Carbonyl sulfide (COS) has been suggested as a useful tracer for gross
primary production as it is taken up by plants in a similar way as CO
Carbonyl sulfide (COS) has been suggested as a potential tracer for
photosynthetic CO
Several past studies on COS have relied on discrete (flask) samples analyzed
with gas chromatographic mass spectrometry
The required measurement precision (in this study we define precision as the
standard deviation over a 2-minute period) for studies of exchange
processes of COS and CO
Measurement instruments for long-term atmospheric trace gas concentration
monitoring need to meet different requirements than, for example,
eddy-covariance measurements. The eddy-covariance technique requires high-frequency data (
Before the actual deployment of the instrument in the field, we performed laboratory tests to assess the accuracy and traceability of the QCLS measurements and to develop procedures for applying corrections as needed. Here we describe the laboratory tests and we give detailed information about the instrumentation and field setup.
The QCL Mini Monitor that we use is a tunable diode laser absorption spectrometer (TILDAS) using a single continuous-wave quantum
cascade laser (Alpes Lasers), which is cooled with a Peltier element to
Simulated transmission spectrum of ambient concentrations of COS
(500 ppt), CO
Response curves determined with NOAA/ESRL calibration standards,
including the residuals of the fit. The NOAA/ESRL calibration standards are
calibrated on NOAA-2004 COS scale, WMO-X2007 CO
The instrument consists of a 0.5 L astigmatic Herriott style multi-pass
absorption cell
The TDLWINTEL software has the option to store raw spectra. These spectra can
later be used for re-analysis using the so-called “playback” mode of the
software. The spectral parameters (line shape and position) for the fits are
taken from the HITRAN database
To allow comparison of QCLS measurements with other instrumentation and across different sites requires traceability to a primary scale. Laboratory tests were conducted to characterize the response of the instrument against ambient air standards from NOAA/ESRL, which were subsequently used to transfer the calibration scale to working standards. Moreover, we performed tests to understand the frequency required for background and reference measurements to ensure reliable and accurate results.
To characterize the response of the instrument for COS, CO
CO
COS calibration values for our real-air working standards as obtained from three different calibration approaches: (1) with response from the two NOAA/ESRL calibration standards and a zero point, (2) with the two NOAA/ESRL calibration standards and the curve not forced through a zero point and (3) using a single bias correction. The NOAA/ESRL calibration standards have COS concentrations 447.8 and 486.6 ppt. The calibration measurement was repeated three times; results are shown as the average over the three measurements and uncertainties indicate the standard deviation over the three measurements.
In this study we use working standards to represent high-pressure cylinders
that are calibrated against NOAA or WMO standards in our laboratory. The
working standards are used for two purposes: (1) to correct for instrument
drift during field measurements and link the measurements to the NOAA or WMO
scales (in this case we refer to these standards as reference standards) and
(2) to assess the accuracy of the measurements (in this case we refer to
these standards as target standards). The working standards used in this study
are aluminum gas cylinders (Luxfer, max. 200 bar) filled to high pressure
with ambient air at the Center for Isotope Research of the University of
Groningen using an oil-free air compressor (RIX SA-3) and are used in
combination with two-stage regulators (Scott Specialty Gases, model 14). The
working standards differ from the NOAA/ESRL calibration standards in the
sense that they are uncoated. To trace the working standards that we used in
the field we refer to these with numbers 1, 2 and 3 throughout the
paper. Other working standards that were used in this paper but not
in the field are referred to as “A”,“B”, “C” and “D”. Using the linear
regression curve that we found in Fig. 2, we determined mole fractions in
these working standards by considering response curves derived from
measurements of the NOAA/ESRL calibration standards. Results of calibrations
of three working standards with the QCLS are shown in Table 1 for CO
As we saw in Sect. 2.2.1, the response curve of COS is difficult to determine due to the lower instrument precision and the narrow COS concentration range of the NOAA/ESRL calibration standards. These factors introduce uncertainty in assigning values to the working standards that will be used on-site, especially for those having COS mole fractions outside of this range. Besides that, the number of available calibrated standards used to transfer the scales to other cylinders may be limited in labs, especially for COS, as this gas is usually not one of the standard measured species. The question is therefore what method is suitable to transfer the scale to the working standards for COS. To test this, we re-analyzed calibration measurements in different ways: (1) with response from the two NOAA/ESRL calibration standards and the curve forced through a zero point, (2) with two NOAA/ESRL calibration standards and the curve not forced through zero and (3) with a single bias correction using a NOAA/ESRL calibration standard that has the concentration closest to the working standard.
