A lightweight balloon-borne laser diode sensor for the in-situ measurement of CO₂ at 2.68 micron in the upper troposphere and the lower stratosphere

Abstract

We report the development of a laser sonde operated under stratospheric balloons and devoted to the in-situ measurement of carbon dioxide in the upper troposphere and the lower stratosphere. In the 2.68 micron region, strong CO₂ transitions are suitable for the in-situ monitoring of carbon dioxide, which gives ∼10% absorption depth and, moreover, antimonide laser diodes are nowadays available that show relevant spectral properties for absorption spectroscopy. The light-weight sensor is based on 50-cm single path configuration and is operated open to the atmosphere. We provide details of the design of the instrument and data processing. The performance and the stability of the instrument were evaluated with the Allan variance technique. The spectrometer was test-flown in the Arctic stratosphere from Kiruna, Sweden and we report preliminary flight results.

1 Introduction

Carbon dioxide is a very long-lived natural tracer that increases with time in the troposphere and has no significant sources or sinks in the stratosphere. Hence, this molecule can be used to determine the “age of air”, i.e. the time needed to transport an air mass in a given location in the stratosphere from its entry point, essentially the tropical convective regions. The distribution of mean age with latitude and altitude from CO₂ observations can be used to investigate the general circulation in the stratosphere [1–5], despite the seasonal variation of this molecule in the troposphere [2]. A better understanding of the stratospheric circulation in the models is needed to understand the response of the ozone layer due to natural or anthropogenic perturbations [6–8]. To address this science issue, we have developed a light-weight laser sensor to provide in-situ measurement of CO₂ in the upper troposphere (UT) and the lower stratosphere (LS). Former in-situ observations, for example with balloon-borne cryo-samplers [9], reveals a decrease in the CO₂ between the UT and the LS, due to the time it takes for transport processes to mix tropospheric air into the stratosphere. Observations are still needed here, but the order of magnitude of the difference is ∼10 ppmv at mid-latitudes. The objective of the laser-sensors is to collect long-term measurements from various latitudes: the instrument is to be launched by non-specialists of the laser field under weather balloons from meteorological stations networks or to be operated as a piggy-back in larger science gondola. We target for CO₂ in the middle atmosphere, a ±2 ppmv precision error for a measurement time of 1 s. The precision error can then be further enhanced by co-adding successive data at the cost of a lower spatial resolution in the vertical concentration profile. A spatial resolution of a few tens of meters still matches our science requirements. The average speed of a balloon is a few meters per second. Apart from the study of the CO₂ in the UT-LS, such a laser sonde could be also useful for the validation of the planned satellite missions devoted to the study of the carbon-cycle.

Several developments have been made previously to monitor CO₂ with laser diodes, in the near infrared with telecommunication InGaAs laser diodes [10–12] as well as in the mid-infrared with lead salt laser diodes [13] or more recently with quantum cascade lasers [14]. For our part, we had explored the capability of antimonide laser diodes to monitor CO₂ at ground levels [15] and now we report in this paper the use of this cutting-edged laser technology to develop light-weight easy-to-use balloon-borne sondes to probe the middle atmosphere. The 2.68 micron spectral region is particularly interesting to monitor in-situ carbon dioxide in the UT-LS. It features the strong v1+v3 vibrational band. At ground level, water vapor is interfering and must be taken into account in the data processing [15]. Nonetheless, the H₂O concentration decreases dramatically with altitudes and above roughly 3 km, CO₂ transitions can be found that are free of overlapping by water vapor lines. In the upper troposphere and in the stratosphere that is strongly dehydrated, the contribution of H₂O to the overall molecular absorption is negligible.

Figure 1 displays the selected spectral region with the contribution of both CO₂ and H₂O at 5 km and 15 km. For the balloon sensor, we have selected the R(18), (10⁰1)_I ←–(00⁰0) transition at 3728.4 cm⁻¹ with a line strength of ∼6×10⁻²⁰ cm⁻¹ molecule⁻¹ cm². Table 1 gives a few numbers to illustrate the variation of the absorption depth with altitude. At 2.68 micron, the absorption depth is roughly 10% up to 25 km over an absorption path length of only 50 cm.

