1. Introduction

Active volcanoes emit large amounts of gases into the atmosphere. In recent years, there has been growing attention towards volcanic gas monitoring in order to gain a better understanding of the precursor signals of an eruptive activity. Indeed, changes in the rate and the composition of volcanic gaseous emissions can be associated to magma movements underneath the crater [1]. The release of volcanic gases occurs through hot fumaroles which are the visible aspect of what can be considered an active or a quiescent volcano. Typically, the most abundant molecular species in volcanic gas are water vapour and carbon dioxide. Trace constituents include sulphur dioxide, hydrogen sulphide, methane, carbon monoxide, and in minor amounts also helium, hydrogen, nitrogen, ammonia, mercury vapour, etc. Useful geochemical parameters are fluxes, concentration ratios and isotope abundance ratios for the molecular constituents of volcanic gas. For instance, the temporal evolution of H₂O, H₂S and HCl, normalised to CO₂, were employed to study the 1982–1984 seismo–volcanic crisis of Solfatara volcano, near Naples, Italy [2]. As another example, aquifer temperatures can be inferred from compositional data using the H₂O–H₂–CO₂–CO–CH₄ geothermometer [3]: CO/CO₂, H₂/CH₄, and CO/CH₄ ratios may be indicative of reactions such as the catalytic hydrogenation of CO₂ and of CO (Fischer–Tropsch process) within hydrothermal systems.

Presently, volcanic gas monitoring is carried out, occasionally or periodically, through field sampling followed by laboratory analysis by means of mass spectrometry and gas chromatography. Instead, geochemical surveillance of volcanoes should be continuous, which demands using gas sensors placed in-situ nearby the fumaroles [4]. In the last decade, optical methods have shown a clear potential to satisfy these needs [5]. Among the different methods based on incoherent sources, Fourier Transform Infrared Spectroscopy (FTIR) has experienced a tremendous development. Indeed, open-path FTIR systems have been successfully employed to study gas emissions from different volcanoes, using natural or artificial infrared sources [6]. Remote measurements of volcanic gas compositions have been demonstrated using solar occultation FTIR spectroscopy, in which observations are possible wherever the path between the Sun and the instrument intersects the volcanic plume [7]. However, these methods have potential problems with selectivity, because of the limited spectral resolution, and evidently with the measurement of species present in the atmosphere, such as carbon monoxide and methane.

In order to obtain molecular densities in the gas phase, a technique of choice is laser absorption spectroscopy, which allows direct non-destructive monitoring of the intensity of absorption lines belonging to one or more molecules present in a gas parcel, with high precision and accuracy. Using a narrowband tunable laser, spectral resolution affords high selectivity when absorption lines are sufficiently narrow or well isolated. For the measurement of small concentrations the preferred strategy is to access the strongest absorption lines of a given molecule. For most molecules an obvious choice is the fundamental vibrational transitions found in the mid-infrared, however weaker overtone transitions at wavelengths extending to the near infrared are exploitable using high sensitivity spectroscopic techniques.

The advantage of shorter wavelengths is that it is possible to use Distributed Feed-Back (DFB) diode lasers which are compact and inexpensive, are power–efficient, and operate at room temperature. These are today readily available at any wavelength in the range 1.27 to 1.7µm from telecommunication companies, and at wavelengths in the mid–IR domain up to 2.7µm from specialized companies. Recently, DFB diode lasers were successfully applied to the continuous analysis of volcanic emissions. Direct detection of absorption in a multiple reflection cell allowed monitoring CO₂ and H₂O in volcanic fumaroles, in spite of the hostile environment (large temperature fluctuations, high humidity and acidic fumes) where the instrumentation had to operate [8]. Furthermore, the first field determination of the ¹³CO₂/¹²CO₂ isotope ratio in volcanic emissions was achieved using a diode–laser spectrometer working around 2µm [9].