Standard deviation (SD) over minute-averaged data of COS, CO
Table 2 shows the assigned values on the working standards considering the different approaches. The calibration measurement was repeated three times and results are here shown as the average over the three measurement. The fact that the cylinder values need to be extrapolated causes larger deviations compared to when the curve is forced through a zero point (see, e.g., cylinder “A” in Table 2). Therefore, when working standards have concentrations outside the range covered by the NOAA/ESRL calibration standards, method 1 is preferred because it avoids extrapolation of the calibration curve by using the zero point. However, as we saw in Sect. 2.2.1 it is difficult to accurately determine a calibration curve due to the lower precision of the measurements and the narrow COS concentration range of the NOAA/ESRL calibration standards, which holds for both methods 1 and 2. In this study we therefore calibrate cylinders and field measurements under the assumption that the data follow a calibration curve with a slope equal to 1 (in Sect. 2.2.1 we saw that the average slope of different calibration experiments was equal to 0.99) and we apply a single bias correction, which is applied in Table 2 as method 3. Theoretically, methods 1 and 2 would give the results closest to reality, but here we observe that the difference of method 3 against methods 1 and 2 is small (on average 1.3 ppt) and the result is often in between those of methods 1 and 2. We therefore consider the single bias correction as a good compromise when a calibration curve cannot accurately be determined.
Background measurements with dry nitrogen are required to reduce the effect
of curvature of the baseline spectra and are typically done every 2, 5 or 30 min by other users of similar QCLS analyzers
Left: water vapor experiment where cylinder air was humidified with
wet silica gel and no water vapor correction was applied to the data. Right:
wet over dry ratio of COS, CO
The concentration measurements of gases determined through light absorption
spectrometry can be affected by water vapor in the sample air in two ways:
(1) by spectroscopic effects (enhanced pressure broadening or direct spectral
interference), which will directly modify the absorption spectrum, and (2) through
dilution of the sample air, which linearly depends on water vapor
concentrations. The water vapor interference can be prevented by drying the
air before measurement. However, this requires adding a drier to the sampling
inlet lines and, depending on the drier, it can require additional maintenance,
which is not favorable for unattended autonomous measurements. As the QCLS
includes measurements of H
Different water vapor correction strategies based on the software
fitting parameters (standard or split fit) and with the TDLWINTEL correction
turned on or off. When the TDLWINTEL correction is turned on the values
indicate the broadening coefficient used for the different species, and when it is turned off the values indicate the slope of the correction
curve as determined in Fig. 3 with
* No broadening coefficient could be derived; however, we found that for a broadening coefficient of 1.5 with the standard fit the slope of the curve for COS is equal to 0.031, which can be applied as an extra correction on top of the TDLWINTEL correction.
Using the playback mode we tried to find the most optimal water broadening
coefficients to sufficiently correct for water vapor using TDLWINTEL. We did
this for the three different water vapor tests and for both the standard fit
and the split fit. For the standard fit we could not find optimized
broadening coefficients for COS because turning the TDLWINTEL correction on
caused an opposite correction and resulted in larger deviations from the
assigned cylinder value due to the effect of the H
Different water vapor correction strategies can now be considered and are
summarized in Table 4. An appropriate direct water correction for COS is not
possible with the combination of the standard fit and the TDLWINTEL
correction on. However, a correction curve can still be applied to these data
with a curve that is determined with the same broadening coefficients as are
used for the original data to be corrected. Here we found that for a
broadening coefficient of 1.5 with the standard fit the slope of the curve
for COS is equal to 0.031. We continued to test the performance of the water
vapor correction based on the standard fit and the TDLWINTEL correction off.
