In complex and urban environments, atmospheric trace gas composition is highly variable in time and space. Point measurement techniques for trace gases with in situ instruments are well established and accurate, but do not provide spatial averaging to compare against developing high-resolution atmospheric models of composition and meteorology with resolutions of the order of a kilometre. Open-path measurement techniques provide path average concentrations and spatial averaging which, if sufficiently accurate, may be better suited to assessment and interpretation with such models. Open-path Fourier transform spectroscopy (FTS) in the mid-infrared region, and differential optical absorption spectroscopy (DOAS) in the UV and visible, have been used for many years for open-path spectroscopic measurements of selected species in both clean air and in polluted environments. Near infrared instrumentation allows measurements over longer paths than mid-infrared FTS for species such as greenhouse gases which are not easily accessible to DOAS.
In this pilot study we present the first open-path near-infrared
(4000–10 000 cm
The cycling of carbon between Earth's surface and the atmosphere is
dominated by carbon dioxide (CO
Point measurements are sensitive to the immediate local environment, and may or may not adequately represent the mean concentrations over the grid-scale of the relevant atmospheric models in non-background environments. open-path (OP) measurements provide spatially averaged concentrations, by measuring an optical absorption spectrum along a path between a light source and the measuring instrument and retrieving component concentrations from the spectra. Spatial averaging at similar scales to those of the finest urban- and regional-scale models should be advantageous in combining measurements and models to deduce the strengths of localised sources and sinks of greenhouse gases. But how accurately can we measure such spatially averaged trace gas concentrations?
The longest established surface OP techniques (i.e. excluding satellite and
ground based total column measurements) are differential optical absorption
spectroscopy (DOAS), typically employing the UV and visible spectral regions
(Platt and Stutz, 2008), and open-path Fourier transform spectroscopy
(OP-FTS) in the mid-infrared (e.g. Tuazon et al., 1978; Russwurm and
Childers, 2002; Griffith and Jamie, 2006; Smith et al., 2011; Laubach et
al., 2013; Flesch et al., 2016; You et al., 2017). While DOAS can operate
over path lengths of several kilometres, suitable absorptions for accurate
and precise measurements of CO
Schematic drawing of the long open-path FT spectrometer and optical system. Radiation from the source is fed close to the focus of the telescope through the outer bundle of six fibres (blue) and transmitted across the open path. The return beam is collected by the central fibre (red) and focussed onto the input aperture of the interferometer. The modulated beam from the interferometer is detected by the InGaAs detector and the resultant interferogram is Fourier transformed to provide the long open-path spectrum.
The recent and rapid development of TCCON, the Total Carbon Column Observing
Network (Wunch et al., 2011) has shown that the near-IR spectrum, with
a ground based FT spectrometer and the sun as a source, is suitable for
highly accurate and repeatable (
The optical system is shown schematically in Fig. 1. The spectrometer and telescope were located in the rooftop observatory
of the six-storey Institute of Environmental Physics (IUP) building on the
University of Heidelberg campus in urban Heidelberg, (49.4172
Aerial view of Heidelberg and the 1.5 km measurement path. IUP is the Institute of Environmental Physics (FTS and telescope, in situ measurements), PI is Physics Institute (retroreflector), PS is power station. The measurement path is mostly over residential areas. There is an extensive small-agricultural area to the north and northwest.
Typical NIR long path spectrum, recorded 1 October 2014.
The return beam from the fibre was focussed by a 75 mm focal length
NIR-coated lens into the 1 mm entrance stop of the FT spectrometer (IRcube,
Bruker Optics, Ettlingen Germany), which had a quartz beam splitter and InGaAs
detector optimised for the NIR spectral region (3800–10 000 cm
The rms spectral signal : noise ratio (SNR) was determined at 6300–6500 cm
Measurements reported here were continuously collected from 10 July to 4 November 2014.
Spectra were recorded with a resolution of 0.55 cm
Atmospheric pressure and temperature for the measurement path are required for the spectrum analysis and to calculate air density. These were measured and averaged over the period of each spectrum measurement by an electronic barometer (Vaisala PTB110) and LM335 diode, co-located with the FT spectrometer. The acquisition of spectral data, pressure and temperature, shutter control and real-time spectrum analysis were executed automatically by the software available for the Ecotech Spectronus in situ FTIR analyser (Ecotech, Knoxfield, Australia). Initially the IUP weather station temperature and height-adjusted pressure were used in the spectrum analysis; the weather station temperature was subsequently replaced by the path-averaged temperature derived from the spectra themselves, as described below.
