Space-based measurements of HCl: Intercomparison
and historical context
D. J. Lary1,2 and O. Aulov1,2
Received 28 March 2007; revised 26 October 2007; accepted 21 November 2007; published 18 April 2008.
[1] The peak in stratospheric HCl was reached in the late 1990s. Between 1998 and 2004
the stratospheric loading of HCl was relatively constant, with some month to month
fluctuation; this was followed by a more pronounced decrease in HCl since 2004. We use
probability distribution functions (PDFs) and scatter diagrams for validation and bias
characterization of Aura Microwave Limb Sounder (MLS) HCl retrievals. Both these
methods allow us to use large statistical samples and do not require correlative
measurements to be colocated in space and time. The bias between the Halogen
Occultation Experiment (HALOE) and Aura MLS is greatest above the 525 K (
21 km)
isentropic surface. The global average mean bias between Aura and the Atmospheric
Chemistry Experiment (ACE) for January 2005 was 2% and between Aura MLS and
HALOE was 14%. The widths of the PDFs are a measure of the spatial variability and
measurement precision. The Aura MLS HCl PDFs are consistently wider than those for
ACE and HALOE, this reflects the retrieval uncertainties. The median observation
uncertainty for Aura MLS v1.51 HCl is 12%, and the median ACE v2.2 uncertainty is 8%.
We also connect Aura MLS HCl with the heritage of HALOE HCl by using neural
networks to learn the interinstrument biases and provide a seamless HCl record from the
launch of the Upper Atmosphere Research Satellite (UARS) in 1991 to the present.
Citation: Lary, D. J., and O. Aulov (2008), Space-based measurements of HCl: Intercomparison and historical context, J. Geophys.
Res., 113, D15S04, doi:10.1029/2007JD008715.
1. Introduction
[2] The Microwave Limb Sounder (MLS) on Aura is
providing the first daily global observations of HCl [Waters
et al., 2006; Froidevaux et al., 2006b]. A preliminary
validation of MLS HCl has been presented by Froidevaux
et al. [2006a].
[3] Evaluation and validation of satellite data are neces-
sary, but sampling issues can make practical application
problematic. In the traditional approach to validation, there
is a comparison of matched pairs of profiles coincident in
space and time. This strong constraint dramatically reduces
the statistical sample sizes we can deal with. The definition
of ‘‘coincident’’ observations varies, but 1000 km or more
often separates such measurements. Establishing instrument
accuracy or precision through such comparisons can be
hindered by the limited number of coincident pairs and the
contribution of atmospheric variability (termed representa-
tiveness). Issues of representativeness arise because the
validation exercises are typically limited geographically. It
is therefore useful to augment the traditional approach to
validation with the use of probability distribution functions
(PDFs) of trace gases over an extended period for a given
spatial domain. In this study, we choose to form PDFs of an
entire month of data and to specify the spatial domain in
terms of Lagrangian flow-tracking coordinates. The width
of the PDFs allow us to quantify the spatial variability for
each analysis grid cell over the month. The analysis starts
with the launch of UARS and continues up to the present.
The scatter diagrams allow us to compare a pair of instru-
ments globally over the entire period of overlap using the
PDFs for each month, for each Lagrangian region.
[4] PDFs have already been used in a variety of tracer
studies [Pierrehumbert, 1994; Yang, 1995; Sparling and
Schoeberl, 1995; Rood et al., 2000; Hu and Pierrehumbert,
2001; Gao et al., 2002; Johnson et al., 2002; Strahan, 2002;
Neu et al., 2003; Hsu et al., 2004] and in estimation of
representativeness uncertainty in chemical data assimilation
[Lary, 2003].