We applied the correction curves to field measurements over a period of 35
days in March and April 2015. In Sect. 3.2 we will compare the dry air mole
fractions with measurements of a collocated CRDS for CO
In July 2014 we deployed the QCLS in the field for measurements at the
Lutjewad monitoring station in the Netherlands (60 m; 6
For our setup, every hour starts with a measurement from a reference and target standard (3 min each). Subsequently, the system alternates between three measurement heights where every height is measured for 8 min, meaning that every height is measured twice an hour. The reference standard was measured every half hour to remove instrument drift. In March 2015 we measured an extra reference gas once every hour, which was used to test the stability of the instrument response over a period of 35 days (the results will be shown in Sect. 3.1). Background measurements were done every 6 h with dry nitrogen over 60 s. Before the actual background measurement is done, the cell is first flushed for 2.5 min to make sure that water vapor is removed from the cell by 99 %.
We noticed that under lab conditions, when the temperature is controlled
within
Schematic overview of the instrument setup for tower profile measurements at the Lutjewad monitoring station. The pressure sensor and pump with shut-off valve were added for flask measurements only (see Sect. 2.6).
Besides testing the QCLS for continuous in situ measurements, we explore the
suitability of the QCLS for measuring air from flasks. We developed a means
to analyze flasks; this allowed a comparison between the QCLS and GC-MS
measurements from the NOAA Boulder laboratory
Mean concentrations of hourly measurements of working standards conducted while the instrument was in the field from August 2014 until April 2015 together with electronics temperature. The data shown here are uncorrected data and are not calibrated with a response curve. The concentrations are therefore not necessarily close to the assigned cylinder values. The solid vertical line at 7 January 2015 indicates the moment when we changed the setup with a solenoid valve to a Valco valve for switching to nitrogen, as the solenoid valve was leaking. The dashed vertical line at 25 March 2015 indicates the moment when we improved the temperature stability by actively cooling the electronics section and putting the analyzer in an enclosed box, which resulted in substantially smaller temperature fluctuations than before. From this moment onwards an extra working standard was also measured every hour to ascertain if the instrument response was stable over a period of 35 days (orange). The gap in the data record in December 2014 and February–March 2015 is because of tests with the QCLS in the laboratory.
Left: standard deviation over 2 min of hourly measurements of
one of the working standards as measured from August 2014 until April 2015.
The gray (black) colored data indicate the period after (before)
modifications were made to improve the temperature stability (such that
Two Allan deviation plots of COS retrieved from cylinder air measurements during two laboratory experiments under similar conditions on the same day. The Allan deviation plots show that the lowest random noise level of drift is not constant over time.
We also measured dry air flask samples to test if the water vapor correction
that we determined in Sect. 2.3 sufficiently removes the effect of water
vapor on calculated mole fractions. To that aim, flasks were filled to
ambient pressure as part of a standard flask sampling routine at the Lutjewad
station
We assessed the measurement uncertainty and accuracy with the hourly
measurements of the reference and the target gases over the period from
August 2014 until April 2015. As mentioned previously, each reference and
target gas was measured for 2 min. The mean value of the hourly
instrument-reported and uncorrected 2-minute measurements are shown in Fig. 5. In Fig. 6 the standard deviation of these measurements is shown. Figure 5
also includes the electronics temperature of the QCLS. The cylinder
measurements show that concentrations can drift substantially; i.e., COS
concentrations easily vary by 50 to 100 ppt. However, on the long term,
concentration changes are not correlated with temperature, which changed by
13 K throughout the year. The concentration shift on 7 January 2015
(especially visible in CO
Figure 6 shows the 1-second standard deviations of the hourly reference gas
measurements between August 2014 and April 2015. It is clear from Fig. 6 that
the instrument precision cannot be captured with one single value due to its
variation. For the period marked in red the room temperature was
