Introduction
Not only does subaerial CO2 Earth degassing affect the global geochemical carbon cycle (Burton et al., 2013), but it may
also pose a threat to human health in its vicinity, e.g., near volcanoes (Farrar et al., 1995; Hansell and Oppenheimer, 2004).
Central and southern Italy in particular is characterized by strong natural CO2 degassing that accounts for 10 % of
the global CO2 released by subaerial volcanoes (Chiodini et al., 2004). Caldara di Manziana (CdM, Fig. 1a) represents one
of the strongest of those degassing sites. CdM was probably formed by a hydrothermal explosion (Rogie et al., 2000) and still
preserves the shape of an ancient explosion crater with an almost flat bottom and rims elevated a few tens of meters.
A means to assess the gas hazard in such areas is the quantification of the CO2 release and the investigation of the
CO2 dispersion into the atmosphere through numerical modeling (Costa et al., 2008; Chiodini et al., 2010). Here we
intercompare modeled and measured CO2 amounts from open path FTIR spectrometry (OP-FTIR) and 1.6 µm
differential absorption lidar (DIAL) at CdM for the following two reasons.
Firstly, the DIAL is a prototype that currently undergoes first validation surveys. In this context, the well-established OP-FTIR
served as a reference instrument for the DIAL. The second motivation is to explore whether for the given situation an Eulerian
dispersion model is able to simulate CO2 amounts that compare to measured CO2 concentrations. So far we know of
only little and limited work on directly comparing modeled and measured volcanic CO2 amounts (e.g., Granieri et al., 2013,
2014). While both the instruments measure column averaged CO2 concentrations the model uses input data from point
measurements. For the background atmosphere agreement between point and path averaged CO2 amounts is expected (e.g.,
Gibert et al., 2008). In this case, using input data from point measurements to simulate column averaged CO2 amounts that
are in line with measured CO2 column amounts can assumed to be straightforward. At CdM, however, we are sensing a highly
non-uniform, non-steady CO2 concentration, comprising of both diffuse and vented CO2 degassing, which is
challenging this kind of comparison and thus makes it particularly interesting.
Instruments and methods
OP-FTIR is an established remote sensing technique to measure CO2 path amounts, including volcanic CO2 (Naughton
et al., 1969; Burton et al., 2000). The OP-FTIR used is a MIDAC M4406-S (MIDAC Corporation, Irvine, USA) with a ∼20 mrad field of view (FOV). Other main components are the Stirling-cooled MCT detector, which is sensitive between 600
and 5000 cm-1 and allows operation without liquid nitrogen cooling (e.g., La Spina et al., 2010), a 0.5 cm-1
resolution mid-infrared interferometer, the signal-processing electronics and an infrared (IR) light source.
The IR source is located at the other end of the open measurement path and can be either natural, e.g., volcanic magma, or
artificial. The measurements at CdM were performed in bistatic active mode using a high emissivity, 24 W SiC element
operating at a temperature of ∼1200 K placed at the focus of a parabolic reflector with 120 mm
diameter. Light from the source traverses the measurement column and enters the interferometer. As one of the interferometer
mirrors is translated an interferogram is being produced, which is Fourier transformed to yield an absorption spectrum. Through
fitting the latter with a forward modeled spectrum the path averaged CO2 concentration can be inverted for (Burton et al.,
1998). The OP-FTIR provides corresponding CO2 concentrations on average every 22 s. The necessity of aligning the
instrument with a light source at the end of the path limits the flexibility of the approach, for instance, when employed at
a volcanic site. Many volcanoes are quiescent and thus offer no magma as light source and may otherwise pose a challenging
environment to align the OP-FTIR with an artificial IR source, e.g., due to opaque or toxic volcanic gases. Even at good
visibility alignment becomes impractical at ranges beyond 200 m.