At the IUP end of the open path, air from a roof-level inlet on the IUP
building was sampled and analysed continuously with an in situ trace gas
analyser described in detail
by Griffith et al. (2012), Hammer et
al. (2013), and Vardag et al. (2015). This analyser is based on an FTIR
spectrometer operating in the mid-IR and provided simultaneous high
precision measurements of CO
Standard measurements of pressure, temperature, humidity, wind speed, wind direction and solar radiation were obtained from the IUP weather station (located on the roof of the building) as 5 min averages and interpolated to the times of the open-path measurements.
Path averaged trace gas mole fractions were retrieved from spectra by iteratively best-fitting a calculated spectrum to the measured spectrum. The forward model, MALT (Griffith, 1996) calculates the transmission spectrum from a set of input parameters including absorption line parameters, trace gas amounts, pressure, temperature and path length as well as instrument parameters including resolution, apodisation function, line shape, spectral shift and a five-term polynomial fit to the continuum, which in these single beam spectra is generally not flat. The line parameters are based on Hitran 2008 (Rothman et al., 2009) updated by Toon and associates for the GFIT software used throughout TCCON (Wunch et al., 2015). The inverse model uses non-linear least squares following the Levenberg–Marquardt algorithm (Press et al., 1992) to retrieve the path averaged concentration of each trace gas species. The path averaged concentrations are converted to mole fractions by dividing them by the concentration of air determined from pressure and temperature. More details are given by Griffith et al. (2012).
Details of spectral windows used for fitting.
* In O
Details of the spectral windows used for the NIR long path analysis are
summarised in Table 1 and typical fits for spectral
regions used to retrieve O
Typical fits for
The fibre optic coupling between telescope, source and detector introduces
repeatable fringing and interferences in the measured spectra at about 1 %
of the measured signal intensity. These spectral structures can be seen in
the residual plots of Fig. 4 and are quite
reproducible over periods of days to weeks. They are larger than the
underlying detector noise but much less than the trace gas absorptions, at
least for CO
Comparison of IUP meteorological station temperature (red) and
spectrum-derived path averaged temperature (blue) for an illustrative period
of 4 sunny days. The differences are plotted in green and range from 0 to 6
Background spectra of stray light measured hourly by blocking the source had
intensities up to 1 % of those of the open-path spectra, maximising in the
early morning and late evening when the solar elevation was low and
direction roughly parallel (east to west) to the open path. Scattered solar stray
light collected by the FTIR spectrometer has an effective atmospheric path
of
Significant differences of up to 5
To check the instrument line-shape function (ILS) of the FTS, we followed
Frey et al. (2015), by measuring the spectrum of water vapour
in a short-path reference spectrum over a path length in air of approximately
3 m and fitting it using both MALT and Linefit (Hase et al., 1999)
programs. Assuming the nominal field of view (FOV) of the FTS is 7.2 mrad,
we found a linear drop in modulation efficiency to 0.67 at the maximum
optical path difference. Alternatively, setting the modulation to its
nominal value of 1.0 and fitting the field of view, we retrieved an
effective FOV of 10.8 mrad. The effective ILS width is thus approximately
30 % broader than the nominal value for a perfect optical system. This is
consistent with the short focal length optics and aberrations in the compact
optical system. The ILS is shown in Fig. 6. The
full width at half height is 0.58 cm
Retrieved instrument line-shape function for the IRcube FTS at
nominal 0.55 cm
All raw mole fractions (except water vapour) were converted to dry air mole
fractions using the path-averaged water vapour amount retrieved from the
same spectrum:
Retrieval of the O
Measured O
The mean mole fraction (excluding evening scattered sunlight anomalies) is
0.217, a bias of
Water vapour provides a further check of the FTS measurements against
weather station humidity. (The in situ analyser does not measure ambient
water vapour as the sample is dried for measurement.) H
Water vapour,
As is the case for O
The bias-corrected OP and calibrated in situ measurements are shown in Fig. 9, together with their differences. Fig. 10 shows the differences plotted (a) against wind speed, (b) against wind direction and (c) as a histogram. The data are discussed in Sect. 4.
Open-path (blue), in situ (red) and difference (OP – in situ,
coloured by time) measurements of CO
CO
Similar analyses for CH
Open-path (blue), in situ (red) and difference (OP – in situ,
coloured by time) measurements of CH
CH
Absorption by the UV quartz retroreflectors below 4600 cm
N
Allan deviation plots for
Allan deviations for open-path and in situ measurements and their differences. The Allan deviation analysis is taken over the period 11 August, 06:00–27 August, 18:00 when diurnal variations were least.
* In the “smooth-subtracted” and standard deviation rows, a smoothed curve through all the data has been subtracted from the raw data to remove the gross atmospheric variability and approximates the measurement noise. See text for further details.