Fig. 1

Simulation of the atmospheric spectrum in the selected spectral region at 2.68 micron at 5 km (a) and 15 km (b). The selected CO₂ rotation-vibration transition is quoted in (a). It is free of overlapping by H₂O over ∼3 km

Table 1 Comparison between predicted absorption depth for the SDLA spectrometer [21] (L=56 m) and the PicoSDLA-CO₂ spectrometer (L=50 cm) in the troposphere to lower stratosphere for mid-latitudes

Table 1 also features the absorption in the standard 1.60 micron spectral range reachable with telecommunication-type InGaAs laser diodes; for instance with our former TDL balloon-borne spectrometer called SDLA [10], we had selected the R(16), (30⁰1)_III←–(000) transition at 6240.1 cm⁻¹ featuring a line strength of ∼1.7×10⁻²³ cm⁻¹ molecule⁻¹ cm². With a 50 m absorption path length provided by a multipass cell, the absorption depth was less than 0.5% at 1.60 micron, to compare to 10% over 0.5 m at 2.68 micron. At 2.68 micron, single-path lightweight spectrometers can be designed by removing the need for a complex optical multipass cell to expand the absorption path.

Moreover, the featured strong absorption permits to get a high signal-to-noise ratio in the spectra with a simple direct detection scheme [16]. A new-generation of distributed-feedback antimonide (GaInAsSb) laser diodes emitting at room-temperature near 2.7 μm is now commercially available to monitor CO₂ in this spectral region that can be combined to efficient InAs detectors. The spectral emission properties of these lasers are well suited to absorption spectroscopy: it features good monochromaticity, absence of mode-hops, wavelength tunability over a few reciprocal centimeters, a laser line width of a few MHz, as well as an optical output power of a few milliwatts. For two years, we have been working on the development of the balloon-borne laser sensor, called PicoSDLA-CO₂, by using this laser technology and with the help of the French space agency and of the CNRS.

In the next chapter, we will describe the laser sonde. Then we will discuss the data processing and the testing of the sensor in the laboratory. Finally, we report the preliminary results from a balloon flight in the Arctic stratosphere in spring 2011.

2 Instrument design and laser spectroscopy

The design of the instrument is very simple: the laser diode and the InAs detector are maintained at a 50 cm-distance using fiber carbon tubes; the laser is propagated in the open atmosphere where the laser flux is partially absorbed by the ambient carbon dioxide molecules; the absorption spectrum is then recorded and stored onboard. Figure 2 is a picture of the PicoSDLA-CO₂ laser integrated in a balloon flight chain. The laser device at 2.68 micron was purchased from Nanoplus Inc. and is mounted in a TO-8 package with integrated Peltier thermo-element. To avoid laser spectral drifts due to the large variation in ambient atmospheric temperature (down to −70^∘C for the reported flight), we have designed a small module (∼5 cm-diameter) combining aluminum and teflon and equipped with thermal heaters. The temperature in the module is maintained within ±10^∘C of the laser operational temperature so that the Peltier thermo-element can properly stabilize the temperature of the laser component. The detector is a Judson InAs photodiode with a diameter of 1 mm. The focusing and the collection of the laser light are ensured with two sapphire lenses featuring a 25 mm focal length. In addition to an anti-reflection coating, the lenses are slightly tilted to avoid fringing due to reflected light. The lenses are heated to avoid the formation of ice on the optical surfaces. The atmospheric pressure and temperature are monitored respectively with a Honeywell pressure gauge (0.01 hPa precision error) and three meteorological VIZ thermistances (0.1^∘C precision error). The location of the gondola in flight is obtained from an onboard GPS. For the first flights of this prototype, we have also at disposal an additional high-precision pressure gauge from Paroscientific Inc. The onboard electronics is based on a Pentium architecture; it controls the lasers and takes the data using 16-digits acquisition cards. The data are stored onboard on flash memories and also transmitted to the ground stations through a custom-made UHF telemetry-telecommand subsystem (the antenna can be seen in Fig. 2). To locate the sensor at landing, a cell phone is integrated that sends the GPS information once the sensor is at ground level. The electronics as well as the laser and detector modules was designed to be operated unpressurized despite the low pressure at high altitude (down to 30 hPa for the reported flight). The weight of the laser sonde (laser, detector, antennas, gondola structure in carbon fibers, pressure sensor, thermal protection and heaters…) is roughly 800 g. The lithium batteries (1 kg), the electronics module, the telemetry-telecommand subsystem and the GPS are installed in a box located above the sensor as can be seen in the picture; this box has a weight of nearly 5 kg (including the batteries), which is mostly due to additional hardwares needed for the test-flight. It will be downsized once the prototype is fully tested. The sensor has not constrains to launching. It can be installed in a flight chain (as in the picture) or as a piggy-back in a larger gondola. The optical alignment does not require particular care; the optics is adjusted to maximize the signal. The instrument can be easily operated by non-specialist of the laser field.

