The Southern Ocean (south of 30
Greenhouse gases, such as
The annual basin-scale Southern Ocean carbon flux is generally well
constrained
For ocean
With few representative locations suitable for measuring atmospheric
A key challenge when measuring atmospheric
This paper focuses on the technical aspects of the Macquarie Island in situ
Macquarie Island isthmus map
Macquarie Island is 34 km long and 5 km wide at its widest
point
The clean-air laboratory is located on a low-lying (6 m a.s.l.) isthmus
between the main body of the island (a plateau 200–400 m a.s.l.) and a
small hill at the northern end of the island (Fig.
Maintaining an in situ instrument on Macquarie Island is logistically
challenging. Since there is no airport, access has been restricted to an
annual resupply voyage in March or April. All instrument servicing must be
completed in the “resupply window”, which is generally less than a week. As
the resupply ship cannot dock on the island, all equipment and personnel must
be transported from the ship to the shore by either helicopter or small boat.
These restrictions make Macquarie Island less accessible than many Antarctic
sites, possibly the most inaccessible of all sites in the current
The clean-air laboratory also houses an atmospheric radon monitor, the output
of which can be useful for interpreting the
Carbon dioxide mole fractions have been measured from
April 2005 to October 2016 using a CSIRO LoFlo Mark2
Dual-stage regulators (high purity, stainless steel, 64-3400 series, Tescom
Corporation, Elk River, Minnesota, USA) are used on all reference and
calibration cylinders, and all fittings and tubing used throughout the system
are stainless steel. Each hour the instrument alternates between 10 min of
reference measurement (when reference gas is passed through both cells of the
Li-COR) and 50 min of sample measurement (reference in one cell and sample
gas in the other). While temperature, pressure and flow rates are tightly
controlled within the system, small variations in flow and pressure occur
following the switch between sample and reference modes. Consequently, the
first 6 min after a switch are excluded to ensure that the flow and pressure
have stabilised. The performance of the instrument over the remaining 44 min
is explored further in Sect.
Despite the remote location of the instrument, instrument performance has been remarkable, with only 3.4 % of collected data points rejected due to poor instrumental performance (software failures and sporadic flow rate and temperature issues). Many of these were in the first year, with the annual average data lost for 2006 onwards being only 2.3 %.
Ambient air is sampled from 7 m a.g.l. (13 m a.s.l.) through an inverted
stainless steel cup with a 4 mm mesh covering the inlet. Quarter inch
polymer-coated aluminium tubing (Dekoron®
“1300”) is used between the inlet and pump manifold with the intake line
positioned so a continuous descent towards the pump is maintained. A simple
manifold system is used, consisting of 2 and 7
Air entering the drying system is immediately split into two: half is dried
using two 200 mL drying towers filled with magnesium perchlorate, the other
half, the air entering the LoFlo, is dried using a Nafion drier. To minimise
Internal drying reagent and
MQA LoFlo measurements are made relative to an assigned concentration of the
reference cylinder consisting of Southern Ocean ambient air (see
Sect.
Calibration runs consist of alternating 10 min reference (reference in both cells) and calibration
(reference in one cell and calibration gas in the other) measurements. As for
the normal sampling measurement procedure, the bracketing reference
measurements are deducted from the calibration gas measurement to remove
short-term instrumental drift. During a calibration run the cylinders are
measured first in ascending and then in descending order of
Error components of the Macquarie Island data site.
Like the reference gas cylinders, calibration cylinders are made using dry
Southern Ocean air collected at Cape Schanck, which is then modified to
achieve targeted mole fractions higher or lower than ambient using aliquots
of pure
The LoFlo2B calibration suite was calibrated directly against the WMO X2007
scale by the CCL on two occasions, 8 years apart. Differences for individual
cylinders varied, averaging 0.01
Calibration cylinder concentrations on the WMO X2007 scale as measured by LoFlo2B or GASLAB. Suite 2G-a was used from 2005 to April 2006 and Suite 2G-b from April 2006 to the present.
Two
Measurement uncertainty is typically composed of multiple elements and
evaluated using a statistical analysis of replicate measurements (Type A) or based on an
alternate source of information (Type B)
It is particularly important to characterise the measurement uncertainty of
the MQA record given the small atmospheric signals at midlatitudes to high latitudes in
the Southern Hemisphere. An earlier study documents the significant impact of
measurement errors and biases of LoFlo, conventional NDIR and flask
measurements on
Here, following the approach discussed earlier, we aim to quantify the
measurement uncertainty of the MQA
MQA measurements were calibrated following a multi-stage protocol
(Fig. The random uncertainty in measuring the the accuracy of the non-linearity correction with changes in the
absolute mole fraction difference between the reference and sample at both
the minutely and weekly timescale (Type 2), systematic within-hour variation in the sample-reference the mole fraction stability of the reference standard over time
(Type 4), the propagation of mole fractions to the 2G calibration suites from the
WMO X2007 scale via the LoFlo2B instrument (Type 5).