Table 2 and Fig. 13 show
Allan deviations (ADs, the square root of Allan Variance, Werle et
al., 1993) for open-path and in situ CO
For both open-path and in situ CO
For CO
For CH
Raw OP measurements are biased high relative to WMO-calibrated in situ
measurements at the IUP (western) end of the path,
Sensitivity of retrieved mole fractions to retrieval inputs in the OP-FTIR measurements. Each input parameter or choice was varied by an estimate of its uncertainty in the MALT spectrum analysis and its effect on retrieved mole fractions calculated.
Table 3 presents the sensitivity of mole fraction retrievals from the spectra to realistic
uncertainties in input parameters and choices in the retrieval. Details are
given in the caption to Table 3. There is no
dominant single source of uncertainty; the main contributors are derived
from uncertainties in spectroscopic data, the instrument line shape, stray
radiation, and details of the fitted spectral window. A simple quadrature
sum of the estimated systematic errors (4.5 % for CO
Data from recent work using broadband DOAS and laser-based long open-path techniques are shown for comparison in Table 4. Compared to conventional DOAS with a grating monochromator, array detector and the same long path fibre-telescope optics (Sommer, 2012; Saito et al., 2015; Somekawa et al., 2011), the FTS system achieves greatly improved repeatability. Compared to more recent work with dual frequency comb laser spectroscopy (Rieker et al., 2014; Waxman et al., 2017), the repeatability is less by about a factor of two. The frequency comb was operated over a longer path length with shorter measurement times and achieved lower bias when compared to co-located in situ measurements, but at this stage of development is less portable for remote field measurements and applicable only to a narrower range of species. The FTS setup has advantages in terms of mobility and costs.
From the preceding discussion, measured differences between open-path and in
situ measurements are only
For bias-corrected CO
Comparison of repeatability and bias of long path techniques in the NIR region.
For bias-corrected CH
The OP–in situ differences and geographical scales of these measurements approach the accuracy and resolution of developing regional-scale models, such as the Weather Research and Forecasting model (WRF) in high-resolution mode (Viatte et al., 2017). A detailed high-resolution modelling analysis of the measurements presented here might help in interpreting the observed OP–in situ differences, but is beyond the scope of this paper.
This study was made with available instrumentation in a restricted timeframe
as a pilot study of the open-path FTS technique in the NIR and did not
optimise some aspects of the measurements. Several options are available to
improve the accuracy and precision of the OP-FTS-NIR measurements:
Interferences from stray radiation: especially at low solar elevations,
background (stray) radiation is modulated and detected by the interferometer
and leads to broad enhancements and spikes in measured concentrations. This
can be almost entirely removed by reversing the source and detector in the
optical system shown in Fig. 1, first modulating
the source in the interferometer before transmission over the open path.
With this option stray environmental radiation such as direct or scattered
sunlight is viewed directly by the detector and not modulated by the
interferometer; it does not contribute to the Fourier-transformed infrared
spectrum. This option was not possible with the available optics and
spectrometer for this pilot study, but will be incorporated in the next
build. With the present system, increasing the frequency of the background
stray light measurements (1 per hour in this work) would allow better
correction for stray light interferences due to short term variations in
stray radiation, but at the cost of lower precision, measurement time and
duty cycle. Increased optical throughput: using a brighter source and/or larger
telescope and retroreflector area will translate directly into lower
measurement noise and improved repeatability. This is particularly true of the
retroreflectors, which had a total area of around 510 cm Extension to include CO: for urban studies the measurement of CO is
advantageous, both for its intrinsic interest and as a tracer for combustion
sources of other trace gases. In this work we used available UV quartz
retroreflectors optimised for UV/vis DOAS measurements. The transmission of
UV quartz cuts off below 4500 cm
We have introduced a long open-path Fourier transform spectrometer operating
in the near-infrared over a 3.1 km return path in open air. The system is
able to make measurements of several species simultaneously by virtue of the
broadband nature of the spectroscopy. We have demonstrated measurements of
CO
We observe significant differences of the order of a few ppm for CO
A text file tabulating all data presented in this paper is available in the Supplement.
The authors declare that they have no conflict of interest.
This work was carried out as a sabbatical leave project by David W. T. Griffith at the Institute for Environmental Physics, University of Heidelberg. David W. T. Griffith thanks Ingeborg Levin, Ulrich Platt and members of the DOAS and carbon cycle groups for their contributions and collaboration in providing the laboratory and long path optical systems for the study. Geoff Toon, JPL, provided updated 2015 versions of GFIT line parameters.Edited by: Pierre Herckes Reviewed by: two anonymous referees