In order to measure minor constituents, highly sensitive detection methods should be implemented, with the additional requirement of a compact and robust design capable of operating in harsh environmental conditions. With this objective we recently developed a Cavity Enhanced Absorption Spectroscopy technique which exploits Optical Feedback (hence OF–CEAS) occurring only at cavity resonances towards a diode–laser source. This is possible thanks to the V–geometry of the high–finesse optical cavity. The effects of this feedback are laser linewidth narrowing and temporary frequency locking to successive cavity resonances during a laser scan [10, 11], providing strong and reproducible cavity transmission signals at a fast tuning rate (~0.2GHz/ms). This technique, recently described in detail [12, 13], has important advantages with respect to other CEAS implementations not relying on optical feedback. Indeed, OF–CEAS detection limits are below 10^-9/cm at more than 1Hz scanning rate, together with a 4-decades dynamic range over an absolute and linear absorbance scale (eliminating the need for a calibration sample), a high precision linear frequency scale, and a small sample volume allowing for fast gas exchange even using small flow rates.

We report here probably the first application of an extended–wavelength (2.33µm) DFB diode laser [14, 15] to CEAS and the first demonstration of a compact CEAS spectrometer to in–situ trace analysis of volcanic gases. The chosen wavelength range is an atmospheric transmission window where absorption by the omnipresent water molecule is weak. Likewise, the other main volcanic gas, CO₂, does not absorb there, while strong absorption lines of several interesting species are accessible even inside the small tuning range of a single DFB diode laser.

2. Experimental

The DFB diode laser used for this work is based on a Ga–Sb multiple quantum well structure designed and grown by molecular beam epitaxy at the CEM2 laboratory of the University of Montpellier [14, 15]. The DFB metal grating was processed at Nanoplus GMBH (www.nanoplus.de), where DFB lasers emitting at up to 2.7 µm are now commercially available.

Our laser sample, one of the first available at this wavelength, presents three regions of mode–hop–free tuning separated by regions of multimode emission when the laser temperature is varied from 0 to 45°C. The first region, from 0 to 11°C, covers the spectral range 4295.5–4302.8 cm^-1 (~2323.4–2327.4 nm). After fixing the temperature, injection current ramps in the range 60–80mA yield mode–hop–free scans over about 1 cm^-1. Simulations based on the HITRAN database (http://cfa-www.harvard.edu/HITRAN) reveal in this range relatively strong absorption lines of CH₄, CO, and NH₃ all with max absorption strength close to 2×10^-21 cm/mol.

The OF–CEAS system is basically the same as described previously [13], except for the thicker and stiffer mechanical layout and the particular operating wavelength. This last required an appropriate set of high reflectivity mirrors for the cavity (by Layertec, www.layertec.de) and extended InGaAs PIN photodiodes (300µm diameter) readily available, e.g. from Hamamatsu. These photodiodes are coupled to 500 kΩ transimpedance amplifier circuits with ~500 kHz bandpass and ~1mV r.m.s. noise (RedWave Labs Ltd, UK).

We give just here just an outline of the working principle of OF–CEAS. A V–shaped high finesse resonator cavity composed of 3 high reflectivity dielectric mirrors works as a long effective–path (~10 km) absorption cell. Light from the diode laser is injected into the cavity through the folding mirror and intracavity optical field build–up occurs when the laser frequency matches one of the cavity resonances (modes). Due to the cavity geometry, some fraction of this buildup field feeds–back to the laser, which makes the laser emission to become spectrally narrower than the resonances (which are a few kHz wide). This optical feed–back also acts as a fast frequency–locking mechanism which works as the laser frequency is tuned across several modes. At each mode the cavity is thus filled up and a CEAS spectrum is obtained by detecting the max transmitted light for each mode. A fraction of the laser power is sampled before the cavity by a beamsplitter and measured to allow normalizing the cavity transmission. OF–CEAS spectra are thus obtained with data points precisely corresponding to the equally spaced cavity resonances. The time the laser spends locked to each resonance depends on the feedback level, which is set by an attenuator. Also, for obtaining a feedback effect as described, the optical field coming back from the cavity to the laser must have an appropriate phase, which is controlled actively by a piezo mounted mirror.