characterized by rapid changes caused by an air conditioner. This was a
period in which the instrument precision was adversely affected, indicating that
temperature stability influences the instrument precision. In the right plot
of Fig. 6 the histograms of the standard deviations are shown for the periods
before and after improvement of the temperature stability, with the black/gray
colors corresponding to that in the left figure. The data show that for COS
and CO
Allan deviation plots are an effective way to show how far the random noise
level can be reduced by averaging and at what timescale the drift effect
starts
Uncertainty contributions and the overall uncertainty for
measurements of COS (447.8–486.6 ppt), CO
Mole fraction offsets in target standards after application of
(left) corrections with a fixed response curve for CO
The overall uncertainty of the measurements consists of uncertainties
associated with the scale transfer, water vapor corrections, drift
correction and the measurement precision. Table 5 summarizes the different
uncertainty contributions as well as the overall uncertainty for measurements
of COS, CO
We have now presented the total uncertainty for long-term concentration
monitoring at an atmospheric station, but not all uncertainties are relevant
for every type of analyses. If the data are to be compared across different
sites, then the data should be on the same scale and the accuracy of the
measurements is important. However, if data from the same site and the same
instrumentation are compared, for example to do flux-gradient analysis, then
there is a less stringent need for accuracy and the short-term precision is
more important. If the uncertainties related to the transfer of the scales
are not taken into account, then the total uncertainty would be 6.0 ppt for
COS, 0.13 ppm for CO
Additional to the uncertainties presented in Table 5, we have observed that
COS can decrease over time in uncoated aluminum cylinders. First, we did not
find indications that our working standards drifted during field measurements
at the Lutjewad station; calibrations in July 2014 and March 2015 showed a
decrease of only 2.2 and 1.1 ppt for cylinder nos. 1 and 2 over this 8-month
period, which is well within the measurement uncertainty. However, a
re-calibration in November 2015 showed a decrease of 18.2 and 24.1 ppt in
these cylinders, a decrease with a rate of 2.3 and 3 ppt per month, while
cylinder no. 3 only changed by 1.9 ppt. Cylinder no. 3 is not different from nos. 1
and 2 (they are all uncoated aluminum cylinders), but cylinder nos. 1 and 2
were stored with a pressure of 25 and 40 bar, which is lower than 130 bar for
cylinder no. 3. We cannot confirm if the drift in the cylinders is related with
the cylinder pressure; however, it is the only difference that we were able
to find. Experience with cylinders over the past 15 years at NOAA indicates
that COS in Aculife treated aluminum cylinders is typically much more stable
than untreated aluminum cylinders. A potential method to improve COS
calibrations is to calibrate these using a ppb-level standard accurately
diluted to a range of desired COS concentrations
Our COS measurements are reported on the NOAA-2004 scale and can be compared
to the observations from the global network of NOAA/ESRL
Flask sample measurements at the QCLS in Groningen (left) and GC-MS by NOAA/ESRL (right). Four paired flasks were filled with dry air from the two NOAA/ESRL calibration standards and two working standards, which were calibrated as in Sect. 2.2.2. The flask pair measurements are averaged and shown as the deviation from the assigned cylinder value. The first two cylinders are those calibrated at NOAA/ESRL and have an assigned value of 447.8 ppt (NOAA no. 1) and 486.6 ppt (NOAA no. 2). The last two cylinders were used as working standards in Lutjewad and have assigned values of 467.6 ppt (no. 3) and 455.5 ppt (no. 4). The error bars show the precision. For the QCLS two measurements were done: in April 2015 (orange) and in January 2016 (blue). The GC-MS measurement at NOAA/ESRL was performed in May 2015.
In this section we make different measurement comparisons. First, we compare
COS flask measurements using the QCLS with those analyzed with the GC-MS at
NOAA/ESRL. Second, we compare QCLS in situ and flask measurements of COS in
order to test if the water vapor correction that we determined in Sect. 2.3
sufficiently removes the effect of water vapor on calculated mole fractions.
Third, we compare the in situ CO
For the QCLS and GC-MS instrument comparison an overview of the measurements
is given in Fig. 9, where the flask pair measurements are averaged and are
shown as the deviation from the assigned cylinder value (see also Sect. 2.6).
