You are on page 1of 9

Chemosphere: Global Change Science 1 (1999) 153±161

Atmospheric carbon monoxide Ôin situÕ monitoring by automatic


gas chromatography
V. Gros 1, B. Bonsang *, R. Sarda Esteve
LSCE, Orme des Merisiers, B^
at.709, CE Saclay, 91 191 Gif sur Yvette cedex, France

Received 15 April 1998; accepted 2 February 1999

Importance of this Paper. CO measurements in the background atmosphere of the southern hemisphere are mainly based
on weekly ¯asks' samples which allow the study of seasonal variations and long-term trends but do not give access to short-
term variations. The development of a fully automatic gas chromatograph described here, allows to perform 10-min step
measurements of CO which can be combined with monitoring of tracers or reactive compounds. This technique is used for
the study of background atmosphere such as at Amsterdam island in the remote Indian ocean or for measurements in the
more polluted atmosphere of the northern hemisphere.

Abstract

In order to develop Ôin situÕ automatic CO measurements, we adapted the method based upon the separation by gas
chromatography and the detection by hot mercuric oxide reduction, previously described in the literature. We applied
this method in order to alternate the injection at the atmospheric pressure of a standard and a sample, with each sample
calibrated against the preceding standard. A frequency of 5 min was chosen for each run, which implied a sample
measurement every 10 min. The evolution of the standards used in the routine experiments was regularly checked
against a primary standard provided by NOAA. First measurements, made at Amsterdam Island (37°S, 77°E) with this
procedure, showed quite a good agreement with other values observed in the remote Southern Hemisphere. Several tests
were then performed in order to determine some characteristics of the measurement. Thus, the standard deviation of the
mean CO mixing ratio, for 100 analyses of a single air sample containing approximately 200 ppbv CO, averaged ‹ 2%,
which determined the stability of the measurement. The reproducibility was found to be at most 2.5% and the overall
uncertainty of the measurement was estimated at 4%. Finally two intercomparison exercises between two similar in-
struments were performed at Amsterdam Island and they agreed within 4% and 5.5%, respectively, that is namely in the
range of the uncertainty of the analytical procedure. Ó 1999 Elsevier Science Ltd. All rights reserved.

Keywords: Carbon monoxide; Gas chromatography; Automatic Ôin situÕ measurement; Short-term variations; Remote
atmosphere

*
Corresponding author.
1
Present address: Max Planck Institute for Chemistry, Air Chemistry Division, Mainz, Germany. E-mail: vgros@mpch-
mainz.mpg.de (V. Gros)

1465-9972/99/$ ± see front matter Ó 1999 Elsevier Science Ltd. All rights reserved.
PII: S 1 4 6 5 - 9 9 7 2 ( 9 9 ) 0 0 0 1 0 - 0
154 V. Gros et al. / Chemosphere: Global Change Science 1 (1999) 153±161