To have an instrument deployable at volcanoes that overcomes these drawbacks we are developing a DIAL in the framework of the
European Research Council funded project CO2Volc, which as an active remote sensing technique carries its own light
source. The prototype is based on integrated path differential absorption (IPDA, Amediek et al., 2008; Kameyama et al., 2009;
Dobler et al., 2013). As a dedicated volcanology tool the instrument is geared towards maximum compactness and transportability in
the field. The compact DIAL prototype is described in detail in Queißer et al. (2015). Table 1 summarizes its key
parameters. The instrument uses two narrow line width fiber lasers (NKT Photonics, Birkerød, Denmark) operating at two
corresponding wavelengths followed by a fiber amplifier stage. The wavelength pair was chosen following Amediek et al. (2008). The
ON wavelength (1572.992 nm) corresponds to a rotational line of the vibrational transition 0000 → 2201 and
hence to a local CO2 absorption maximum of CO2, while for the OFF wavelength (1573.160 nm) there is
negligible CO2 absorption (Rothman et al., 2012). Photons of one wavelength at a time are alternatingly emitted at
a 1 kHz switching rate. The laser light transverses the measurement column, is backscattered by a hard target, such as the
ground, traverses the path again and is received by a commercial Schmidt–Cassegrain telescope with 200 mm
diameter. A detection unit comprising of a PIN photodiode and a high gain transimpedance amplifier converts the photon flux to
a voltage that is being digitized at 50 kSampless-1 with high resolution (24-bit). By computing the ratio of the
received light intensities associated with the ON and the OFF wavelengths we obtain the differential optical depth. The latter
along with knowledge of the CO2 absorption cross-section, moist air number density and the path length permits to directly
retrieve the path averaged CO2 concentration. The CO2 absorption cross-section is calculated using spectroscopic
data from the HITRAN 2012 database as well as air temperature and pressure data. Temperature, pressure and relative humidity
needed for the air number density are taken from a meteorological station mounted at the edge of CdM (Fig. 1a).
The integration time of the DIAL per measured CO2 concentrations value is 1 s per wavelength, so 2 s in
total, corresponding to 50 000 acquired samples for the ON and 50 000 samples for the OFF wavelength. In the post processing the
data are normalized by the transmitted signal strength to correct for laser power fluctuations and averaged to arrive at a mean
intensity ratio, yielding a CO2 concentration value each ∼3 s. Depending on the atmospheric state, we find
the serial wavelength emission scheme to be quite susceptible to atmospheric turbulences (Queißer et al., 2015). Consequently, to
mitigate the associated scintillation noise statistical dependence of subsequent data is reduced by using only a fraction of the
data and skipping adjacent data following Grant et al. (1988). We use data chunks of up to 10 ms s-1, leaving
99 ms in between the chunks.
The distance between the OP-FTIR and the IR source was 49 m and defined the length of the measurement column (path #1,
Fig. 1a). Figure 1a also shows a second path of 126 m length (path #2) we will refer to further below. Path #1
included the water pool (CO2 vent) at its end, which represents the target CO2 concentration probed and also the
strongest vented CO2 emission for both paths. While the OP-FTIR line of sight ran 0.5 m above ground, the DIAL
stood 1 m above ground aiming at the floor just behind the IR source, hence its optical axis crossed the OP-FTIR axis half
way (Fig. 1b). As the instruments were placed side by side, the lines of sights enclosed an angle of the order of 10 mrad,
crossing each other at the IR source.
The CO2 concentrations at CdM strongly depend on how the CO2 disperses. We performed numerical simulations of
CO2 dispersion using the Eulerian DISGAS code (DISpersion of GAS, Costa et al., 2005; Granieri et al., 2014). DISGAS is
coupled with the mass-consistent diagnostic wind model (DWM, Douglas and Kessler, 1990), which can describe wind fields over
complex topography. The model is able to reproduce atmospheric dispersion of a dense gas cloud released by point sources or
diffuse sources, accounting for topographic effects, soil-cover types, variations of atmospheric conditions and wind
direction. The spatial distribution of the CO2 sources is a very significant modeling input. To obtain it we acquired in
situ CO2 fluxes at 152 points throughout CdM (Fig. 1a) using the accumulation chamber technique (Chiodini et al., 1996;
Carapezza and Granieri, 2004). From these measurements we derive the total flux as well as a soil CO2 flux map using the
sequential Gaussian simulation procedure (sGs, Deutsch and Journel, 1998), following the approach by Cardellini
et al. (2003). Further input data are the topography, terrain roughness as well as meteorological data, such as wind speed, wind
direction, air pressure and temperature taken from the meteorological station. DISGAS outputs air CO2 concentrations at
horizontal x, y points and at heights selected by the user, i.e., it produces a 3-D map. The computational domain is
1.0km×1.0km with an x and y grid spacing of 10 m. Each CO2 concentration is
expressed as a value in excess above the local background CO2 concentration of 370 ppm, as measured during the
fieldwork far from the main degassing area. Time-lapse concentration 3-D maps are computed for every 2 min, constrained by the
maximum acquisition rate of the meteorological station.
From the 3-D concentration maps for each time step we derive spatial mean CO2 concentrations by averaging values along the
grid coordinates that are associated with the line of sight of the instruments and at a fixed height of 0.5 m above
ground. The resulting mean CO2 concentrations are directly compared with the path averaged CO2 concentrations
measured by DIAL and OP-FTIR.
