1. Introduction

Carbon dioxide (CO₂) is the basis for photosynthesis and thus also the basis for life on Earth [1]. Apart from the natural abundance or background level, currently ∼0.04% [2], CO₂ is a trace gas and indicator for all metabolism (~4% concentration) [3], and a major product in all hydrocarbon combustion processes (typical concentrations on the order of 10%), e.g., from anthropogenic activity [4]. Carbon dioxide traces can display fine details in space and time, such as the features exploited in the natural phenomenon of host seeking of the world’s deadliest animal [5]. Technological applications include, for example, search and rescue [6] and the CO₂-soot particle ratio can be used to evaluate combustion quality from vehicle emissions [7]. In respect to the Earth’s energy budget [8,9], CO₂ is criticized for its long-waved absorption bands, reabsorbing thermal radiation bound for the universe [8,10]. At the same time, absorption bands are also the only regions in the electromagnetic spectrum where gases are black, have an emissivity, and thereby the ability to radiate heat from the Earth [11]. From this it is understood that not only the integrated vertical CO₂ column, but also the profile governs the thermal photon transport in the atmosphere. Although CO₂ is heavier than air it can display distinct vertical distributions [12–14] and a step increase over the planetary boundary layer [15,16].

The CO₂ absorption spectrum contains a number of infrared absorption bands, which are increasing in line strength towards the long-waved infrared regime. Despite the increase in vibrational band intensities with increasing wavelength, the associated decrease in photon energy implies noisy detection, requires complicated cooled sensors and results in low atmospheric backscattering. Several passive approaches for monitoring CO₂ have been invented. On a coarse and slow scale, satellite based spectroscopy provides global coverage of the CO₂ column [17–20], and atmospheric limb occultation provides coarse estimation of vertical profiles [21]. Ground based stand-off detection of CO₂ on fine scales can be accomplished real time by gas correlation imaging [22], and other non-dispersive methods, such as Non-Dispersive InfraRed (NDIR) [23] or scanning Fourier Transform InfraRed (FTIR) imaging [24]. The FTIR technique is spectrally broadband and can thus capture multiple molecular species simultaneously, but relying on the thermal emission at CO₂ bands has its limitations. Active laser-based methods have been developed both for space- and air-borne instruments [25–29]. These methods are derivatives of Tunable Diode Laser Absorption Spectroscopy (TDLAS) [30], also known as Integrated Path Differential Absorption (IPDA). The IPDA technique relies on hard target reflectance from ground [29] or clouds [16]. A common feature for all of the aforementioned techniques is that the ambition is somewhat lowered; i.e. to provide an integrated CO₂ concentration rather than a range-resolved concentration profile.

One method that does provide a range-resolved profile is DIfferential Absorption Lidar (DIAL). Atmospheric Lidar (Light Detection And Ranging) have sufficient sensitivity to retrieve a distributed backscattered echo from molecules and particles in the air, and range is normally determined by the round trip time-of-flight (ToF) delay. In DIAL the lidar alternates between firing laser pulses tuned on and off an absorption line (λ_on and λ_off) of the atmospheric species of interest. Thus, one echo is attenuated by scattering and absorption, while the other is solely attenuated by scattering. By dividing the two echoes, all instrument responses and the impact of present atmospheric conditions cancel out, and the molecular concentration can be found as the range derivative of the natural logarithm of the measured on-off-ratio, divided by twice the difference in absorption cross section for the two wavelengths. As opposed to other methods, such as FTIR, TDLAS and IPDA, then DIAL does not compare signals towards a baseline or source intensity. In integrating methods, such as FTIR and TDLAS, source stability and interference fringes in the instrumentation are major concerns. Contrary to this, DIAL compares intensity over range and is, in other words, self-referencing.