[5] Not only does a PDF characterize the tracer distribu-
tion, its shape tells us about mixing barriers, how complete
the mixing is, and chemical processes such as ozone
depletion [Sparling, 2000; Pierrehumbert, 2000; Strahan,
2002; Neu et al., 2003]. For example, a narrow peak in
the concentration PDF indicates that the air is well mixed
and significant variability generating processes have not
recently occurred (e.g., long-range transport). A multimodal
distribution indicates air of different origins (e.g., polar
and midlatitude). In general, broad peaks indicate recent
variability-generating processes such as photochemistry or
transport (horizontal or vertical).
JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 113, D15S04, doi:10.1029/2007JD008715, 2008
Click
Here
for
Full
Article
1Goddard Earth Sciences and Technology Center, University of
Maryland Baltimore County, Baltimore, Maryland, USA.
2Atmospheric Chemistry and Dynamics Branch, NASA Goddard Space
Flight Center, Greenbelt, Maryland, USA.
Copyright 2008 by the American Geophysical Union.
0148-0227/08/2007JD008715$09.00
D15S04 1 of 10

[6] Measurement imprecision is one factor that affects
the widths of the PDFs, and precision of the measurements
is certainly a parameter that needs validation. Because a
major component of the variability of trace gases is due
to atmospheric transport we make our comparisons in
equivalent PV latitude–potential temperature coordinates
[Schoeberl et al., 1989; Proffitt et al., 1989; Lait et al.,
1990; Douglass et al., 1990; Lary et al., 1995; Schoeberl et
al., 2000]. Using these coordinates also extends the effec-
tive latitudinal coverage of the measurements, although
there will always be limitations arising from sampling
issues. In the period 1991–2004 we use the UKMO
meteorological analyses, and after that we use NASA GEOS
meteorological analyses.
[7] Section 2 describes the HCl intercomparison using
PDFs and scatter diagrams and the cross calibration of HCl
retrievals using neural networks. Section 3 presents a
continuous time series of HCl from the launch of UARS
to the present. Section 4 gives a summary.
2. HCl Intercomparison
[8] We compare measurements of HCl from the different
instruments in Table 1. Table 1 also gives the median
observation uncertainty over the entire record of each
instrument. The Halogen Occultation Experiment (HALOE)
provides the longest record of space based HCl observa-
tions. Aura MLS has a vertical resolution which is 3 km in
the lower stratosphere increasing to 5–6 km near 1 hPa and
7 km near 0.22 hPa. ACE and HALOE have a vertical
resolution of about 4 km. The ACE and HALOE retrievals
are given on a much finer altitude grid, with a spacing of
1 km or less. The Aura MLS retrievals used are given on
a pressure grid with an approximate altitude spacing of

2.5 km.
2.1. PDFs
[9] Figure 1 shows example HCl PDFs for the three
instruments HALOE, ACE and Aura MLS. In each case
the PDFs are for all observations made by that instrument in
a Lagrangian region for three isentropic levels centered on
an equivalent latitude of 55
N during all the Januarys that
the instruments observed (Figures 1a to 1c) or for the
observations made only during January 2005 (Figures 1d
to 1f). A consistent picture is seen in these plots: Aura MLS
agrees very well with ACE, while HALOE HCl retrievals
are lower than those from the other instruments. There is a
general increase in the bias with increasing altitude, partic-
ularly noticeable at the 525 K (
21 km) surface and above.
Previous comparisons among HCl data sets reveal a similar
bias for HALOE [Russell et al., 1996; McHugh et al., 2005;
Froidevaux et al., 2006a].
2.2. Width of the PDFs
[10] The width of the PDFs, srep, gives us a measure of
the spatial variability (representativeness) in the tracer field
[Lary, 2003]. A robust estimator of the width of the
distribution is the average deviation [Press et al., 1992],
srep ¼ ADev y1 . . . yNð Þ ¼
1
N
XN
j¼1
yj  y
  ð1Þ
where y is the tracer volume mixing ratio (VMR), and the
overbar indicates the mean of the N observations consid-
ered. It is interesting to look at a cross section of the
representativeness as it highlights the regions with large
gradients. Figure 2 shows the PDF width for HCl
observations made by Aura MLS during January 2005.
