678 Journal of the Meteorological Society of Japan Vol. 61, No.

High Quality Measurements of the Concentration

of Atmospheric Carbon Dioxide

By MasayukiTanaka,TakakiyoNakazawaand ShuhjiAoki
UpperAtmosphere
ResearchLaboratory,
TohokuUniversity,Sendai980,Japan
(Manuscript
receivedMarch3 1983,in revisedformJune5, 1983)

Abstract
The technique for high quality measurements of the atmospheric CO2 concentration has
been developed by improving many facets of the commercial nondispersive infrared CO2
analyzers. The precision of better than *0.05ppm was attained for both continuous and
flask-sample analyses. The standard gases of CO2-in-air mixture have been prepared by the
gravimetric method using an extremely precise balance; the absolute accuracy of the CO2
concentrations were estimated to be better than *0.3ppm. The problems of great impor-
tance in the flask sampling technique, i.e., possible deterioration risk of air samples stored
in the flasks and effect of water vapor condensed in the sample flask on measured values
of the CO2 concentration, have been examined.

standard gases differs from that in the air sample

1. Introduction
to be measured (Bischof, 1975; Pearman and
Recently, the requirement for high quality Garratt, 1975; Keeling et al., 1976a, b; Pearman,
measurements of the concentration of atmos- 1977; Griffith, 1982; Griffith et al., 1982). The
pheric CO2 has increased noticeably as a result carrier gas effect arises from the difference in
of wide recognition of a potential CO2-climate collision-broadened widths of the CO2 absorp-
problem (WMO, 1981). For the atmospheric tion lines among different collision partners, and
CO2 monitoring, proper installation and opera- depends on the design of the analyzer, e.g., the
tion of the analysis device as well as discreet absorption band used, the dimension of the
air sampling are essential. In general, the at- sample (or reference) cell and the amount and
mospheric CO2 concentration is measured by pressure of CO2 filled in the detector cell. The
using a non-dispersive infrared (NDIR) gas standard gases used at all monitoring stations,
analyzer. However, most of the usual commer- except for the Stockholm University (Bischof,
cial gas analyzers are originally designed for 1977) which uses CO2-in-air mixtures, are CO2-
engineering use such as measurements in the in-N2 mixtures. Although their concentrations
exhaust gas or in the greenhouse and not always are calibrated against the absolute system of the
suitable for baseline work unless specially im- Scripps Institution of Oceanography (Keeling,
proved. 1958), CO2 data obtained at different stations
Since the NDIR gas analyzer is for relative require different corrections for the carrier gas
measurement, it requires standard gases defined effect properly determined for respective ana-
against absolute calibration system to relate the lyzers. Such a correction is not an easy task
analyzer response to the absolute value of the and undesirable.
CO2 concentration. The absolute concentration We have initiated an atmospheric CO2 moni-
of usual standard gases prepared by gas manu- toring program in December 1978 using the
facturers is guaranteed only for an accuracy of NDIR analyzers. The program comprises of in
a few ppm which is insufficient for atmospheric situ continuous measurements and continual col-
CO2 monitoring. lections of air samples with subsequent analysis
The NDIR gas analyzers indicate erroneous in the laboratory. Preliminary results of our
results when the carrier gas composition in the aircraft measurements have been presented in
August 1983 M. Tanaka, T. Nakazawa and S. Aoki 679