Results and discussion
Modeling results
We find a large spatial variability of soil CO2 flux, between 4 to 27 500 gm-2d-1, with an average of
890 gm-2d-1. The average total CO2 output from 100 sGs simulations is 222±40 td-1,
with the strongest emissive areas located in the central and northern portion of the CdM crater floor (Fig. 2). Highest values
were measured in the vicinity of the water pool (CO2 vent, Fig. 2) where we estimated an emission of 6.3 td-1
of CO2 through measurement of the air CO2 concentration gradient above the pool.
Figure 3a shows one of the simulated time-lapse air CO2 concentrations maps for 15 October 2014, 16:00 CET, at
0.5 m height above ground. The wind was blowing from 235∘ N, which corresponds quasi to the mean direction until
16:00 CET. The model suggests higher CO2 concentrations in the northern sector of CdM. The NE inner wall of the crater likely
obstructs the dispersion of the plume. Figure 3b shows a map for 30 min after, when the mean wind direction veered to E (∼90∘ N) and remained constant until the end of the recording. The CO2 plume is stretched towards the open western
side of the caldera. Consequently, the CO2 concentration lowers inside the crater.
In contrast to this steady wind scenario, Fig. 3c shows a snapshot of the CO2 concentrations during the morning of the
following day, on 16 October, when the wind direction was fluctuant between 120 to 290∘ N, with a dominant
direction of ca. 203∘ N.
Comparing measured CO2 concentrations with each other
Figure 4a presents the measurement results between 15:50 and 17:00 (CET), 15 October 2014 along with the simulated CO2
concentrations that are associated with the model scenario Fig. 3a and b are referring to. The DIAL started recording when the
wind direction was already changing (around 16:00 CET). After 16 min it was paused to increase the integration time from 2 to
4 s, thus the gap in the time series.
The CO2 concentrations from OP-FTIR are on average 430 ppm above the local background value and in line with those
from the DIAL. Since the temporal resolution of the DIAL is on average 7 times higher than the FTIR sample spacing the FTIR series
is smoother. Nonetheless, the DIAL series trend follows the OP-FTIR trend, that is, it slightly decreases, with a mean
concentration 30 ppm higher during the first 10 min of DIAL recording than during the following 10 min. To obtain
a measure for the match between OP-FTIR and DIAL result we compute the residuals of the OP-FTIR series relative to the DIAL
series. Taking the mean of the residuals yields 44 ppm.
The FOV of the DIAL is ∼1/20 of the FTIR FOV, sensing a much smaller slice of the CO2 vent. Consequently, a number
of short lived CO2 dispersion effects and heterogeneities of the degassing that contributed to the FTIR signal have not
been sensed by the DIAL. This explains why at places large amplitude, small time-scale fluctuations are not
coinciding. Disagreement of this sort may furthermore be due to the different angles under which both instruments were probing the
gas plume.
Due to a malfunctioning of the DIAL reference detecting unit that occurred in the field we did not normalize the transmitted
signal for power fluctuations, which increases the uncertainty of the resulting CO2 concentrations. Therefore, the SD in
Fig. 4b also includes fluctuations of the laser power. The latter may fluctuate by up to 1 % of its mean corresponding to an
uncertainty of up to 255 ppm in excess of the background concentration, which would make laser power fluctuations by far
the highest source of uncertainty. However, the match with the FTIR data indicates that fluctuations of this magnitude were
absent during the measurement.
The serial wavelength switching scheme of the DIAL may lead to higher data variability at the presence of atmospheric dynamics,
such as turbulences or lifted water droplets from the CO2 vent. As a consequence, the latter may cause spike-like errors,
such as near 16:09 CET (Fig. 4a and b). In this regard, the experiment was important for the further development of the DIAL as it
showed evidence that at the current serial wavelength switching scheme atmospheric effects such as these may strongly affect its
performance. Owing to the flat terrain, noise due to atmospheric turbulence was significantly lower than at a previous test with
the same prototype (Queißer et al., 2015), but in general still contributed to the uncertainty of the CO2 amount (Fiorani
and Durieux, 2001). To minimize common mode noise caused by atmospheric variability the DIAL is currently modified to allow for
simultaneous sensing of ON and OFF related signal.
Comparing measured with simulated CO2 concentrations
For the beginning of the comparison, both the CO2 concentrations from OP-FTIR and from DIAL agree well with those
predicted by DISGAS (Model, Fig. 4a). In particular, the simulated CO2 concentrations decrease during the first ∼20 min coinciding with the major change in wind direction. Remarkably, the CO2 concentrations from OP-FTIR and
DIAL do so as well. However, after 16:10 CET their mean remains quasi constant and roughly 150 ppm above the modeled
values. Given the model grid spacing of 10 m and since the simulated CO2 concentration was very non-uniform after
the wind direction had changed (Fig. 3b), even a small mismatch between instrumental FOV and the model spatial domain may have led
to the systematic discrepancy we see after 16:10 CET in Fig. 4a. Under this conditions spatial mismatch coming from the GPS
coordinates of the instruments, which is of the order of 3 m, becomes relevant too.