Optical components in this setup are installed inside a massive aluminium frame as shown in Fig. 1 and described in the corresponding caption. This setup fits inside a 19” chassis and weights around 16 kg. The new V–shaped cavity design was developed for airborne applications [16]. It is obtained by joining (by epoxy glue and screws) 2 stainless steel bars carrying grooves to form half of the V channel each. At the ends of the resulting assembly we install 2 mirrors on one side and 1 on the other, using small metal holders which are screwed in place. Each mirror is prealigned and glued inside its holder, which is vacuum tight to the cell thanks to o–ring seals. This gives a very compact and stiff cavity, and allows easy mirror removal for cleaning without loosing optical alignment. This cavity is stable after months of operation and repeated mirror cleaning, which is actually needed only after procedure errors when modifying the gas flow and pressure in the cell, especially evacuating or filling the cavity too fast or using a non–filtered gas flow for too long. Mirror cleaning is not normally needed even after several weeks of continuous operation with a flow of filtered gas. In particular, during our measurement campaign which extended over 2 days we did not notice any degradation of the signal. The flow of rather aggressive dried volcanic gas was simply passed through a metal–sponge filter with pores of a few micrometers.

Not shown in Fig. 1 are the pressure controller (EL–PRESS P602C by Bronkhorst) placed at the cell gas inlet, and the small membrane pump connected at cell outlet. Also, heating ribbons with a controller (by Minco) are installed on the bottom of the aluminium frame to allow stabilizing the device temperature.

 

Fig. 1. OFCEAS setup inside its 19” rack chassis. Gas inlets and outlets to the V–shaped high–finesse cavity are next to the high reflectivity mirrors (M1,M2,M3). Photodiodes PD1 and PD2 collect signals for cavity input (given by beamsplitter BS) and output, respectively. A polarizer (P) is used to attenuate the laser light to set an appropriate optical feedback level. One of the steering mirrors is mounted on a PZT disk to allow fine control of the feedback phase. The 2.33µm DFB laser is installed on a dove–tail translation stage.

Download Full Size | PDF

The spectrometer was installed in a mini–van vehicle, parked inside the Solfatara crater about 20 meters away from the strongest fumarole. Here, the volcanic gas is mostly composed of H₂O (~80%) and CO₂ (~20%), with other gases in trace amounts [2]. The gas temperature can be as high as 160°C. A compact gasoline–engine generator and a voltage stabilization unit provided the necessary power supply. Volcanic gas was sampled through a home–made gas dryer and cooler, where most water vapour condensed and was collected in a reservoir placed at the bottom of the cold trap. In this setup, the hot volcanic gas blowing from the fumarole vent was collected by a glass funnel and delivered to the trap through a 1m long rubber tube. The trap consisted of a glass coil immersed in a styrofoam tank filled with ice and water. The coil was designed to function with the strongest gas fluxes that are expected at Solfatara volcano. The measured efficiency of this setup is close to the thermodynamic limit, with CO₂ gas accounting for about 98% of the dry gas emerging from the trap [9]. A small peristaltic pump was used to maintain a constant flow of a few litres per minute. During several hours of continuous sampling, no water condensation was observed in the 20m Teflon tube (4mm internal diameter) carrying the dry gas to the spectrometer.

3. Results

OF-CEAS absorption spectra are obtained from the peak transmission values of about 200 TEM₀₀ cavity modes scanned during 100ms laser current ramps. A ringdown measurement is also performed every few seconds on a single selected cavity mode to obtain information on the cavity finesse and determine an absolute and stable absorption scale [13, 16, 17]. The spacing of data points in a spectrum is then constant and the resulting frequency scale is perfectly linear. This spacing is given by the cavity free spectral range (about 150MHz here) and is sufficient to describe the profile of absorption lines at low pressure [12]. In order to better resolve the congested methane spectrum we then stabilized the pressure inside the cell to about 80mbar, which gives Doppler limited lineshapes.

