging with air at 2 l min −1 from a pressurized air cylinder. During this test, the pressure and the temperature were maintained constant, while the fan was running at full speed. Figure 4a summarizes the results of this measurement. The Allan-Werle variance technique (Werle et al.,145 1993) applied to the data collected over one hour shows that the device achieves a precision of 1.1 ppb within one second, corresponding to an absorption noise level of 5.8 × 10 −6 . The sensitivity can be further improved by averaging over at least another 100 s. After this time, the two-sample variance, deviates from the expected 1/τ line, which indicates that drifts start to dominate the system. Nevertheless, the measurement precision stays bellow 0.5 pbb for 20 min, corresponding to the duration of a common drone flight. Repeated precision tests indicate a robust stability of the system. The inset in Fig. 4 shows the The histogram is slightly skewed to the right with the highest count located at 1.2 ppb. It shows that the 1 s noise level is highly reproducible and adds confidence that the 1.5 h Allan-Werle deviation plot shown in Fig. 4 is representative for the performance of the instrument.

155
The chamber was also used to determine the response of the spectrometer to CH 4 concentration changes. For this purpose, a certified calibration gas with high CH 4 concentration (200 ppm ± 1 %, PanGas, Switzerland) was dynamically diluted with dry nitrogen (N 2 ) in a step-wise fashion using mass flow controllers (see Fig. 4b). The derived linear calibration slope was then used throughout the field campaigns. A critical and limiting factor for trace gas laser spectroscopy is the appearance of unwanted etalon effects, generally referred 160 to as fringing. This effect is induced by multiple reflections from partially reflecting plane-parallel surfaces (windows, focusing optics) interfering with the main beam. As they differ in phase due to the different path lengths traversed, they can efficiently translate optical frequency into transmitted intensity. This inherent transfer function may exhibit temporal variations due to mechanical vibrations or changes in ambient temperature or pressure. This is especially critical under field conditions, i.e.
during flight of the spectrometer, as it tends to impair the measurement. To investigate the impact of such influences, the 165 instrument was repeatedly exposed to sudden pressure and temperature variations generated in the climate chamber. The span of these changes corresponds to situations expected during realistic flight scenarios, e.g., assuming variations in flight altitude up to 500 m relative to ground, related temperature gradients as well as diurnal temperature changes of up to 10°C. The results of these experiments are summarized in Fig. 5.
It was found that pressure variations do not influence the performance, as the real-time on-board pressure measurements 170 can be used to fully account for the respective impact on the spectral data, i.e. changes in absorption line width (pressure broadening) as well as changes in the number density of molecules (ideal gas law). In contrast, variations in temperature had a significant effect on the retrieved CH 4 mole fraction. This is mainly attributed to changes in the opto-mechanical system. In particular, we found that changes in the retrieved CH 4 mole fraction were linearly correlated (23 ppb K −1 ) with the temperature of the detector module. Therefore, we implemented a TEC-based solution to maintain its heat sink temperature at a constant 175 value of 32 ± 0.05°C even under varying environmental conditions. Further influencing factor, but less significant in magnitude, was the effect of dynamic fringe structures. The broad line-profile of the CH 4 absorption line was susceptible to slight shifts in the laser emission frequency, which caused subtle baseline variations. Given their complex behaviour, it is not possible to fully account for these fringes, which, therefore, may bias the retrieved concentrations. Our strategy to minimize their impact was to implement an active frequency locking scheme using the water absorption line position as a reference. Thereby, the peak 180 position is continuously determined by real-time spectral fitting and compared to the initial value. Whenever the difference is larger than ±2 × 10 −4 cm −1 , the laser heat sink temperature is repeatedly adjusted in small steps (mK) with a latency time of 10 s to compensate this drift. By implementing all the above optimisation strategies, the effect of sudden temperature changes on the retrieved CH 4 concentration was reduced to about 4 ppb K −1 , as indicated on Fig. 5. These perturbations are rather of random nature, i.e. difficult to account for, and most likely reflect the susceptibility of the entire electronics to abrupt 185 temperature changes. Nevertheless, considering the typical flight scenarios (e.g. emission screening), the temperature changes in such cases are only a few degrees, and correspondingly, the instabilities in the CH 4 concentration are expected to stay bellow 10 ppb. Nevertheless, for high-altitude flights, such as PBL determination, it is definitely advisable to adopt a more elaborated thermal stabilization scheme. In principle, there are two ways to further improve the temperature stability of the instrument: active (e.g. TEC-based) and passive. While the former can be very compact, it requires significant electrical power. A passive 190 isolation although can efficiently dampen sudden temperature fluctuations, it has limited flexibility for the range of supported temperatures. Thus, during cold season a good isolation can be beneficial, while on hot summer days can lead to thermal rollover. Lately, we were successfully applied a combination of TEC and phase-change material (as thermal buffer medium) solution, but with considerable cost of weight and electrical power (Graf et al., 2020). In this case, the instrument demonstrated temperature stabilization in a range of nearly 80°C. Similar approach could be implemented also for CH 4 spectrometer, when 195 the target application requires the highest precision and accuracy.
Finally, we also tested the influence of mechanical vibrations and air turbulences generated by an operating drone. For this purpose, the instrument was installed on the hexacopter, which was then turned on and off repeatedly to generate typical vibrations and turbulence. It became immediately clear that a direct mechanical contact with the drone-body propagates the rotor vibrations onto the spectrometer, generating significant optical noise. Therefore, a gimbal frame with anti-vibration 200 rubber dampers was added beneath the hexacopter and the CH 4 spectrometer was then firmly mounted on this platform.
Followup measurements showed no effect on the retrieved methane mole fractions within the measurement precision and natural variability, proving the efficiency of the gimbal system.