Fig. 2

A picture of the PicoSDLA-CO₂ spectrometer. The laser flux emitted by an antimonide laser diode at 2.68 micron is propagated over 50 cm in the open atmosphere and recorded with an InAs detector. See text for more details

The diode laser is scanned over the molecular line shape by ramping of its driving current within 10 ms at constant laser temperature and 256 sample points are taken in the absorption spectrum using a 16-bit digitizer. For the flight configuration reported in this paper, twenty elementary spectra are co-added to give one CO₂ spectrum. Hence the overall measurement is 200 ms. The laser diode is regularly switched off during the flight to give the electrical zero-level. Given the strong absorption depth (10%), the detection scheme is a simple direct detection combination whose details and performances may be found in reference [16]: after propagation over 50 cm, the laser flux is simply converted into an electrical signal with one single InAs photodiode and the signal is digitized with a 16 bits analogic-to-digital converter that provide a sufficient dynamical range of the measurement for the atmospheric molecule under study. Figure 3 is an example of stratospheric CO₂ spectrum achieved during the balloon flight-test in Kiruna (67^∘N) on the 12th of March 2011 in the Arctic stratosphere at 22 km. As can be seen in the figure, the signal-to-noise ratio is high, of the order of 200. To extract the molecular absorption (Fig. 3(b)) from the direct absorption spectrum in Fig. 3(a), a polynomial interpolation (3rd order) is applied on zero-absorption regions on both sides of the direct spectrum to get the baseline (quoted F ₀ in Fig. 3).

Fig. 3

Example of in situ CO₂ spectrum taken in the Arctic stratosphere with the PiocSDLA-CO₂. The direct atmospheric spectrum (a) is recorded within 200 ms. The molecular absorption (b) is retrieved from the direct spectrum in (a). The absorption depth is 10% at 22 km. A Fabry–Pérot signal (c) is used to scale the wavenumber axis in the spectrum. See text for more details

The baseline is what would be the laser signal in the absence of molecular absorbent. The achieved noise is ∼5×10⁻⁴ expressed in absorption units for a 200 ms measurement time. In Fig. 3(c), a Fabry–Pérot (F–P) signal is shown that is used for the scaling of the wavelength frequency and to take into account non-linearities in the laser spectral emission. For the prototype, we do not have the F–P onboard. We record before the launching the F–P signals over the spectral ranges selected for the flight by placing a germanium 5-inch (12.7 cm) etalon in the optical path.

3 Data processing and laboratory test

The carbon dioxide concentration is retrieved from the absorption spectra by applying a non-linear least-squares fit to the full molecular line-shape (Fig. 3(b)) in conjunction with the in-situ atmospheric P and T measurements. The details of the inversion procedure and the theoretical formula for the laser spectroscopy technique used in this work may be found in [17]. The model is based on the Beer–Lambert law, which relates the absorbed laser energy to the molecular densities, and uses the Hitran 2008 molecular data [18]. The revisited molecular parameters, particularly the temperature dependence of the pressure-broadening parameters will be the subject of a forthcoming spectroscopic paper. This temperature dependence is to be revisited using the available cryogenically-cooled optical cells available at the GSMA laboratory [19] as the sensor explore large range of atmospheric temperature during a stratospheric flight, typically from +20^∘C to −80^∘C. To report these preliminary data, we have made use of the Hitran data and the stored spectra will be reprocessed with the revisited molecular parameters in a next step.