For the higher time series frequencies, a great deal of the disagreement can be attributed to the different domains of the
approaches, in particular differences in temporal and spatial resolution between numerical model and instruments. Vented emissions
possess dynamics that are challenging to model. Rather than being in a constant flow the main CO2 vent is characterized by
pulsed degassing, which undoubtedly contributed to the CO2 concentration measured by the instruments and partly explains
peaks and troughs in the time series.
The difference in tempo-spatial resolution is furthermore associated with a different sensitivity of the model with respect to the
instruments to relatively rapid and small-scale changes in atmospheric conditions. The model assumes a simultaneous effect of 152
emissive points spread over CdM, including the strong contribution of the CO2 vent. These points are roughly 50 m
apart and only a couple of them are located close to the instrument measurement paths. Furthermore, the model has a spatial
resolution of 10 m horizontally and 0.5 m vertically. The instruments, on the other hand, are designed to sense
CO2 concentrations within a small solid angle corresponding to tens of cm at the end of their path, thus sensing a rather
narrow cone, which is slicing the target CO2 vent. Clearly, they are expected to be sensitive to small-scale fluctuations
of CO2 concentration the model does not account for. For instance, while the CO2 vent is associated with one grid
point only, in the area above the vent CO2 concentrations are governed by a number of short-lived processes, such as
lateral eddies pushing up air.
Furthermore, the model has a temporal resolution of 2 min, i.e., the acquisition rate of the meteorological station. In
contrast, the acquisition times of the DIAL and the OP-FTIR are of the order of seconds. DISGAS is geared towards resolving
variations in CO2 concentration linked to atmospheric pressure and temperature over hourly or diurnal time scales. It
performs worse at capturing sudden CO2 fluctuations over relatively short time scales of seconds to minutes associated
with changes in wind speed and wind direction. The instruments do this very well.
Despite the limitations of the Eulerian DISGAS code to reproduce dispersion in the proximity of the gas source, the numerical
simulations provide a satisfactory agreement with the measurements along path #1. Note that apart from the change in wind
direction the wind field for path #1 was quite steady and thus the atmosphere rather stable.
Figure 4a also shows the result of the second comparison associated with the morning after where the atmospheric situation was
quite different (Fig. 3c). The path was 126 m long (path #2, Fig. 1a) and included the main CO2 vent as
well. The OP-FTIR and the DIAL time series agree reasonably well, with a mean residual of the OP-FTIR series relative to the DIAL
series of 34 ppm. As for path #1, the maximum simulated CO2 concentrations reach 800 ppm. However,
while the latter have a mean of 480 ppm, the instrumental CO2 concentrations are around 200 ppm only and
so the match between instrumental and model result is poor.
To analyze this we consider the result of a second simulation, which accounts for the flux around the CO2 vent only,
therefore representing the lower limit of modeled CO2 concentration (Model CO2 vent, Fig. 4a). For path #1 this
result is on average closer to the default modeling scenario than for path #2, suggesting that the path #1 modeling result
is dominated more by the CO2 vent than the path #2 modeling result. An explanation can be found by confronting the path
lengths and the atmospheric stability. Unlike for path #1 no major change in wind direction occurred during the
acquisition. However, as mentioned above, the wind direction was swiftly fluctuating around its mean suggesting a less stable
atmosphere during the path #2 acquisition. This is consistent with the fact that the recording took place in the morning. In
the morning, the near-surface layer of the atmosphere is characterized by positive buoyancy due to the soil heating as opposed to
the later afternoon when air is negatively buoyant and turbulences are reduced. This suggests that effects contributing to the
instruments responses are associated with tempo-spatial scales below the model resolution. For instance, sampling the wind every
2 min was likely missing important changes in the wind direction that, however, governed the CO2 distribution sensed by
the instruments.
In addition, unlike for the shorter path #1, the target CO2 vent less dominates the CO2 concentration. So under
the given turbulent atmospheric conditions the measured CO2 concentrations were strongly governed by small-scale
dispersion processes and the vent concentration was being perturbed more by diffuse CO2 than at path #1. These
conditions are harder to account for by DISGAS than those for path #1 and so the agreement between measurements and model is
worse than for path #1.
In sum, the outcome of the comparison suggests that the match between modeled result and instrumental result is better for stable
atmospheric conditions and when the target vent probed by the instruments dominates the CO2 concentration within the
measurement volume. The result for path #2 illustrates the limits of DISGAS for this application.