Field deployments
Prior to flying aboard a drone, the spectrometer was installed and operated continuously outside in the vicinity of a monitor-  Figure 6. Intercomparison of the CH4 spectrometer with a reference methane analyzer at the NABEL monitoring station. The device was placed outside, fully exposed to changing environmental conditions, in the proximity of the air inlet located at 2 m above ground. The cessfully captured sudden concentration variations in atmospheric methane. The observed spikes were reoccurring regularly in the evening hours, but their origin/source is unknown to us. However, they offered a good opportunity to assess the dynamic response and sensitivity of our CH 4 instrument. The observed differences between the two analyzers (see middle plot in Fig. 6) 215 are mainly due to the unsynchronized clock's and the difference in response times as well as in sampling strategies, i.e. openand closed-path.

Flying above an artificial source
For the first test flight we used a small grassland area located nearby the Empa campus. Here, we installed an artificial CH 4 source and released pure methane at a constant flow-rate of 10 l min −1 . To simulate the concentration field of methane and 220 calculate the footprint emission map, the GRAL dispersion model (Oettl et al., 2002) was used. This model computes highresolution wind fields and turbulence in the presence of 3D obstacles (buildings and vegetation) using a computational fluid  dynamics (CFD) approach. The dispersion of pollutants is then computed in a Lagrangian framework by releasing virtual particles at prescribed sources (Anfossi et al., 2006). The forcing for the simulation is obtained from flow fields computed by averaging locally observed meteorological data. In this simulation, meteorological data during the flight were obtained by 225 placing a 3D sonic anemometer METEK GmbH,Germany) in the middle of the field. The anemometer, mounted on a 5 m high mast, sampled the wind direction, ambient temperature, and wind speeds in the x, y, and z directions at 12 50 Hz. Meteorological data from the anemometer were then averaged to regular 5 min intervals to obtain turbulent parameters such as Obukhov length, friction velocity, and wind speed fluctuations (u , v , and w ), and served as input to compute the flow fields. Simulated concentration field of methane at 5 min intervals were finally averaged to obtain a footprint emission map of 230 the whole modelling domain.
The CH 4 spectrometer was operated prior the flight for 20 minutes on the ground, including warm-up time, recording an etalon spectrum, setting up the spectral fit, initiating the data recording and the wireless communication. Afterwards, the drone was flown at heights above ground ranging from 2 to 15 m for about 20 min in the proximity of the artificial methane source (see Figure 7). The time-series recorded along the trajectory intersecting the plume exhibit large spikes reflecting methane 235 enhancements from the release, which is in good qualitative agreement with the GRAL model, which predicted methane concentrations enhancements of up to 10 ppm at 5 m above the ground.

Atmospheric profiling
A challenging but important application for drone-based CH 4 -sensors is to interrogate the atmospheric boundary layer (ABL) frequently and at multiple locations to provide feedback for ABL model developments. Therefore, we conducted a field test 240 to assess the capability of our CH 4 spectrometer to detect small, natural inhomogeneities in the atmospheric methane concentrations at different altitudes throughout the ABL. These field experiments were conducted at Beromünster, which is located in a moderately hilly environment at the southern border of the Swiss Plateau. Here, a former radio tower (47°11 23 N, 8°10 32 E, 212.5 m tall, base at 797 m a.s.l.) is equipped with air inlets at 12.5, 44.6, 71.5, 131.6 and 212.5 m, respectively (see Fig. 8). A cavity ring down spectrometer (G2401, Picarro Inc., USA) monitors the mole fractions of CH 4 at these five 245 different heights sequentially, sampling each height for 3 min (Satar et al., 2016). The field-test took place on a day with stable atmospheric conditions, when methane accumulates during nighttime in a shallow nocturnal boundary layer near the ground. Figure 8 shows the CH 4 time series provided by the CRDS analyzer for the selected day. Consistent with our expectations, the data indicate a pronounced gradient among the inlet heights during the time-period before sunrise. This concentration build-up is then slowly disappearing after sunrise due to break-up of the nocturnal boundary layer by convective vertical mixing. The 250 sensor's precision was adequate to easily resolve the vertical gradients of atmospheric methane. The CH 4 mole fractions typically varied between 1940 and 2200 ppb, with some exceptional cases, when sudden increase of CH 4 mole fractions of up to 3200 ppb were observed close to ground over time periods of tens of minutes. These short-term spikes were attributed to pollution events, due to emissions from the farmsteads (ruminants) in the close vicinity (200 m) of the tower.

255
A compact and lightweight laser spectrometer capable of atmospheric CH 4 measurements at high sensitivity and time resolution (1 Hz) was developed to be deployed aboard UAVs. The open-path system has an overall weight of 2.1 kg (including battery) and an electrical power consumption of 18 W. It can easily be carried by commercial drones, providing unique opportunities for mobile sensing and spatial mapping of CH 4 mole fractions at unprecedented precision. The fully autonomous device uses a QC laser emitting in the mid-infrared at 7.83 µm, a versatile and robust segmented circular mirror multipass 260 cell (SC-MPC) with 10 m effective optical path length, and a custom-developed low-power SoC FPGA based data acquisition system. The instrument was characterized in the laboratory in terms of precision, linearity, and its dependence on various environmental factors, and then validated under field conditions and flown in two different application scenarios. With a precision of 1 ppbv CH 4 at 1 Hz, the spectrometer represents the very first portable instrument of its class regarding precision, weight, and real-time analysis capabilities. It outperforms existing solutions based on near-and mid-infrared spectroscopy using CRDS 265 (Berman et al., 2012;Martinez et al., 2020) and wavelength modulation (Golston et al., 2017)

280
Data availability. The data used in this manuscript are available from the corresponding author upon request.
Competing interests. The authors declare that they have no conflict of interest.