Tanaka et al. (1983a), and those of continuous the reference cell of the VIA analyzer from
measurements at a ground station will be given enclosed N2 gas to a constantly flowing reference
in Tanaka et al. (1983b). In this paper, tech- gas to set optical bias for accurate measurements
nical aspects of our monitoring program are in an arbitrary concentration range of 50ppm
described in detail. span; (5) rearrangement of the electric system
of the EIA analyzer to convert the original range
2. Non-dispersive infrared (NDIR) CO2
of 0-500ppm to an arbitrary range of about
analyzers
150ppm by using a differential amplifier; and
The non-dispersive infrared CO2 analyzers of (6) insertion of interference filters (*max=4.3*m,
the types EIA and VIA, manufactured by 1/2=0.3*m), which pass light only* in a spec-
Hitachi-Horiba Co., Japan, have been adopted tral region of the 4.3*m CO2 band, into both
for continuous measurements and analyses of sample and reference beams to reduce spectral
flask samples, respectively. The optical system contaminations due to atmospheric water vapor
of these analyzers is composed of double light absorption, especially by the 2.7 and 6.3*m
sources, an optical balancer, parallel reference bands.
and sample cells, an optical chopper and a single Since the spaces for the optical chopper and
detector with parallel absorption cells (Fig. 1). the optical balancer are open to the analyzer
The following facets of these analyzers have housing, the fluctuation of CO2 content of the
been improved to satisfy the precision required air enclosed in the housing becomes an addi-
for the atmospheric CO2 monitoring: (1) Sta- tional source of the instrumental drift. To avoid
bilization of the electric voltage supplied to the such a defect, the analyzers were purged of CO2
light sources to minimize the fluctuation of the by continuously flowing the CO2-free N2 gas
intensity of the light sources; (2) stabilization of with a pressure slightly higher than the environ-
the voltage and frequency of the electric sources ment.
supplied to the chopper motor to minimize the
irregularity in the rotation rate of the optical 3. Standard gases
chopper; (3) improvement of the temperature To maintain the consistency of measurements
control of the analyzer with a precise propor- for a long time, the standard gases are classified
tional regulation; (4) alternation of the gas in into three categories i.e., the primary, secondary
and working standards. All standard gases are
CO2-in-air mixtures and stored in 47 l high pres-
sure gas cylinders of chrome-molybdenum-steel.
Therefore, the correction for the carrier gas
effect is not necessary. The working standard
gases are used for the calibration of the analyzer
in usual measurements, and their concentrations
are compared with the secondary standard gases
before and after their uses. The secondary
standard gases are also compared carefully with
the primary standard gases for about once a
year. The primary standard gases were calibrated
by gases prepared by the gravimetric method
with an extremely precise balance of accuracy
better than 3*=*4.6mg (* being the rms error)
for a wide range from 1mg to 100kg. For this
purpose, the original gas of about 1 % concen-
tration was first prepared by mixing the air
which was free from CO2 and H2O (<0.1ppm
for CO2 and dewpoint temperature of -70*
for H2O) with accurately known amount of CO2
and, then, the 9 calibration gases ranging from
310 to 450ppm were obtained by dilution. An
Fig. 1 Non-dispersive infrared gas analyzer. example of the preparation of the calibration
680 Journal of the Meteorological Society of Japan Vol. 61, No. 4