Here we quantify each of these five contributions to measurement uncertainty, thus providing a framework for defining uncertainties specific to data applications, e.g. involving different averaging periods or comparison with other data sets. Combining uncertainties of all five types in quadrature defines the overall measurement uncertainty when comparing measurements, including those of other laboratories, that are independently calibrated against the WMO X2007 scale. Comparisons of measurements made within the CSIRO network on similar instruments relative to LoFlo2B will have a significantly smaller Type 5 component. The uncertainty analysis uses only data with stable instrumental temperature and pressure and also excludes measurements made shortly after valve switches to minimise line conditioning effects. Uncertainties inherent in the sample handling or intake system, involving potential modification of sample air before being admitted to the LoFlo instrument, have not been examined.
These two uncertainty types were assessed using regular measurements of the
second suite of calibration standards (2G-b) as a proxy for in situ air data.
This analysis was based on 80 calibration runs between 2006 and 2013.
Each calibration run included between 16 and 144 (mean
Minute-mean mole fractions of the calibration standard data (i.e. the proxy air samples) were calculated for each run using the non-linearity correction determined in the previous calibration run. This represents a worst-case scenario, as in situ mole fractions will generally be calculated using a non-linearity correction determined much closer in time and will not be affected by any regulator or gas handling or switching effects.
First we examined uncertainty in the non-linearity correction characteristic
of the 1 min timescale. The minute-mean 1
The slope of the line is 0.0001, indicating an uncertainty of 0.01 % of the
sample-reference mole fraction difference at a 1 min timescale. This Type
2 mole fraction dependent component of uncertainty is negligible for the vast
majority of in situ measurements since at MQA, 99.9 % of minute
measurements are within 10
The same data set was used to evaluate uncertainty in the non-linearity
correction over timescales of a few weeks, which relates to the time period
between calibration runs. For this case we calculated the mean
Standard 994 235 was a clear outlier in this analysis (low open circle
Fig.
Interestingly the
As for the Type 2 uncertainty in minute means, this component is again
typically very small, less than 0.008
Between calibration runs, which are
performed several weeks apart, the instrument operates in routine in situ
monitoring mode. This involves an hourly cycle of alternating measurements of
reference and ambient MQA air. The first 10 min of each hour are used for
reference measurements (reference in both cells) to determine the difference
in output between cells. This difference is used by the data-processing
algorithm to define a background signal, interpolated between successive
reference measurements made every hour, against which ambient
The first 6 min of data from both the reference and ambient air measurement
periods are excluded from further processing due to stabilisation of flow
rate and pressure in the sample side cell after the valve switch. For ambient
air,
In order to resolve these small instrumental artefacts in ambient
The curves for different years are very similar in shape, with deviations
being largest in the early minutes and then decaying to zero at around
minute 45. There is a suggestion that the magnitude of deviation has
increased over time, with 2006 showing the smallest deviation at minute 16 of
0.02
We assume here that the latter, more stable part of the ambient measurement
period provides the most reliable
Definition of the Type 3 uncertainty applicable to minute means is more
complex, as it comprises both random and systematic components, varies with
minute number within the hour, and in some respects, increases with time
(i.e. increasing maximum deviation between 2006 and 2014 as displayed in
Fig.
The uncertainty inherent in assuming that the
The short-term variability of each cylinder (Fig.
The mole fractions of the 2G calibration suite were linked to the WMO X2007
scale using measurements made on LoFlo2B against the 2B calibration suite,
which is, in turn, linked to the WMO X2007 scale (Fig.
Similarly to the earlier discussion for LoFlo2G, the remaining LoFlo2B
uncertainties can be separated into Types 1, 2, 3 and 4. Combining in
quadrature the 2B propagation uncertainty with Type 1, 2, 3 and 4
uncertainties estimated based on the worst-case 2G uncertainties, the 2G WMO
X2007 propagation error was estimated as 0.024
Combined uncertainty estimates in
By geometrically combining appropriate uncertainty types and selecting key
factors, it is possible to give a series of examples of the expected minute-mean and hourly uncertainties for different situations
(Table
Typically the uncertainty is dominated by the Type 5 uncertainty component,
which in turn is comprised mainly of the propagation uncertainty to the WMO X2007 scale. As
such, the applicable uncertainty is highly dependent on the network choice,
decreasing by up to 40 % when considering within-network
Minute mean (black dot, left axis) and SD (blue dot, right axis) of
MQA
The minute-mean
Hourly wind speed
Figure
Figure
Hourly wind speed
Finally we examine a period without large deviations (Fig.