To test and calibrate the sensor, it was placed inside a vacuum cell as shown in Fig. 4(a). The cell was filled with a calibrated mixing of carbon dioxide and dry air. The mixing was measured at 503.9 ppmv ±1 ppmv of carbon dioxide by use of a LI-COR 7000 reference spectrometer that was itself calibrated with three primary CO₂ etalons with an inaccuracy of ±0.1 ppmv. This mixing ratio of 500 ppmv is slightly higher than the standard atmospheric CO₂ concentration (390 ppmv), but this does not alter the discussion on the detection limit. Pressure and temperature were monitored simultaneously to the absorption spectra using the onboard pressure gauge (Honeywell, 0.01 hPa precision error) and a MKS Baratron gauge with 100 torr full scale (accuracy: 0.25%) and the three onboard meteorological thermistors (VIZ, 0.1^∘C precision). In order to scale the frequency axis of the spectra, we have recorded the Fabry–Pérot signals from the 5-inch germanium etalon for each of the current ramping used for this laboratory test. The free spectral range of the Ge etalon is 9.422×10⁻³ cm⁻¹ (accuracy: 2.355×10⁻³ cm⁻¹). Figure 4(b) is an example of CO₂ spectrum achieved with the laboratory set-up at a total pressure of 190 mbar (approximately 11 km in the atmosphere). For the spectrum in Fig. 4(b), the measurement time is 400 ms corresponding to the co-addition of forty elementary 10 ms-spectra. The precision error in the retrieved concentration was obtained from processing a large set of spectra at the same P and T conditions. With the used processing technique, several standard sources of error contribute to the overall precision error [17] in the molecular concentration retrieval process: the error in the baseline determination (<0.2% considering the achieved high signal-to-noise ratios and considering the pressure range of the UT-LS), the precision error in the pressure and temperature in situ monitoring (<0.1%) and the achieved high signal-to-noise ratio in the spectra (∼0.1%). Regarding the absolute accuracy, it is mostly driven by the accurate knowing of the absorption path length (0.2%), of the line strength (the reported HITRAN inaccuracy is ∼2% for the selected transition). as well as by the use of the Voigt line shape model whose systematic bias was estimated to be 0.7% by calculating the difference between the maximum and minimum of the main structure of the residual and by comparing this value to the maximum of the considered line. Regarding the spectroscopy, we are currently revisiting the line strengths in this spectral region (targeted inaccuracy <1%) and the temperature dependence of the pressure broadening parameters. To process these preliminary data reported in this paper we have used the HITRAN 2008 data. Nonetheless, for the science objective that consists of estimating a difference between a mean content of the UT and a mean content of the LS, the absolute accuracy, assuming it causes a constant bias in these region of the atmosphere, is of less impact. Series of measurements at various pressures were done using this laboratory cell to investigate the achieved overall precision error. For instance, Fig. 5(a) is an example of a 5-hour measurement set done at a pressure of 112 mbar (corresponding to an altitude of ∼14 km) and at room temperature. The achieved one-σ standard deviation for a measurement time of 400 ms is nearly ±2.2 ppmv (i.e. ±0.45% of 503.9 ppmv). It is consistent with our objective of a ±2 ppmv precision error for a ∼1 s measurement time as mentioned in the introduction. For measurement times longer than one second, in order to enhance the precision error, we prefer to co-add concentration data rather than to co-add successive spectra to avoid artifacts due to spectral drifts in the laser emission wavelength. Hence, a moving average was applied over the concentration set by co-adding 25 successive concentration data, corresponding to a 10 s integration time (in red on Fig. 5(b)). For an integration time of 10 s, the standard deviation is then roughly ±0.7 ppmv. To further assess the performances of the instrument, series of Allan variances were calculated [20, 21] to discriminate the different sources of noises and to estimate the temporal stability of the instrument. The Allan variance ($\sigma_{\mathrm{Allan}}^{2}$) permits one to specify noise in a time series, following the relation

$$ \log\sigma_{\mathrm{Allan}}^{2}(\tau) = \mu\log\tau\quad \mbox{for}\ \tau = \tau_{0},2\tau_{0}, 4\tau_{0}, \ldots $$

(1)

τ ₀ is the sampling period in second. The slope of this log-to-log relation (μ) relates to the nature of the predominant noise:

• μ=−2, super flicker noise

• μ=−1 white noise

• μ=0 flicker noise

• μ=1 random walk

Fig. 4

(a) A picture of the cell used in the laboratory to test the stability and the performance of the laser sensor. The PicoSDLA sensor is located inside the cell that is filled with a known mixture of CO₂ and dry air at various pressures. (b) An example of laboratory spectrum

Fig. 5

(a) A long-term measurement of CO₂ over several hours. The data were smoothed using a moving average over 25 successive elementary measurements (in red). (b) A plot of the Allan standard deviation. See text for more details

Figure 5(b) shows a plot of standard deviation σ _Allan, according to the integration time. For our science objective, an integration time up to 10 s is still acceptable as it gives a spatial resolution of a few tens of meters in the vertical concentration profile as mentioned previously. Up to 10 s, the Allan variance ($\sigma^{2}_{\mathrm{Allan}}$) decreases with a slope close to −1 (∼−0.86), indicating that a white noise is predominant. For a one second measurement time, the Allan standard deviation is 1.6 ppmv and it decreases down to ∼0.6 ppmv for a 10 s integration time. At longer time (in our case about above 900 seconds), the variance is rising again because of undesirable drift noises. The minimum of the variance correspond to σ _Allan=280 ppbv at τ _Allan=900 s, which corresponds to the optimum time of integration. The expected performance of the prototype with a 1 s measurement time from this laboratory characterization does match the requirements in terms of a precision error in the range ±2 ppmv. The precision error can then be enhanced to ∼±0.5 ppmv by co-adding successive concentration data.