gases is given below. The 10l chrome-moly- Table 1 Comparison of the original and cor-
bdenum-steel cylinders were used for this pre- rected CO2 concentrations of the pri-
mary standard gases (units: ppm).
paration. The mass of the CO2 gas with a purity
of 99.995% was measured by the precise bal-
ance to be 30.073g after correcting for the
fluctuation of the zero-point of the balance due
to the fluctuation of the air conditioning and
for the buoyancy for which the correction was
minimized by roughly balancing the volumes of
both sides. The carrier gas (i.e., dry air without
CO2) was then added to the CO2 gas, and the
mass of the total mixture was measured to be
1417.833g; the mass of the carrier gas being
1387.760g, by the same procedure as the CO2
gas. Assuming normal abundances of air mole- tion gases were corrected by a least square fit
cules and their isotopes (Miller, 1961), the technique assuming a quadratic relation between
molecular weights of the carrier gas and CO2 the response of each analyzer and the CO2 con-
were estimated to be 28.960 and 44.011, respec- centration. An assumption of quadratic relation-
tively, and accordingly the CO2 concentration of ship was found to be adequate for an arbitrary
the original gas prepared above was determined range of about 50ppm. In Table 1 are shown
to be 1.406% by volume. The mass of 33.977g the results of comparison of the original calibra-
of the original gas thus obtained was diluted by tion values and those corrected by the above
adding the carrier gas of 1285.965g to obtain mentioned procedure for both VIA and EIA
the calibration gas of the CO2 concentration of analyzers. Most of the corrections were less
359.3ppm by volume. than *0.2ppm. Although the concentration drift
Although the accuracy of the balance we used of the standard gases is a critical problem for
is very good as mentioned before, an overall the interpretation of the secular trend of CO2
accuracy of the gravimetry is degraded to be increase in the atmosphere, our primary stand-
better than *25mg due to adsorption of water ard gases have been confirmed to remain stable
vapor and especially adhesion of dust particles over 2 years by the comparison with the calibra-
on the outer wall of the cylinder. Thus, the tion gases prepared by the same procedure in
accuracy of the calibration gas is given by February 1981. However, further work will be
necessary for the more precise determination,
which is required for the examination of the
potential long-term drift in the concentration.
The cross comparison of our primary standards
for the preparation of the original gas, and
with the Scripps standards is also desired for
comparability of the data obtained.
The working and secondary standard gases
were used after aging over more than a month
to minimize the concentration drift. The usage
for the preparation of the diluted gas,and
of all gases is terminated at a pressure of about
30kg/ cm2 because the drift in concentration is
for an overall accuracy. The estimation of the likely to occur with further decrease of the pres-
accuracies for the other calibration gases is very sure in the cylinders.
similar to this example; the absolute accuracy
4. Flask sampling procedure
of the calibration gases is of the order of 0.1%
or, equivalently, about 0.3ppm. Uncertainties Air sample containers are cylindrical Pyrex
due to the residual CO2 in the carrier gas and glass flasks of 550ml capacity with greaseless
the molecular abundances are expected to be stopcocks fitted at the both ends. Cleaning of
smaller than the above limit. the flasks was first performed at the time of
In order to minimize the uncertainties in cali- their fabrication by washing in an aqueous de-
brated values, CO2 concentrations of the calibra- tergent solution activated by ultrasonic waves
August 1983 M. Tanaka, T. Nakazawa and S. Aoki 681

and by heating to about 460* for annealing. for the operation at the ground station for con-
Subsequent cleaning is usually done by the same tinuous measurements. The analyzer has the 12
procedure as above at about half year intervals inlets with solenoid valves; 8 channels are usual-
during the long-term use of flasks. However, the ly alloted to the sample air and remaining 4
flasks used for the air sampling on ships are channels to the standard gases. Tubing system
cleaned whenever they are used. Before sam- was made of a number of different materials
pling, the flasks are evacuated to a pressure be- including stainless steel, brass, glass and flexible
low 1 * 10-3mmHg and heated at approximately plastic for which permeability and/or exchange
100* for more than 5 hours. among different gases are of no problems. Es-
The air samples are pressurized up to about pecially, flexible plastic was used only for the
3 atm with an electric or manual diaphragm tubing of the air intake in order to avoid uptake
pump. The intake for the air sample is mounted and release of CO2 depending on the pressure.
so as to avoid unnecessary contaminations such The inner wall of the tubing system was washed
as expiration of the observers. In the air sam- by chemicals and dried out sufficiently before
pling on aircraft and ship, a certain care is also its arrangement. To achieve rapid gas exchange
taken to avoid the contamination from the en- in the system and to save the usage of standard
gine exhaust and cabin air. The sampling time gases, the tubing was made as short as possible.
is taken to be long enough to ensure a complete Additional attention was also paid to confirm
air exchange in the sample flasks. Our sample no leakage in the tubing and connection system.
flasks, not evacuated, require the air flow which Water vapor contained in the sample air is
corresponds to about four times their volume removed in two steps, by specially designed glass
for the complete air exchange. The flask filled traps cooled by electric freezers. The rear trap
with the air sample is protected from breakage is always operated at temperatures lower than
by a shock absorbing material and stored in a - 30* , while the temperature of the fore trap
dark inside of a thick paper box. is arbitrarily set according to the ambient air
temperature. The concentration of CO2 is there-
5. Measurement system and procedures
fore defined as part per million (ppm) by volume
5.1 Continuous measurement dry air. Aerosol particles are removed from the
Figure 2 shows the infrared analyzer system sample air by thread-glass and membrane filters

Fig. 2 Non-dispersive infrared gas analyzer system for continuous

measurements of atmospheric CO2.
682 Journal of the Meteorological Society of Japan Vol. 61, No. 4

inserted into the tubing system.