1. Introduction tion. A 4-port valve, option of the RGA3, allowed


us to make a stream selection between the stan-
Carbon monoxide (CO) has been measured for dard and the sample and to select the frequency of
more than 50 years even if its key role in the the calibration. It has been decided to inject the
tropospheric chemistry, via its reaction with OH, sample alternately with a standard and to calibrate
has only been recognized since the 1970s. From each sample with the preceding standard. Two
solar spectroscopy to non-dispersive infrared gas kinds of standards, both in the range 100±200
®lter correlation, several methods have been used ppbv, were used in our study. Primary standards
to measure CO and they are reviewed in (WMO, provided by NOAA (CO-in-air, standardized
1994). The CO measurement consisting of sepa- against the CMDL/NOAA primary, gravimetric
ration by gas chromatography and detection by scale (Novelli et al., 1991) were used to calibrate
hot mercuric oxide reduction is assumed to have a secondary standards provided by Air-Liquide
very low limit detection (2 ppbv) and a good ac- (CO-in N2 ). In the automatic runs of the system,
curacy (1% at 100 ppbv) (WMO, 1994). This the secondary standard is continually circulating
method has been described by Seiler et al. (1980) with a low ¯ow rate (under 10 ml/min) and
and Novelli et al. (1991,1992) and can be used therefore is able to last over a year.
either for in situ measurement or for CO analyses For sampling, a Reciprotor pump located
based on ¯asks' sample collection. In situ mea- downstream the sample loop allowed the circula-
surements are necessary for continuous monitor- tion of ambient air in the system. For the standard,
ing particularly in remote locations which are too we mounted a 6-port valve at the rear of the in-
far away from the laboratory. Moreover, the sta- strument. The four possible con®gurations of the
bility of CO in ¯asks strongly depends on the system ± sample load, sample injection, standard
nature of the ¯asks and on the delay between in load and standard injection ± are depicted in Fig. 1
situ sampling and analysis in the laboratory (a±d).
(Novelli et al., 1992). In order to perform auto- In sample load position, sample ¯ows through
matic and in situ measurement, we adapted a GC the sample loop while the carrier gas is back-
instrument in the laboratory. However, in order to ¯ushed in column 1 in order to remove the com-
check and improve its performances in low pol- ponents from the former injection which will not
luted conditions the instrument was operated at be eluted. The correct timing of the 10-port valve
Amsterdam Island (37°S, 77°E) in the Indian switching enables CO2 , NMHC to be back¯ushed.
Ocean. The following step, sample injection, consists in
¯owing the carrier gas through the sample loop
which drags along the sample to column 1, column
2 and ®nally to the detector. During these two
2. Instrument description and analytical procedure steps the standard gas circulates at a low ¯ow rate
(less than 5 ml/min) controlled by a metering
The instrument used, a RGA3 (Trace Analyti- valve. The two other steps, standard load and in-
cal), allowed the separation and detection of CO jection, are based on the same principle, with the
and H2 with a typical retention time of 1.5 min. for standard circulating at atmospheric pressure in the
CO. The carrier gas used was Nitrogen 5.0 (purity sample loop at a ¯ow rate of 5±10 ml/min.
of 99.999%) at a ¯ow rate of about 20 ml/min. In order to avoid measurement of the pressure
Temperature of the back¯ush column (1/8 in ´ 30 in the loop and subsequent calculations, the loop
1/4 in unibeads 1S 60/80) and the analytical col- must be loaded at the same pressure for standard
umn (1/8 in ´ 30 1/4 in mole sieve 5A 60/80) was and sample. Loading the loop at ambient atmo-
95°C whereas temperature of the detector was spheric pressure appears to be the easiest way. In
265°C. the con®guration described above, the sample
The RGA3 analyzer was adapted for in situ pressure in the loop is lower than the ambient at-
continuous monitoring of CO including calibra- mospheric pressure, resulting from a pressure drop
V. Gros et al. / Chemosphere: Global Change Science 1 (1999) 153±161 155

Fig. 1. The 4 possible con®gurations of the system: a) sample load, b) sample inject, c) standard load, d) standard inject.

in the sampling and injection lines. In order to pressure drop when the electrovalve was not used,
avoid this problem, an electrovalve, controlled by i.e. from March to December 1996 for Amsterdam
the software Borwin, was mounted between the 6- Island measurements. They consisted of the injec-
port valve and the pump. This electrovalve isolates tion of the sample with and without the electro-
the pump a few seconds before the sample injec- valve operating. The pressure drop was found to
tion and allows the pressure in the loop to equili- be randomly distributed over the range 15±25% of
brate with the external pressure. In this new the ambient pressure which will result in a cor-
con®guration, we checked that the loop was rective factor of 20 ‹ 5 %. In order to adjust more
loaded at the same pressure for both sample and precisely this corrective factor, we have made
standard within the accuracy of the pressure sen- comparisons with the data obtained at Crozet Is-
sor (i.e. 0.1 Torr). Several tests were conducted lands from ¯asks samples (Gros et al., this issue).
over a period of 3 months in order to estimate the For these two sets of data the best ®t was obtained
156 V. Gros et al. / Chemosphere: Global Change Science 1 (1999) 153±161

for a corrective factor of 25% i.e. in the upper


range of our estimate. This factor (+ 25 %) was
applied to Amsterdam Island measurements from
March 1996 to December 1996. Besides the un-
certainty on the measurement calculated in section
3.5, the data obtained at Amsterdam Island for
1996 present an additional uncertainty due to this
problem of pressure drop and are estimated at
‹ 5%.
Data acquisition and instrument control were
made by `Borwin' gas chromatography software.
This software was used in order to follow the
chromatograms in real time and compute directly,
from the peak area of each sample and its pre-
ceding standard, the concentration of CO. This
software was also used to start the analytical
procedure of the RGA3 instrument, allowing to
choose the frequency of the measurement. Each
injection which alternates a standard and a sample
was made every 5 min. Consequently the fre-
quency of the measurement as well as of the
standard was ten min. An example of chromato-
gram for a standard and for a sample is given in
Fig. 2.