4 Preliminary flight results and conclusion

The PicoSDLA was launched on 12 March from Kiruna (67^∘N), Sweden. The laser sensor was flown as a piggy-back of the ELHYSA frost-point hygrometer (PI: Dr. G. Berthet, LPC2E, CNRS-Orléans). It was installed in the flight chain as shown in Fig. 2. The scientific gondola was flown by a 100000 m³ open stratospheric balloon operated by the CNES and reached a 23-km float altitude before starting a slow-descent in the stratosphere with a separation at 15 km followed by a descent under parachutes. The flight lasted four hours. The flight occurred on the edges of the Arctic vortex. The reported CO₂ spectra were recorded only during the slow descent in the stratosphere and under parachutes in the troposphere. During the ascent we have experienced unpredicted contamination of the laser by our own UHF telemetry; the 400 MHz telemetry contaminated the laser driving current causing the apparition of additional side-lobs in the laser spectrum (as in high-frequency wavelength modulation technique). At float, the telemetry was turned-off and we switched to the CNES telemetry subsystem as a back-up. This problem will be solved for the next flight of the instrument by de-locating the telemetry module from the laser electronics module. The temporal resolution was fixed at 200 ms: the co-addition of 20 10 ms-spectra gives one average spectrum as shown in Fig. 3. The overall set of thousands of spectra was processed after the flight. The Paroscientific Inc. pressure gauge was used for the retrieval in conjunction with the HITRAN-2008 database. A non-linear least square fits was applied on the full line shape using a Voigt model and the F–P signals to scale the wavenumber axis. Figure 6 reports CO₂ spectra at different altitudes in the stratosphere and in the troposphere with the simulated spectrum yielded by the fit procedure. The amplitude of the driving current is changed with altitude to match the broadening of the molecular line shape with increasing pressure (the line width decreases by a factor ∼6 from ground to 20 km).

Fig. 6

In-situ CO₂ spectra yielded by the PicoSDLA-CO₂ sensor in Kiruna (67^∘N) at various altitudes in the Arctic troposphere and stratosphere. The spectra are recorded in 200 ms. The simulated spectra achieved by fitting the full molecular line shapes using the HITRAN 2008 spectroscopic data in conjunction with the in situ P and T measurements are superimposed to the experimental spectra. The absorption depth keeps nearly constant at 10% on average up to 20 km. The line width is strongly broadened by pressure-effect as the altitude decreases and the tuning of the laser must be adapted accordingly. An offset was added to the spectra at 16 and 20 km for the sake of clarity

As can be seen in Fig. 6, the absorption depth is nearly constant in the middle atmosphere despite the decrease with increasing altitude in the number of CO₂ molecules. This is due to the fact that the decrease in the number of CO₂ molecules with increasing altitude is compensated by the narrowing of the line due to a diminishing pressure broadening effect. The achieved signal-to-noise ratio is high in the full altitude range, of the order of 200. The noise is a few 10⁻⁴ expressed in absorption units for a 200 millisecond measurement time. The performance in-flight does match what was observed during the laboratory calibration with a detection limit in the 10⁻⁴ range (in absorption units) for this prototype, which is sufficient for our science application under balloon platforms. With the various sources of errors taken into account, which were discussed in detail in the previous paragraph, the precision error for the 200 ms measurement time is ∼±2 ppmv. With the achieved in-flight performance as well as the simulation in Table 1, the maximal altitude reachable with maintaining a satisfying performance is 30 km.

These preliminary flight results are very satisfactory as the selected spectral range near 2.68 micron offers strong absorptions by carbon dioxide which permits to develop compact and simple sensors based on single path over a few of tens centimeters and to get high signal-to-noise ratio in the in situ spectra. The next step of the project consists of doing an in-flight comparison with another instrument, for instance with a balloon-borne cryo-sampler. It would be more convenient to test-fly the sensor at mid-latitudes rather than in a complex meteorological situation like the Arctic vortex. We are going further in the downsizing and in the reduction of the overall weight to reach a weight of less than 3 kg for the complete instrument including the batteries. The spectroscopy and particularly the temperature dependence of the pressure-broadening coefficients are also to be revisited in our laboratory.