In theory, DIAL represents an elegant way of gas sensing and profiling. Nevertheless, in practice there are only a handful operational DIAL systems worldwide [31–33], most of them being bulky truck- or container-sized systems based on high-power solid-state lasers, although some progress have been made with laser diodes and micro-lidar [34–36]. Most of the obstacles for widespread DIAL usage can be attributed to the cumulative required features of the laser source, as it must be: a) pulsed (~10ns), b) tunable (Δλ~1nm), c) narrowband (δλ~1pm), and d) providing high peak power (~100MW). Several of these features are in direct conflict with each other. Whereas DIAL measurements undertaken in the UV regime [32] provides high backscattering, the challenges escalate when pursuing CO₂ monitoring towards the IR region. The atmospheric echo is weaker in the infrared [37], the competition with background radiation from the surrounding environment is more challenging, photons are less energetic, which implies higher noise, and commonly employed cascade detectors (PhotoMultiplier Tubes or Avalanche Photo Diodes) are easily saturated by the strong background radiation. Only a few DIAL studies in the infrared wavelength regime are reported which are mainly based on high-power pulsed lasers and ranging through ToF detection [14,38]. At the same time, CW laser sources (fiber-, disk-, and gas-based) in the IR regime are widely available, even with high enough power for laser machining, and many of them are also tunable and narrowband.

The Scheimpflug-lidar method allows ranging with CW lasers [39–42], thus circumventing the aforementioned challenges with pulsed laser sources. Atmospheric lidar with inexpensive CW laser diodes in the Short Wave InfraRed (SWIR) has been demonstrated [43], e.g. profiling of O₂ and NO₂ using CW-DIAL systems [44,45]. In the present work, we demonstrate, for the first time, range-resolved CW-DIAL in the SWIR regime for CO₂ profiling using a high-power tunable CW-seeded Erbium fiber laser and an InGaAs array detector implemented into a lidar system configured in Scheimpflug geometry. Successful measurements of CO₂ concentrations were carried out in the atmosphere along a 2-km laser beam path directed towards the sky with a small upward inclination angle.

2. Methodology and Instrumentation

The atmospheric CO₂ concentration was monitored from 17:00 (UTC + 1) on the 7th of January 2018 until 08:00 (UTC + 1) on the 8th of January 2018. The experiment was carried out from a laboratory at the Department of Physics (55°42′37.1”N 13°12′17.1”E) at Lund University, Sweden. The transmitted laser beam, defining the test range, propagated into the atmosphere at an azimuth of 34.3° and an upward inclination at 4.3°, as illustrated in Fig. 1. It was a clear sky and the temperature, logged by a weather station with a thermometer located outside the laboratory, had a mean value of 0.2°C (ranging from a minimum of −1.3°C to a maximum of 14°C) during the experiment.

 

Fig. 1 Map and schematic drawing of the test range for the CO₂-DIAL measurements.

Download Full Size | PDF

2.1 Instrument

The experimental setup, schematically depicted in Fig. 2, is based on the Scheimpflug lidar principle [39,40,46] and configured for DIAL measurements [44,45]. The transmitting and receiving optics are mounted on a motorized equatorial tripod mount (50 kg payload capacity, SkyWatcher Eq. (8). Transmitter and receiver are mounted on a rectangular aluminum baseline and separated by 814 mm. The optical path starts with a narrowband single-mode distributed feedback diode laser (1573 nm, 20 mW, QDFBLD-1580-20, QPhotonics). The emission wavelength is tuned coarsely by a temperature controller, while spectral scanning is accomplished through current control. The light then seeds an Er/Yb-doped fiber amplifier (SM-EYDF-6/125-HE, NUFERN) which is pumped by a 4.5-W CW multi-mode diode laser (Lumics, LU0975T090), emitting at 975 nm [47]. In order to prevent back reflections, low- and high power isolators are connected before and after the fiber amplifier. The output emission is narrowband, 3 MHz linewidth (FWHM), with a tuning range of 2 nm in the vicinity of its center wavelength at 1.572 µm. The output power was measured to be a maximum of 1.3 W. The amplifier fiber tip is coupled to the beam expander and transmitted by a ø75 mm achromatic SWIR coated lens (Edmund Optics, 47319). The beam divergence can be controlled by an astronomical focuser (Crayford Monorail, AstroShop) and the expander is mounted on a tangential mount (Baader Stronghold, AstroShop), allowing optimization of the overlap between the laser beam and the field-of-view.