The upwelling air over the tropics is visible as is the large
spatial variability in the lower stratospheric polar vortex at
high northern latitudes.
[11] The width of the PDF could be characterized in
several ways. Two other common measures are the vari-
ance, or its square root, the standard deviation. Both of these
measures have been tried and give essentially the same
results. The reason for choosing the average deviation is
that the variance and standard deviation depend on the
second moment of the PDF and can be unduly affected by
a few outliers. The average deviation is a more robust
estimator that does not suffer from this problem [Press et
al., 1992].
[12] The PDF width is actually a function of several
things including spatial variability, sampling, and retrieval
uncertainty and noise. The all years January plots in
Figures 1a–1c clearly include a wider variety of situations
than do the specific January plots (Figures 1d–1f), so as
would be expected, these PDFs are wider. It is interesting
to note that even though the median values of the Aura
MLS and ACE PDFs are very similar, the width of the
Aura MLS HCl PDFs are consistently larger than those for
ACE and HALOE, this may well reflect in part the retrieval
uncertainties shown in Table 1. The median observation
uncertainty for Aura MLS HCl is 12%, which is 50% larger
than the median ACE v2.2 uncertainty of 8%. However, the
instruments generally have a similar spatial distribution,
e.g., both Aura MLS and ACE have wide PDFs for the
lower stratosphere vortex, and narrow PDFs in the upper
Table 1. Instruments Used in This Studya
Instrument Temporal Coverage Reference
Median Observation
Uncertainty
ACE v2.2 2004–2006 Bernath et al. [2005] 8%
ASUR 1993–1994, 1999, 2000 Mees et al. [1995]
ATMOS v3 1992, 1993, 1994 Zander et al. [1992] 8%
Aura MLS v1.51 2004–2006 Froidevaux et al. [2006a] 12%
HALOE v19 1991–2005 Russell et al. [1993] 4%
MkIV 1996, 1997, 1999, 2000, 2002–2005, 2007 Toon et al. [1999] 10%
aThe uncertainties given are the median uncertainty of all the level 2 product uncertainties for all the observations made by each instrument. The HALOE
uncertainties are only estimates of random error and do not include any indications of overall accuracy.
D15S04 LARY AND AULOV: HCl INTERCOMPARISON AND HISTORICAL CONTEXT
2 of 10
D15S04

Figure 1. Example HCl PDFs for the three instruments HALOE, ACE, and Aura MLS. In each case
the PDFs are for all observations made by that instrument in a Lagrangian region for three isentropic
levels centered on an equivalent latitude of 55
N during all the Januarys that the instrument observed.
For ACE the plots include January 2004–2006, for HALOE the plots include 2004–2005, and for MLS
the plots include 2005–2006. (a) Plot of a PDF for all observations in the range 410 K < q < 450 K
(70 mbar < P < 110 mbar), 49
< fel < 61
. (b) Plot of a PDF for all observations in the range 460 K < q <
590 K (30 mbar < P < 60 mbar), 49
< fel < 61
. (c) Plot of a PDF for all observations in the range
1100 K < q < 1500 K (2 mbar < P < 4 mbar), 49
< fel < 61
. (d) to (f) Analogous to Figures 1a to 1c for
the observations made only during January 2005. The number of observations used to form each PDF is
shown in parentheses in the legend.
D15S04 LARY AND AULOV: HCl INTERCOMPARISON AND HISTORICAL CONTEXT
3 of 10
D15S04

stratosphere. Likewise, the HALOE PDFs for a given
month, are narrower than the ACE PDFs and the median
HALOE uncertainty is less than median ACE observation
uncertainty (e.g., Figures 1a–1c). As pointed out by
E. Remsberg (personal communication, 2007), the width
of the PDFs also depends on the spatial extent of the
observations. For example, HALOE made sunrise measure-
ments in the northern hemisphere from 1–9 January and
from about 16 January to the end of the month. HALOE
obtained data in the regions shown in Figure 1 for only a
few days of January 2005. Because there is normally a lot of
wave activity in the northern hemisphere in January, it
would be easy for HALOE not to see a lot of the variability
that was present for the month of January for that latitude
zone. And that lack of coverage for the month is another
reason for the narrower HALOE PDFs in Figures 1d–1f.