The precision within the flask pair is shown by the error bars. The
comparison demonstrates the uncertainties associated with the transfer to the
NOAA scale (see Table 5). For the first measurement at the QCLS (orange),
three of the four flask pairs are within 1.5 ppt of the assigned value and
one flask pair deviates by 7 ppt. This is similar to the GC-MS measurements
at NOAA where one of the four flask pairs deviates further from the assigned
values. However, the deviating flask pair measured by the GC-MS is not the same as the one deviating at the QCLS.
For the second QCLS measurement (blue) one flask pair has drifted on average by 12 ppt, where other flasks remained stable within 2.5 ppt. It is unclear why the
two flasks have drifted as all flasks were filled, measured and stored in the
same way; however, we have kept the pair to monitor the potential drift in
the future and to find out what has caused the drift. Moreover, the
consistency of the flask measurements between April 2015 and January 2016 is
dependent on the stability of the reference standard which was used for all
flask measurements. Although we observed that COS can drift in cylinders, we
did not find indications that the particular reference standard used for this
analysis drifted over the 9-month period. The flask pairs have a mean
deviation from the assigned cylinder values of
Minute-averaged measurements of COS at the Lutjewad measurement tower (60 m) between 25 March and 29 April 2015 as measured in situ by the QCLS (blue), compared with dry air flasks samples measured at the QCLS (orange). The left plots show the time series; the right plots show the difference between the in situ and flask measurement.
The next comparison in Fig. 10 considers the in situ COS measurements between
25 March and 29 April 2015 and 11 dry air flask samples measured by the same
QCLS. The flasks are flushed for an hour before closing, but because of
mixing in the flask we assume that the flask sample represents the last 15 min; therefore we average the in situ measurements over these 15 min.
The flask samples have an inlet at 60 m height, but because the in situ 60 m measurements only cover the period between 9 and 15 min before flask
closure we also include 40 m measurements to cover the last 9 min before
the flask is closed. The average difference of COS mole fractions between the
40 and 60 m level is 0.7
Minute-averaged measurements of CO
Hourly-averaged measurements of COS, CO
COS seasonal cycle of four sites: Wisconsin, USA (LEF); Mauna Loa, USA (MLO); Mace Head, Ireland (MHD); and Lutjewad, the Netherlands (LUT); the data of the latter site are presented in this study. COS mole fractions for the LEF, MLO and MHD sites were measured from flask samples at a GC-MS by NOAA/ESRL (Montzka et al., 2007). The NOAA/ESRL data are shown as flask pair means from individual sampling events. All NOAA measurements are plotted as function of time of the year and cover a period between 2000 and 2015 for LEF, MLO and MHD. In situ COS measurements with the QCLS at the Lutjewad site during 2014–2015 are shown as daily averages (black). A two-harmonic seasonal cycle is fit through the data.
Last, we compare the minute-averaged QCLS measurements for CO
The COS, CO
The location of the Lutjewad station along the coast of the province of
Groningen in the Netherlands allows the measurement of marine background air
during northerly winds and continental air during southerly winds
In this study we have tested a QCLS for its suitability for making accurate
and high-precision measurements of COS, CO
The QCLS was set up for continuous in situ measurements at different heights
at the tower of the Lutjewad monitoring station. Hourly target measurements
were used to assess the accuracy and precision of the measurements. After
application of a calibration response curve for CO
The following data are available as a Supplement to this paper:
water vapor experiment data used in Fig. 3 time series of COS used for the Allan deviation plots in Fig. 7.
Other data from laboratory tests are available upon request
(huilin.chen@rug.nl).
The flask measurements made by NOAA/ESRL as part of a COS monitoring network
(Montzka et al., 2007) are available at
We would like to thank B. A. M. Kers, M. de Vries, H. G. Jansen and H. A. Been
for their help in preparing the system for installation in the field and for
maintenance of the instrumentation in Lutjewad. We thank D. Paul for the
valuable discussions and suggestions, H. A. Scheeren for providing the CO