 

Fig. 2 Schematic illustration of the Lidar instrument. A single-mode laser diode is amplified by an Erbium fiber amplifier, expanded and transmitted into the atmosphere. Atmospheric backscattering is collected by a Newtonian telescope and detected on a linear InGaAs array implemented in Scheimpflug configuration. The strobe of the sensor modulates the amplifier pump and controls the seed wavelength.

Download Full Size | PDF

The backscattered light from the illuminated air column is collected by a ø200 mm, f = 800 mm Newtonian telescope (Sky-Watcher, Quattro-8CF). The collected light passes a longpass filter (Edmund Optics) with 50-mm diameter, optical density OD = 4, and cut-on wavelength 1550 nm. The illuminated air column is imaged sharply onto an InGaAs array detector (Xenics, Lynx-2048-250um-GigE) with 2048 pixels of 12.5 × 250 µm size. The linear array is tilted 45° off the optical axis. Data from the sensor are transferred to a PC via Ethernet, and the strobe synchronization signal is fed to two Digital to Analog Converters (DACs) on a single Data AcQuisition (DAQ) board (National Instruments, NI-6512). The DACs are configured from the PC via USB2, such that the strobe pulse updates the analog values from two table registers. The lengths of the registers correspond to the number of time slots in a measurement cycle. The sequence controlling the pump diode laser includes both on- and off-values. An on-value correspond to that the pump laser is on, whereas an off-value means that the laser is switched off. The laser-off setting is used for dark acquisitions, allowing subtraction of background light and pixel-specific dark noise. The sequence controlling the seed laser current also includes dark timeslots and a staircase (see sketch in Fig. 2) with current levels specifying the relative spectral band positions of the laser. The absolute band position is controlled by the seed laser temperature.

2.2 System operation

The data presented here were (Fig. 3) acquired with the system operating in hyperspectral mode, with 50 entries in the DAC registers and thus 50 time slots, 2 dark time slots and 48 spectral bands (Fig. 3(c)). The spectral bands were selected to cover a wavelength range of 1 nm, including three CO₂ absorption lines, namely the R14, R16 and R18 lines, of the (301←000) absorption band. The line array detector captured frames of 2048 pixels by 1000 exposures. The exposure time for the data presented here was 30 ms. With 50 time slots that implies 50 × 30 ms = 1500 ms acquisition time per cycle, and one frame takes 1000 exposures × 30 ms = 30 s. The sensor does an analog dark current subtraction prior to digitalization. The InGaAs detector was controlled from LabView (National Instruments) with real-time data visualization.

 

Fig. 3 Collected raw data represented as time-range maps on three different time scales. (a) Recorded data throughout the entire night, illustrating the long-term stability of the system. (b) Data revealing an exhaust plume emitted by a vehicle located on Tornavägen (see map in Fig. 1). (c) Data showing a full measurement cycle, initiated by two background measurements, λ_dark1 and λ_dark2.

Download Full Size | PDF

2.3 Range calibration

The range was calibrated by back-extrapolating from a hard target at a known far distance [41,46]. Further, this calibration relies on the system parameters: sensor tilt, pixel number and size, focal length of receiver and baseline separation of transmitter and receiver. From this a numerical vector of corresponding ranges from each pixel is obtained. As in other triangulation methods, range sampling follow a tangential function which can be approximated by a squared range [46]. For this reason the conventional r⁻² attenuation known from ToF lidars is absent and atmospheric echoes can be flat when extinction is negligible. In practice, range resolution is also limited by beam width [41]. When the beam expander size is chosen optimally the range resolution deteriorates linearly [40], which is close to the case in this study, where ranging accuracy is on the order of 5% of the range. The lidar near limit is 35 m and the range of complete overlap is 120 m, see [40]. The calibration point was located at a distance of 980 m. Based on tests, range estimations between the lidar system and the calibration point are accurately determined, whereas range estimation beyond the calibration point are subject to increasing uncertainty and deviation from the true value.