In the remainder of this section and in Sect.
Frequency histograms of
The distribution of minute SDs of
Figures
Since 1992, pairs of air samples have been collected fortnightly at MQA, in
0.5 L glass flasks using flask sampling techniques described by
Filled flasks are stored and then shipped back annually to CSIRO GASLAB
(Aspendale, Australia), where they are analysed for
All measurements derived from CSIRO flask samples require a correction for the loss of
Storage time
These corrections are especially significant for CSIRO's low-volume (0.5 L)
flasks and at sites such as MQA, where storage times can exceed a year. Loss
rates have been determined by comparing data from CSIRO's southern high-latitude sites, where flasks can be stored for a year or so before analysis,
with smoothed baseline concentrations at Cape Grim, Tasmania, derived from
flask sample data with relatively short storage times. Using data from
1992 to 2007, a correction of 0.002
LoFlo observations were compared to individual flask sample data by taking
the mean of the hours before and after the flask filling time or either hour
if only one was available. This identified 361 matching records after flagged
flasks had been excluded. Flask-LoFlo concentration differences are shown in
Fig.
The aim of most long-term atmospheric
Local flux influences on the
Proportion of hours lost (blue) and proportion of rejected hours
that are outliers (red) for a given minute
We test this selection technique by rejecting data for a range of MMSD
values. Effective selection is demonstrated by a reduction in short-term
variability in the data through removal of outliers without excessive data
rejection. The short-term variability is determined by fitting a smooth curve
to the hourly data (Sect.
Figure
Figure
Atmospheric radon concentrations are simulated as in
Figure
Figure
A smooth curve was fit to the hourly
Figure
Using the maximum minute
Macquarie Island
The long-term trend in MQA LoFlo
The seasonal variation in
It is important to note that the interpretation of the interannual variations in
the MQA LoFlo seasonality cannot only consider interannual variations in
Southern Ocean fluxes. Tropical and Northern Hemisphere fluxes also make a
significant contribution to seasonality across the Southern Ocean
The Southern Ocean plays a key role in the global
The in situ nature of this record (unlike the traditional flask measurements)
results in an increase in the temporal frequency of the data and hence a far
richer data stream. The in situ record and its statistically derived products
(baseline, growth rate, long-term trend and seasonality) are consequently
more robust than those of the co-located flask record, which is also impeded
by long sample storage times. The increased temporal frequency has revealed
diurnal and synoptic variations in atmospheric
The fortran version of the curve-fitting code used in this
paper is not publicly available. However, a C language programme version can
be found at
The MQA LoFlo
ARS and RML analysed the data, performed the model simulations and wrote the paper with input from co-authors; ARS and MVS serviced and managed the instrument; MVS installed the instrument; RF led the development of the LoFlo instrumentation and calibration strategy; RLL contributed to the uncertainty analysis and provided the MQA flask data and its analysis; ARS, MVS, DAS, PBK and RTH contributed to calibration, database and analysis software development; SDC, AGW and SW contributed the radon data.
The authors declare that they have no conflict of interest.
The authors would like to thank L. Paul Steele for his simulating discussions and suggestions in relation to this paper and acknowledge his significant involvement in the development and calibration of the LoFlo system and the GASLAB flask sampling programme. This research was funded in part by the Australian Government Department of the Environment, the Bureau of Meteorology, and CSIRO through the Australian Climate Change Science Programme and directly by CSIRO. The authors would also like to acknowledge the in-kind support of the Australian Antarctic Division, under project no. 4167 – Greenhouse gases in the southern atmosphere, and the Australian Bureau of Meteorology. CCAM modelling was undertaken on the NCI National Facility in Canberra, Australia, which is supported by the Australian Commonwealth Government. Back trajectories were calculated using the HYSPLIT transport and dispersion model from NOAA Air Resources Laboratory. Ot Sisoutham has provided support to the Macquarie Island and Southern Ocean radon programme. Maps used in Fig. 1 are courtesy of the Australian Antarctic Division. Edited by: Keding Lu Reviewed by: two anonymous referees