According to what can be found in the literature, the volcanic dry gas at Solfatara volcano is expected to contain H₂S (~1%), N₂ (~0.3%), H₂ (~0.2%), and trace amounts of CH₄ (~250 ppm) and CO (~2 ppm) [4]. As we first explored the Solfatara dried gas absorption spectrum using the whole available temperature-tuned spectral range of our laser, we could easily identify absorption lines of CH₄, CO, and residual H₂O. This was possible by comparing to HITRAN simulated spectra, where a good agreement in positions and relative intensities was found for all major lines. NH₃ lines seemed also to be present, but always blended with some methane feature. We already noted that no CO₂ features are present in this spectral range according to HITRAN. We verified the absence of even very weak CO₂ transitions in the laboratory by injecting pure CO₂ through our instrument. On the other hand, we did observe several unidentified weak features in the Solfatara gas absorption spectrum, but these are likely just weaker lines from water, methane, and their isotopes.

Given the sparse CO spectrum, we had to choose between only two major lines available in the laser tuning range. One of these was accessible for a laser temperature closer to ambient, which fixed our choice of a reduced spectral region to run fast OF-CEAS scans for real time acquisition (shown in Fig. 2). In this region, methane lines are much weaker than those available at other wavelengths accessible by our laser, but this was compensated by the large trace level of this molecule. On the other hand, this region includes one of the strongest ammonia features available, even if partially hidden behind two methane lines. We will see that the detection of this molecule poses other more serious problems.

During the Solfatara measurements, spectra taken in this selected region were streamed to a file for later processing by a Voigt lineshape fitting routine. Fig. 2 displays a typical recorded spectrum, compared to a HITRAN simulation using the concentrations determined from line fitting. The agreement is excellent, as is seen in the baseline profile obtained after direct subtraction of the simulation from the spectrum. For processing the recorded spectra, we chose a reduced set of lines including and around the CO line, then fitted these over a small set of spectra to obtain averaged values of the collisional linewidths and the relative line positions. These were then used as fixed parameters to perform faster fits of all recorded spectra. With respect to the sinusoidal baseline oscillation, it was included in the fit together with a baseline offset parameter. This is a well-known effect often observed in CRDS or CEAS, due to optical interference with parasitic reflections from the back surface of the cavity mirrors. We could recently mostly eliminate this artefact by using mirrors with antireflection coating on the back.

In order to obtain concentrations as a function of time, we used the intensity of the CO line and that of the strongest CH₄ line in the fit together with the corresponding line strength factors from the HITRAN database. We repeated the same type of analysis for the ammonia line at 4297.0 cm^-1 and 4 methane lines partially blended to it. This also provided a new CH₄ concentration trace, which is found to be in good agreement with the previous one, as shown in Fig. 3 where fumarole gas concentration traces for a 1 hour period are plotted.

The consistency of methane levels obtained from independent spectral regions indicates that HITRAN relative line strengths may be very precise. However, the absolute accuracy of HITRAN line strengths is often not better than 10%, thus a more accurate determination of concentrations would demand re-measuring a few line intensities under controlled conditions of pressure, temperature, and concentration. While we did not consider it necessary at this stage, this could be done by injecting a sample of known concentration in our instrument. For the rest, no periodic calibrations are needed with OF-CEAS to guarantee reproducible measurements, as the excellent reproducibility over long time spans has already been assessed [13].

 

Fig. 2. Typical single-scan OF-CEAS spectrum (thick black line) of fumarole dried gas, taken for a sample pressure of 80mbar. In order to extract concentrations, a section of the spectrum around the CO line was fit using 10 Voigt profiles (more details in the text). In colour are HITRAN simulations for the 3 contributing species for the same pressure and concentrations values as obtained from the fit. Their sum is plot as a dotted black line and is subtracted from the experimental data to give a residual sinusoidal baseline (short-dash black line).