2.4 Spectral calibration

The background recordings, stored in the dark time-slots, were subtracted for each measurement cycle before averaging the signals in time. The slots, i.e. spectral bands, that experienced absorption by CO₂ were identified by evaluating the path-integrated absorption signals together with wavelength information from a wavemeter. The absorption lines of CO₂ at 6358.65 cm⁻¹ (R14), 6359.97 cm⁻¹ (R16) and 6361.25 cm⁻¹(R18) were identified, and from this information the wavelength scale was calibrated.

Range-resolved CO₂ number density, N_CO2, was obtained by using the DIAL equation:

Where Δσ is the difference in absorption cross section on and off the absorption line, i.e. Δσ = σ_on - σ_off, P(λ_off,r) and P(λ_on,r) are the backscattered signal off and on the absorption line at the distance r. The second factor in Eq. (1) is obtained by defining a range interval Δr = r_n2-r_n1, where n₁ and n₂ are pixel numbers, and performing a linear fit of ln(P(λ_off, Δ_r)/P(λ_on,Δ_r)). This procedure was repeated for each spectral band and the extracted absorption coefficients from the fits result in an absorption spectrum, as shown in Fig. 4(a). The peak values obtained from a Lorentzian lineshape fitting procedure together with tabulated absorption cross sections from HITRAN [48] (Table 1) were used to calculate a mean CO₂ number density from the absorption lines. The CO₂ concentration in ppm could then be extracted by relating the measured CO₂ number density to the total number density of air, determined from the ideal gas law with the temperature information from the weather station as input. The pressure used to calculate the number density in air was set to 101.325 kPa, i.e. 1 standard atm. The decrease in pressure with increasing altitude was neglected, since the present maximum increase in altitude, i.e. 150 m, merely decreases the pressure by less than 2% (assuming a constant temperature).

 

Fig. 4 (a) Measured spectrum due to absorption from CO₂ over a path length of 2000 m. (b) Each of the spectral bands corresponds to a backscattered signal.

Download Full Size | PDF

Table 1. Molecular data for the three CO₂ lines, taken from [48].

View Table

3. Results

The CO₂ concentration was evaluated over a distance of approximately 2000 m. Due to a non-linear mapping of camera pixel number to range, the nearest 400 m will cover about 1500 of the 2048 pixels on the camera. When evaluating the CO₂ for a certain range interval, the number of data points will therefore vary depending on the range of interest. Figure 5 shows evaluated CO₂ concentrations versus time for four different range intervals, all being ∼500 m long, namely 39 - 539, 540 - 1050, 1060 - 1570, and 1600 - 2120 m, with a time averaging of 1 hour. As can be seen in Fig. 5, the variation in CO₂ concentration is smallest in the nearest region, i.e. 39 - 539 m, where the concentration varies between 371 and 414 ppm. The variation then increases with increasing distance, thus being largest for the range interval 1600 - 2120 m, where the CO₂ concentration varied between 223 and 422 ppm. The highest evaluated CO₂ concentration was roughly equal for the different range intervals and was observed during the evening, i.e.17:00 −18:00. The largest differences in CO₂ concentrations were observed during the night, with a maximum concentration decrease of 150 ppm when going from near to far field. The CO₂ concentration is decreasing with time from 17:00 to 19:00 for all range intervals. The measurement was interrupted around 07:00 dawn when the sunlight background caused the detector to saturate.

 

Fig. 5 CO₂ concentrations evaluated and monitored from 17:00 to 08:00. Here CO₂ concentration is evaluated in range intervals of 500 m.

Download Full Size | PDF

Figure 6(a)-6(d) shows evaluated CO₂ concentrations versus range for different range bins Δr, namely Δr = 200 m (Fig. 6(a)), 400 m (Fig. 6(b)), 500 m (Fig. 6(c)), and 1000 m (Fig. 6(d)). The mean value and standard deviation are based on measurements performed during the night between 22:00 and 03:00. For Δr = 200 m, i.e. the result shown in panel a), the CO₂ concentration varies between a maximum of ∼390 ppm, at the closest distance, and a minimum of ∼350 ppm at a distance of ∼1200 m. As indicated by the error bars, the uncertainty in the measured concentration increases significantly for distances beyond the range calibration point at 1000 m. As can be seen in panels Fig. 6(b) –6(d), for analysis based on range bins of 400, 500, and 1000 m, the evaluated CO₂ concentration decreases monotonically with increasing distance, starting at ∼380 ppm at the closest distance, and ending at ∼280 ppm at the farthest distance. As expected the uncertainty in the measured concentration decreases with increasing range bin.