3. Characteristics of the measurement

3.1. Linearity

The linearity of the measurements was ®rst


studied over a wide range with injections of a
standard of 9.5 ppmv provided by the National Fig. 2. An example of CO chromatogram for standard and
Institute of Standards and Technology (NIST) at sample, respectively.
various pressures in the sampling loop. The re-
sponse of the detector was checked up to a pres-
sure of 760 Torr. As described in Fig. 3a a good the linearity could change with time (Novelli et al.
linearity can be observed (with a correlation co- 1992), it is not expected to a€ect the range of in-
ecient of 0.99985) up to a pressure of 80 Torr terest, i.e 0±200 ppbv, very close to the origin of
corresponding to 1 ppmv of CO at ambient at- the response of the instrument.
mospheric pressure. Afterwards, we studied more
speci®cally the linearity in the range of expected 3.2. Stability
concentrations in background air (i.e 50±500
ppbv). These tests were performed with another The stability of the measurement was checked
source of standard (standard with 365 ppbv/air by continuously ¯owing a secondary standard (Air
provided by NOAA). As shown in Fig. 3b, a good Liquide) during a time period of 5 h and recording
linearity was observed for the range 0±600 ppbv the peak area every 3 min (Fig. 4). The stability of
with a correlation coecient of 0.99976. Even if the measurement over this period is statistically
V. Gros et al. / Chemosphere: Global Change Science 1 (1999) 153±161 157

determined as the standard deviation of this


peak area and is about 2%. The variability of the
measurement includes several factors as variability
of the carrier gas ¯ow rate, of the ambient pres-
sure and of the eciency of the HgO to Hg con-
version. As the ambient pressure variability was
not signi®cant during the period of study (<0.2%),
the 2% variability observed is representative of the
stability of the measurement itself. Many other
similar tests were performed with very similar re-
sults.

3.3. Reproducibility

The reproducibility of the measurement was


checked by calibrating standards in the range of
50±200 ppbv against the primary standard
(NOAA). The reproducibility over about 100
measurements is statistically determined as the
relative standard deviation of the calculated stan-
dard concentration, and ranges from 1% to 2.5%
with generally highest values corresponding to the
lowest concentrations. Other calibrations regularly
performed showed a similar or better reproduc-
ibility and this value of 2.5% can therefore be
considered as the maximum analytical uncertainty
of the measurement.
Fig. 3. (a) Detector responses to injections of a standard of 9.5
ppmv at various pressures in the sampling loop. A good lin-
earity is observed for the concentration range 0±1 ppmv. (b) 3.4. Calibration
Detector responses to injections of a standard of 365 ppbv at
various pressures in the sampling loop. A good linearity is Evolution of the secondary standard is con-
observed for the concentration range 0±600 ppbv. trolled about every two months against the pri-
mary standard and its value is readjusted if

Fig. 4. Recording of the peak area of the Air Liquide standard Fig. 5. Calibrations of the secondary standard (Air Liquide)
during a time period of 5 h. against the primary standard (NOAA) at Amsterdam Island.
158 V. Gros et al. / Chemosphere: Global Change Science 1 (1999) 153±161

 2  2  2
necessary. For example, variations of the standard DCAL DCNOAA DRAL;NOAA
ˆ ‡
Air Liquide used at Amsterdam Island since May CAL CNOAA RAL;NOAA
1997 is given in Fig. 5. Primary standard was
cylinder CA01995, CO-in-air with 151.5 ‹ 1.5 ppbv and ®nally
CO and was assumed to be stable with respect to  2  2  2
CO for periods of at least 1.5 year (Novelli et al., DCsample DCNOAA DRsample;AL
ˆ ‡
1991). The secondary standard does not seem to Csample CNOAA Rsample;AL
present any signi®cant drift but its value ¯uctuates  2
DRAL;NOAA
around 193.1 ‹ 4.5 ppbv which represents a 2.5% ‡ :
RAL;NOAA
variability. Therefore, the secondary standard
variability observed over one year is within the An uncertainty of 1% on the concentration of
analytical uncertainty of the measurement. Note the primary standard NOAA is provided by the
that the evolution observed can vary from one manufacturer (Novelli et al., 1991). The uncer-
secondary standard to another. tainty on the ratios Rsample ,AL and RAL;NOAA were
experimentally evaluated at 2.5%, as described in
section 3.3. Finally CO measurements are made
3.5. Overall uncertainty of the measurement with an overall uncertainty of about 4%.