Download Full Size | PDF

The noise-equivalent absorption on the recorded spectra, without averaging, is about 10^-8/cm. This is about 10 times worse compared to previous OF-CEAS results in the telecom range, which is partly due to the relatively modest cavity ringdown time (~6.5µs), partly to the presence of small fringes on the spectra, again due to stray reflections, which we did not succeed to reduce at the time of this campaign. Spectral noise mostly accounts for noise on the concentrations of CO, CH₄, and NH₃, visible in Fig. 3, with standard deviations of 16 ppb, 0.4 ppm, and 14 ppb respectively.

No variations of the concentrations of CO and methane are observed in the fumarole emission over time scales of 1 hour. This allows excluding the alteration of molecular densities of these species due to the long tube delivering the sample from the fumarole to the device, since there would have been time for equilibration with respect to wall adsorption processes. The situation is not so simple for molecules such as ammonia, with a good solubility in water and high sticking propensity to surfaces. It is expected that a sizable fraction of ammonia remained in the water condensing in the cold trap and also was adsorbed on the walls of the long tubing. In fact, the apparent variation in its concentration (large compared to those of CO and CH₄) is likely an artefact associated with manual purging of the cold trap reservoir, occasionally needed. The residual ammonia detected should thus largely underestimate the level present in the fumarole emission. A calibration of the ammonia throughput for the cold trap in controlled conditions (including an auto-purging system…), and placing the instrument much closer to the fumarole, are solutions which should allow a correct ammonia determination by direct observation of its concentration in the dried fumarole gas.

 

Fig. 3. Concentrations of CO, CH₄ and NH₃ during a 1 hour period of continuous monitoring of the fumarole gas. These are obtained from multiline fits of spectra as the one shown in the previous figure. The flow conditions were slightly perturbed by the need to clean up excess water and ice clogs accumulating inside the cold trap, which might account for the fluctuating NH₃ value. Indeed, for this very polar species we cannot be conclusive about its concentration due to the sampling scheme. 2 superposed traces are given for CH₄ which are obtained by fitting different sets of lines in the spectrum.

Download Full Size | PDF

An interesting observation we would like to report was obtained while testing the device with a flow of ambient air from inside the Solfatara. Bursts of CO and methane were observed, as seen in Fig. 4. These bursts are at times strongly correlated, while sometimes only CO is observed. On one hand, this indicates that the spectral fitting procedure works well: Changes in the concentration of a species do not affect the determination of other species possessing absorption lines in the same region. Also, a conclusion may be draws about the presence of at least two different sources for these species, whose emission is carried by wind eddies to the system inlet, which accounts for the large spiky fluctuations. The first source is evidently the volcanic gas from the several fumaroles present all around our installation. The second source is likely our own electric power generating engine, which was placed nearby and somewhat upwind. It is well known that gasoline engines produce large amount of CO due to incomplete combustion reactions.

 

Fig. 4. Concentrations of CO and CH₄ as a function of time while sampling ambient air at Solfatara volcano. The methane concentration appears anomalously large probably as a result of the methane emission from the fumaroles, while the CO concentration of about 200 ppb appears close to the normal ambient level. Both methane and CO are perturbed by sudden spikes which sometimes are perfectly correlated. This indicates the presence of independent sources (see text). The response time of the device obtained by taking the rise and fall time of the narrowest features is about 0.6s.