Since the infrared absorption is greatly af-
fected by the gas pressure and temperature, the
sample air and standard gases should be pre-
pared in the sample cell of the analyzer at the
strictly equal pressure and temperature. It is
also necessary to maintain a constant flow rate
because heat exchange between the inner wall
of the cell and flowing gas depends on the flow
rate. In order to attain the above requirements,
the sample air is pressurized up to 2 atm by the
electric diaphragm pump, while the standard
gases are depressurized down to the salve pres-
sure as the sample air, i.e., 2 atm, by a pressure
regulator mounted to the outlet of the cylinder.
Then, pressures of both gases are adjusted to
the assigned pressure (1 atm) and flow rate
(500ml/ min) by a constant pressure controller
and a constant flow controller, respectively. Be-
Fig. 3 Non-dispersive infrared gas analyzer
cause of a finite conductance of the tubing-cell
system for flask sample analysis.
system, only a slight difference of the pressure
between the sample air and the standard gas
affects appreciably on the pressure in the sample of 10ml/min. The tubing system is fabricated
cell if the above mentioned controllers are not from stainless steel, except for the Hg reservoirs
used. The buffering volume attached behind the and water vapor traps which are made of Pyrex
pressurizing pump absorbs efficiently the pressure glass.
pulsation due to vibration of the diaphragm of The sample flask is connected with the gas
the pump. Since the sample air and standard handling system of the device by an air-tight
gases are pressurized in the common tubing sys- coupler, and the Hg reservoirs as well as the
tem, residual gas of the preceding channel is connection tubings between the sample flask and
released to the ambient air by opening the sole- the inlet of the analyzer are evacuated to a
noid valve (No. 14) in a moment, to attain rapid
pressure of less than 1 * 10-7mmHg by a rotary
gas exchange. pump. After evacuating sufficiently, the stop-
An apparatus for flask sampling is added to cock of the flask is opened and the sample air
this system. The sample air collected by this is brought into the Hg reservoir through the
apparatus is analyzed in the laboratory and com- warer vapor trap cooled at -78*. The sample
pared with the results of in situ measurements air stored in the Hg reservoir is, then, flown
just before and after the sample collection to into the analyzer at a flow rate of about 300ml/
confirm an overall validity of the system. min with the introduction of the ambient air
The operation of all solenoid valves is con- into another reservoir. This procedure is re-
trolled by a micro computer. The signal from
peated, if necessary, for the analysis of the low-
the analyzer and other meteorological parame-
pressure samples. Standard gases are directly
ters, i.e., temperature, solar radiation, wind introduced into the analyzer at constant pressure
direction and speed, and humidity, are recorded and flow rate.
with a strip-chart recorder and a paper-tape
The response of the analyzer is calibrated at
puncher. The timing of registration is also con- about 1 hour intervals during the analysis by
trolled by the micro computer. using three or four standard gases. The standard
5.2 Flask sample analysis gases are usually chosen so that their CO2 con-
A so called "static air analysis" is employed centrations distribute in a range of about 30ppm
for the analysis of flask samples. The arrange- including the concentration of the sample air.
ment of the device is shown in Fig. 3. The VIA No additional uncertainty is introduced into the
analyzer is used and the reference gas is flown final result by this calibration procedure, since
through the reference cell at an uniform rate the instrumental drift of the analyzer has been
August 1983 M. Tanaka, T. Nakazawa and S. Aoki 683

reduced to less than *0.2ppm/ day.

All valves are manually operated, and the
response of the analyzer is read by a digital
voltmeter with monitoring by the strip-chart
recorder. This device is also used to compare
the concentration of a standard gases.