 

Fig. 6 The CO₂ concentration was evaluated and averaged over five hours between four range intervals of (a) 200 m, (b) 400 m, (c) 500 m and (d) 1000 m.

Download Full Size | PDF

As evident from Fig. 6, the evaluated CO₂ concentration is generally decreasing with increasing distance. Since the laser beam is transmitted at an inclination angle of 4.3°, which correspond to a maximum altitude change of 150 m, the observed increase in CO₂ concentration could possibly be due to a vertical CO₂ concentration profile. The values reported here are somewhat lower than most of the values reported in the literature, for example 405 ppm as reported in [2]. However, it should be emphasized that most concentration values reported in the literature, including the one stated in [2], are global yearly averaged CO₂ column concentrations, while our measured concentrations are neither vertical column nor averaged over all seasons. Concentration variations of the same order of magnitude as observed in our study has been observed by Cadiou et al. [38], who measured the CO₂ concentration with DIAL in the atmospheric boundary layer over the Paris metropolitan area. For our results, it should be noted that possible errors in the range calibration, i.e. the transformation of pixel number to range, introduce an error in the calculated derivative in the DIAL-equation (Eq. (1)), and thereby an error in form of a bias in the evaluated CO₂ concentration. In addition, since the function describing the relation between pixel number and range is non-linear, the systematic bias error will drift for ranges beyond the range reference point (980 m in this case), and, therefore, beyond this point the concentration error will increase non-linearly with increasing distance. Inaccuracies due to pixel indexations or calibration target inhomogeneity can also cause the estimated range to deviate in a tangential manner. This systematic bias is not included in the error bars in Fig. 6(a)-6(d).

4. Conclusions

Atmospheric CO₂ monitoring using CW-DIAL in Scheimpflug configuration has been successfully demonstrated for a test range of 2000 m. The setup is based on a fiber-amplified diode laser, generating wavelength-tunable radiation in the vicinity of 1.57 µm. The concentrations evaluated for the closest distances (within the nearest 400 m) are close to 400 ppm, which is in good agreement with the global average CO₂-concentration at the Earth’s surface [49], and thus clearly demonstrates the functionality of the Scheimpflug-lidar concept for CO₂ detection. Generally, the range-resolved CO₂ concentration profiles are decreasing with increasing distance. It is possible that this decrease, at least partly, is due to errors in the range calibration beyond the calibration point. In the future, such errors could possibly be corrected by making time-of-flight measurements in a fast acquisition mode, which the current driver electronics does not allow.

Since molecular concentration is evaluated from the range derivative of the on/off signal ratio, the signal-to-noise ratio of the recorded echo has a large impact on the evaluated concentration, and consequently limits the achievable range resolution for CO₂ measurements, despite the inherently high range resolution of the lidar system (compare the lidar signal from aerosols in Fig. 3(b) with the DIAL measurements of CO₂ molecules in Fig. 6). Higher signal-to-noise ratio and higher time resolution can be obtained by using a more powerful laser, which also would facilitate the real-time challenge of aligning the system. Increased background suppression and daytime monitoring is possible using a narrow bandpass instead of a long-pass filter [50–52]. The light cone from the receiving F/4 telescope allow narrow bandpass filtering at 1573 nm down to 7 nm FWHM. Such filter requires custom made coating which was outside the scope for this feasibility study.

The absorption bands at 1.57 µm are weak and therefore not suitable for studies of fine tempo-spatial details, such as exhaust plumes from vehicles and exhalation after metabolism. The sensitivity of the method could however be significantly improved by using the CO₂ absorption bands at 2.0 or 2.7 µm, which have up to a factor of 1000 times higher absorption line strengths than the absorption band used in the present study, i.e. at 1570 nm. Such a measurement system would enable detailed range resolved CO₂ profiling over e.g. cities or vegetation, but would require a Sterling-cooled shortwave HgCdTe focal plane array detector.