As the CO concentration of a sample is deter-


mined as 3.6. Intercomparison of two analyzers
Asample
Csample ˆ CAL  ˆ CAL  Rsample;AL ;
AAL We performed at Amsterdam Island two series
where; Rsample;AL ˆ Asample =AAL of intercomparison between two identical RGA3
analyzers. Each analyzer was working indepen-
with dently with its own sample line and secondary
AAL standard. Only the primary standard was com-
CAL ˆ CNOAA  mon for both analyzers. The two intercompari-
ANOAA
sons were made in March 1997 and in November
ˆ CNOAA  RAL;NOAA ; 1997 with respective CO level of about 40 ppbv
and 60 ppbv. Fig. 6 a and b present the distri-
where butions of the di€erence between the analyzers.
RAL;NOAA ˆ AAL=ANOAA ; On an average, this di€erence (analyzer 1ÿ ana-
lyzer 2) was 1.59 ‹ 2.09 ppbv for the ®rst inter-
where Csample , CAL and CNOAA stand for the CO comparison and ÿ3.3 ‹ 2.0 ppbv for the second
concentration of the sample, the secondary stan- intercomparison. The secondary standard associ-
dard (Air Liquide) and the primary standard ated with the ®rst analyzer was changed between
(NOAA), respectively, Asample , AAL and ANOAA the two intercomparisons and this could explain
represent the CO peak area of the sample, the Air why analyzer 1 presented higher values in the ®rst
Liquide and NOAA standards, respectively. exercise and lower values in the second. Finally,
As the terms are independent and the corre- results between both analyzers agreed within 4%
sponding uncertainty is rather low (a few percent), for the ®rst run and 5.5 % for the second run.
the uncertainty on the measurement can be cal- The uncertainty on the primary standard being
culated as the same for both analyzers, it is not taken into
 2  2  2 account in the di€erence observed. Therefore, the
DCsample DCAL DRsample;AL uncertainty on each chain of measurement is less
ˆ ‡ ;
Csample CAL Rsample;AL than 4% and the di€erence observed between the
analyzers remains within the global uncertainty
with of 8%.
V. Gros et al. / Chemosphere: Global Change Science 1 (1999) 153±161 159

were then aggregated into hourly and monthly


averages.
A ®rst set of measurements was performed be-
tween March and December 1996 at Amsterdam
Island. Since January 1997, an electrovalve, which
solved the problem of pressure drop, has been in-
stalled. A 8-day time series during September 1997
is given in Fig. 7a. CO background levels are about
55 ppbv, which is in good agreement with other
data published for this area. For example CO
monthly mean was 55 ppbv at Cape-Grim (Tas-
mania, 40.6°S, 144.68°E) in September 1995
(Steele et al., 1996). The increase of CO which can
be observed on 8th September results from trans-
port of continental air from Southern Africa (de-
tails in Gros et al., this issue).
This CO measurement system has been also
used for in-situ measurement in the Northern
Hemisphere during the two phases of the EC

Fig. 6. Intercomparison between CO analyzers at Amsterdam


Island (a) 10±19 March 1997 and (b) 10±14 November 1997.

4. Time series

Continuous measurements every 10 min implies


the management of a large amount of data and it
was necessary to apply a ®lter in order to suppress
the results out of range. A visual inspection of the
chromatograms have shown that every outlier data
was due to a wrong peak integration which was
often linked to electrical spikes and artefacts. At
Amsterdam Island, in 1996, the rejected data for
each month were less than 10% of the total, except
for June 1996 where about 12 % of the data were
removed. In 1997, this ®gure decreased to less than
1% of the total (except for April 1997 where it
increased to 2.5%), this signi®cant improvement Fig. 7. CO measurements at Amsterdam Island (a) and in
was probably due to the installation of the elec- Greece in comparison with black carbon (BC) measurements
trovalve previously described. The remaining data (personal communication) during the AEROBIC campaign (b).
160 V. Gros et al. / Chemosphere: Global Change Science 1 (1999) 153±161