6. Accuracy of measurements

The precision of our analyzers was examined

by determining the concentration of the working
standard gases against the secondary standard
gas system, by directly flowing respective gases
into the analyzers. The number of analysis made
for individual working standard gas ranged from
3 to 13, but mostly between 4 and 6. The results
are given in Figs. 4 and 5 for the VIA and EIA
analyzers, respectively. The result for the VIA
analyzer indicates that about 80% of all stand-
ard deviations are found within *0.01ppm
range. However, the precision of the EIA ana-
lyzer is worse by approximately one order of
magnitude as compared with that of the VIA Fig. 5 The same as in Fig. 4 but by the EIA
analyzer, although it is still the same order of analyzer. The total number of gas sam-
magnitude as that of most analyzers used in the ples is 50.
BAPMoN (Baseline Air Pollution Monitoring
Network) stations (Pearman, 1977). This differ-
fact that the sample cell of the EIA analyzer
ence in their precisions is probably due to the
is shorter by 30cm in length than that of the
VIA analyzer. In the above examination, the
center value of the analyzer response was care-
fully read by the digital voltmeter with moni-
toring by the strip-chart recorder. Taking into
account the fact that the noise in the response
of the EIA analyzer is about 0.1ppm, the un-
certainty in our continuous measurements may
be somewhat larger than that of the analyzer
itself; in the continuous measurements, the an-
alyzer response is automatically recorded without
any sophisticated statistics according to the com-
mand from the micro computer.
In order to assess the accuracy of the flask
sample analysis, examinations were also made
for dry standard gas sample and for those
saturated with water vapor. Dry standard gas
samples filled in the flasks were analyzed by the
procedure described in the previous section and
by introducing gas samples directly from the
flask into the analyzer without using the mercury
pump system. The result was almost the same
Fig. 4 Distribution of standard deviation of as that for analyze itself.
the CO2 concentration of working standard Since the standard gases are very dry, the
gases determined by the VIA analyzer above examination is not adequate to judge the
against the secondary standard gas system. uncertainty in the analysis of samples taken from
The total number of gas samples is 54. the real atmosphere containing water vapor.
684 Journal of the Meteorological Society of Japan Vol. 61, No. 4

Fig. 7 Deterioration of standard gas samples

filled in the sample flasks over a long time.
Solid circles represent the results for the
sample pressure of 3 atm and open circles
for the sample pressure of 2.5 atm. The
Fig. 6 Influence of liquid water in the sample CO2 concentrations are shown as a devia-
flask on the measured CO2 concentration tion from that of the standard gas sample,
(solid circles). The results were obtained i.e., 330.74ppm.
by using a 1.7l capacity flask filled with
3.1 atm standard gas with CO2 concentra-
tion of 330.74ppm and 0.3ml distilled examination, the flask samples collected in the
water. Open circles show the results for middle and upper tropospheres and the lower
dry standard gas sample. All values are stratosphere where water vapor concentration is
shown as a deviation from the calibrated low and no condensation occurs, are analyzed
value of 330.74ppm. without using the mercury pump system to save
analyzing time.
Indeed, the condensation of water vapor on the Air samples are usually analyzed within a few
inner wall of the flask was observed frequently days after their collections. However, in some
when the stopcock was opened to introduce the projects, the lapse in time between sampling and
air into the flask and/or the air was compressed subsequent laboratory analysis sometimes reached
into the flask, especially in the lower troposphere several weeks or more. A deterioration risk of
and in the summer season. In such cases, the sampled air was therefore examined by
measured values of the CO2 concentration in- analyzing the standard gas samples stored in our
crease gradually with repeating the analysis as sample flasks. The measured CO2 concentrations
seen in Fig. 6, if the sample air is directly in- are plotted against the storing time in Fig. 7.
troduced into the analyzer without using the The CO2 concentration decreases rapidly by
mercury pump system. It was found that this about 0.15ppm within 1 day, and about 30 days
phenomenon depends on the amount of water after it reaches a value of about - 0.5ppm
added, its distribution on the inner wall of the rather gradually. This effect is probably due to
flask, the pressure of the sample air, and the a selective adsorption of CO2 on the inner wall
flow rate of the sample air introduced from the of the flask; a long period of time of a month
flask into the analyzer. The phenomenon is is required to establish an equilibrium of CO2
probably attributable to the fact that a new between the gas and the inner wall of the flask.
equilibrium of CO2 between the air and liquid Different flasks indicate different deterioration of
water is established instantly according to the the CO2 concentration within a range of *0.05
pressure in the sample flask. The influence of ppm. The pressure difference of the gas samples
condensed water on the measured CO2 concen- between 2.5 and 3 atm appeared to be insignifi-
tration was reduced to be negligible by using the cant for the deterioration risk.
mercury pump system, which is able to store
most of the sample air in the flask before analy- References
sis. However, the uncertainty in the analysis
Bischof, W., 1975: The influence of the carrier gas
may be increased to about *0.05ppm, which on the infrared gas analysis of atmospheric CO2.
is still sufficient for the atmospheric CO2 moni- Tellus, 27, 59-61.
toring. ,, ,1977: Comparability of CO2 measure-
Taking into account the results of the above ments. Tellus, 29, 435-444.
,