Download Full Size | PDF

Finally, the observation of these fast concentration changes in ambient air constitute a nice illustration of the possibility to use OF-CEAS for eddy-correlation trace measurements. Typically, the response time of trace measurements based on laser absorption techniques, when not limited by the rate at which absorption spectra are acquired to achieve sufficient sensitivity, is limited by the sample exchange rate inside the measurement cell. This is particularly the case with multipass cells whose volume is on the order of 1 litre. Our sample volume is only about 18 cm³: Considering a gas flow of about 200 sccm and a cell pressure of 80mbar, it is easy to estimate a gas exchange time of 0.4 s compatible with the CO concentration transient times observed in Fig. 4. While we did not try to push for faster measurements at the time of this campaign, in the laboratory we could raise the flow rate up to about 1000 sccm before observing perturbation to the cavity signal due to turbulence. In addition, the current 8Hz laser scanning rate can be increased by using smaller scans over only one or two absorption lines (thus making a compromise on the simultaneous observation of 3 species).

4. Conclusions and perspectives

The technique of OF-CEAS with its simple optical layout is well adapted to the realization of compact, low power, and affordable trace gas analysers. We demonstrated its application to in-situ volcanic gas monitoring. For such field measurements, we developed a mechanically stable high-finesse V-cavity design.

This work also illustrates a first in-situ application of GaSb diode lasers, capable of room temperature operation in the 2–2.7 µm range. Our laser sample is a DFB device able to perform single frequency mode-hop free scans around 2.33 µm (4298 cm^-1). In this spectral region strong absorption lines of several important atmospheric and volcanic trace species are available, with little interference from water vapour. No interference from other major volcanic gas constituents, notably CO₂, is present either.

With respect to the concentration values we measured in the Solfatara volcano emission, on the day of our measurements (10 march 2005) the CO trace is found to be 3 ppm (±10%), which agrees rather well with the previously reported 2 ppm value [4]. Instead, previous methane concentrations from the same work are much larger (250 ppm) than what found by us (76 ppm), which we believe to be within 10% of the real value. In particular we can exclude alteration of the concentration of this unpolar and insoluble species due to our fumarole dried gas sampling scheme. Since our measurements are made after water vapour removal, these concentrations are to be taken as relative to CO₂, this being the major remaining component (98%) in the dried gas. This discrepancy could also indicate an intrinsic large decrease of methane emission by Solfatara volcano over the 20 years elapsed since the previous determination [4].

Detection limits (1 standard deviation) achieved for CO, CH₄ and NH₃ are 16 ppb, 0.4 ppm, and 14 ppb respectively at a 8Hz rate. However, the methane absorption lines exploited here are not the most intense in the spectral range accessible with this laser. If we allow time (~10–20 s) to switch from an ‘optimal’ spectral region to another by changing the laser temperature, the detection limit for CH₄ can be lowered by a factor of 10.

Thanks to the small sample volume the time resolution was ~0.5 s, still limited by sample exchange time. For demanding applications such as eddy correlation measurements this can be easily improved to 0.1 s by increasing the flow rate, which was here only about 200 sccm. Sample volume could also be reduced (to about 10 cm³) and the scan rate increased.

It should be remarked that after choosing a convenient spectral region and determining those lineshape parameters which do not change during sample measurements (linewidths, relative line positions…), a constrained multiline Voigt fit is sufficiently fast that we could recently include it in our data acquisition program to deliver concentrations directly during measurement without degrading the data rate.

The performance of this instrument was recently significantly improved. A new set of high reflectivity mirrors (Layertec, Germany) yielded a 3 times larger cavity finesse while parasitic fringes were reduced by carefully damping stray light reflections (on black paper surfaces). The current noise equivalent absorption is close to 10^-9/cm on single scans (taken at 8Hz rate), which is practically the same we achieved at telecom wavelengths with a cavity of similar finesse [13].

Finally, OF-CEAS could be applied to detect other species of interest for volcanic monitoring. Considering the availability of DFB diode lasers in the range 1–2 µm and up to 2.7 µm HITRAN simulations permit to estimate, for example, the following detection limits: 1 ppm for SO₂ and H₂S (at 2.45 µm), 10 ppb for HBr (1.96 µm), 1 ppb for HCl (1.74 µm), and even 10 ppt for HF (again at 2.45 µm).