campaign AEROBIC which took place in Portugal Territoire des Terres Australes et Antarctiques
(August 1996) and Greece (July±August 1997). A Francßaises (TAAF) and the Institut Francßais de
time series is given for the second campaign in Fig. Recherche pour lÕExploitation de la Mer (IF-
7b in association with the measurement of black REMER) for ®nancial and logistical support of
carbon (T. Nunes and C. Alves, personal com- the monitoring station at Amsterdam Island
munication). The very good agreement between (grant RACEA 146). This work was also sup-
the variations of the two compounds which have ported by Centre National de la Recherche
the same anthropogenic sources ascertains the Scienti®que (CNRS) and Commisariat a lÕEnergie
quality of our CO measurement. Atomique (CEA). Data for black carbon are from
courtesy of T. Nunes and C. Alves. The authors
are indebted to Pierre Nadeau and Beno^õt Pelczar
5. Conclusion for technical assistance in measuring CO at Am-
sterdam Island. This is a LSCE contribution
We adapted the CO measurement based on no.227.
separation by gas chromatography and detection
by hot mercuric oxide reduction in order to per-
form continuous monitoring , with a step time of
References
10 min and to reach a very low detection limit
adapted to the ambient measurements in pure Novelli, P.C., Elkins, J.W., Steele, L.P., 1991. The development
marine atmosphere. One of the ®rst applications of and evaluation of a gravimetric reference scale for mea-
this technique was the continuous monitoring of surements of atmospheric carbon monoxide. J.Geophys.Res
CO in a remote location (Amsterdam Island: 37°S, 96, 13109±13121.
Novelli, P.C., Steele, L.P., Tans, P.P., 1992. J.Geophys.Res 97,
77°E) where logistical problems occur. CO data
20731±20750.
are now available in this location since March Seiler, W., Giehl, H., Roggendorf, P., 1980. Detection of
1996. Measurements are made automatically every carbon monoxide and hydrogen by conversion of mercuric
10 min and are based on the alternate measure- oxide to mercury vapor. Atmos. Tech. 12, 40±45.
ment of the standard and of the sample, with a Steele, L.P., Langenfelds, R.L., Lucarelli, M.P., Fraser, P.J.,
Cooper, L.N., Spencer, D.A., Chea, S., Broadhurst, K.,
simple injection procedure at ambient atmospheric
1996. Atmospheric methane, carbon dioxide, carbon mon-
pressure. This technique requires very little man- oxide, hydrogen, and nitrous oxide from Cape Grim ¯asks
power, and needs only to replace the carrier gas air samples analysed by gas chromatography, in Baseline
every two months, and the secondary standard Atmospheric Program Australia 1994±1995, R.J. Francey,
every year. Comparison of our measurements with A.L. Dick and N. Derek (Eds.), Bureau of Meteorology and
CSIRO Division of Atmospheric Research, Melbourne,
other observations in a remote area of the South-
Australia, pp 107±111.
ern Hemisphere shows a quite good agreement for World Meteorological Organization (WMO), report of the
the monthly average. The high frequency of our WMO meeting of experts in global carbon monoxide
measurements allowed us to observe CO episodes measurements, P.C. Novelli and R.M. Rosson (Eds.), No
on a short time scale (see Gros et al., this issue). 98, Boulder, 8±11, 1994.
This system was used during two ®eld cam-
paigns in Europe, where the CO level is of the Valerie Gros obtained her PhD from the University of Paris 7
in December 1998. She worked on the study of ozone (O3 )
order of 200 ppbv, and the results presented very and carbon monoxide (CO) variabilities in the marine boun-
good correlations with other anthropogenic trac- dary layer of the southern hemisphere. She particularly stud-
ied the marine atmosphere of Amsterdam Island, Indian
ers such as black carbon. ocean, with pointing out the key processes involved in the O3
and CO budgets over di€erent time scales. She has several
publications in international referred journals and conference
proceedings.
Acknowledgements
Bernard Bonsang has worked in the ®eld of atmospheric sci-
ences since 1972. He is Directeur de Recherche (CNRS) since
We are grateful to the Institut Francßais pour la 1993. He studied the oceanic source of reduced sulphur gases
Recherche et la Technologie Polaires (IFRTP), the and hydrocarbons. One part of his activities concerned also the
V. Gros et al. / Chemosphere: Global Change Science 1 (1999) 153±161 161

evaluation of the production of carbon monoxide, methane and on long range transport of ozone and its precursors over the
NMHC's by biomass burning in the tropics, with particulate oceans. He has been coordinator of EC projects dealing with
emphasis on the tropospheric ozone budget. His main activities oxidation processes of VOC's. He is author of more than 60
are focused on the oxidizing capacity of the troposphere in the publications in international referred journals and conference
frame of national and European projects and on a global scale proceedings.

You might also like