August 1983 M. Tanaka, T. Nakazawa and S. Aoki 685

Griffith, D. W. T., 1982: Calculation of carrier gas book of Geophysics (revised edition), The Mac-
effects in non-dispersive infrared analyzers. I. millan Company, New York.
Theory. Tellus, 34, 376,-384. Pearman, G. I. and J. R. Garratt, 1975: Errors in
atmospheric CO2 concentration measurements
Guenther and R. B. Bacastow, 1982: Calculation arising from the use of reference gas mixtures
of carrier gas effects in non-dispersive infrared different in composition to the sample air. Tel-
analyzers. II. Comparisons with experiment. lus, 27, 62-66.
Tellus, 34, 385-397. Pearman, G. I., 1977: Further studies of the com-
Keeling, C. D., 1958: The concentration and iso- parability of baseline atmospheric carbon dioxide
topic abundances of atmospheric carbon dioxide measurements. Tellus, 29, 171-181.
in rural areas. Geochim. et Cosmochim. Acta, Tanaka, M., T. Nakazawa and S. Aoki, 1983a: Con-
centration of atmospheric carbon dioxide over
, ,R. B. Bacastow, A. E. Bainbridge, J r., Japan. J. Geophys. Res., 88, 1339-1344.
C. A. Ekdahl, P. R. Guenther, L. S. Waterman , and , 1983b:
and J. F. S. Chin, 1976a: Atmospheric carbon Atmospheric carbon dioxide variations in hilly
dioxide variations at Mauna Loa Observatory, district in the suburbs of Sendai, Japan, submit-
Hawaii. Tellus. 28. 538-551. ted to Tellus.
, J. A. Adams, Jr., C. A. Ekdahl and WMO, 1981: .Report of the WMO/UNEP/ICSU
P. R. Guenther, 1976b: Atmospheric carbon Meeting on Instrumennts, Standardization and
dioxide variations at the South Pole. Tellus, 28, Measurement Techniques for Atmospheric CO2.
552-564. Geneva, 8-11 September, 1981.
Miller, L. E., 1961: Atmospheric composition, Hand-

大 気 中 の 二 酸 化 炭 素 濃度 の 高 精 度 測 定

田 中 正 之 ・中 澤 高 清 ・青 木 周 司
東北大学理学部超高層物理学研究施設

赤 外 線 吸収 法 を 用 い て,大 気 中 の二 酸 化 炭 素 濃 度 を 高 い 精度(*0.0ppm以 内)で 安 定 して 測定 出来 る装 置

を 開発 し た。 分 析 計 の 出 力 と濃 度 を 関 係 づ け るた め に 必 要 な 標 準 ガ ス(Primary standards)は,高 純 度 の空 気
と二 酸 化 炭 素 の混 合気 体 で あ り,超 高 精 度 天 秤 を 用 い て重 量 法 に よっ て製 造 した。 濃 度 の絶 対 精 度 は*0.3ppm
以 内 で あ っ た。 フ ラ ス コに よる 試 料 空 気 の 採 集 方 法 や試 料 空 気 を フ ラス コに 長 期 保 存 し た場 合に 生 ず る変 質,さ
らに は試 料空 気 を採 集 す る際 に フ ラス コ 内に 凝 結 した 水蒸 気 が 分 析結 果 に 及 ぼ す 影 響 等 に つ い て も詳 細 に検 討 を
行 った 。