2.3. Biases
[13] We can take the difference between the medians of
the PDFs as a measure of the interinstrument bias. This bias
is really only significant if it is larger than the atmospheric
variability in the Lagrangian region we are considering.
[14] Figure 3 shows interinstrument biases for January 2005
for Aura, ACE and HALOE. Figures 3a, 3c, and 3e show
the bias as a volume mixing ratio (VMR). Figures 3b, 3d,
and 3f show the percentage bias. Figures 3a to 3d are for the
biases between ACE v2.2 and Aura MLS v01. In Figures 3a
and 3b we show all available Lagrangian regions where
both ACE and Aura made observations during January
2005. In Figures 3c and 3d we only plot Lagrangian regions
where the bias was greater than the natural HCl variability
in that region of the atmosphere, we have called this the
useful bias. The natural variability (representativeness) has
been diagnosed by taking the width of the PDF as measured
by the average deviation of the PDF.
[15] We note that for the January 2005 example, the bias
between Aura and ACE in the lower stratosphere is less than
the natural variability. The average mean bias between Aura
and ACE for January 2005 was 2.3%. Figures 3e and 3f
show the analogous bias for HALOE and Aura, only those
Lagrangian regions where the bias was greater than the
natural HCl variability have been plotted. We note that for
the January 2005 example, the bias between HALOE and
Aura is greater than the natural variability throughout most
of the stratosphere. The average mean bias between Aura
and HALOE for January 2005 was 13.9%. In general, the
number of points from both of the solar occultation instru-
ments (HALOE and ACE) is much less than the number of
points from the microwave emission instrument (MLS).
2.4. Scatter Diagrams and Cross Calibration
[16] So far, we have compared the PDFs for all over-
lapping Lagrangian regions for a given month. However, we
can use a single scatter diagram to compare all the overlaps
globally for all the months observed by each pair of instru-
ments. Such a scatter diagram has the advantage of a huge
sample size, it encompasses the entire period that a pair of
instruments were making contemporaneous observations.
The scatter diagram is intended as a big picture summary for
all contemporaneous observations made globally. Figure 4
shows two scatter diagrams for all the contemporaneous
observations of HCl made globally by two pairs of instru-
Figure 2. PDF width characterized by the average deviation for HCl observations made by Aura MLS
during January 2005. The width of the PDFs, srep, gives us a measure of the spatial variability
(representativeness) in the tracer field and highlights the regions with large spatial gradients.
D15S04 LARY AND AULOV: HCl INTERCOMPARISON AND HISTORICAL CONTEXT
4 of 10
D15S04

Figure 3. Interinstrument biases for January 2005. These biases are the differences in the PDF medians
for each instrument. (left) Bias as a volume mixing ratio (VMR). (right) Percentage bias. The bias is only
significant if it is greater than the natural variability in that region of the atmosphere. The natural
variability (representativeness) has been diagnosed by taking the width of the PDF as measured by the
average deviation of the PDF. (a) to (d) For the biases between ACE v2.2 and Aura MLS v01. In Figures 3a
and 3b we show all available Lagrangian regions were both ACE and Aura made observations during
January 2005. In Figures 3c and 3d we only plot Lagrangian regions where the bias was greater than the
natural HCl variability in that region of the atmosphere, we have called this the useful bias. It can be seen
that for most of the Lagrangian regions observed by both ACE and Aura the bias is less than 10%. (e) and
(f) Analogous bias for HALOE and Aura where the bias was greater than the natural HCl variability.
D15S04 LARY AND AULOV: HCl INTERCOMPARISON AND HISTORICAL CONTEXT
5 of 10
D15S04

ments. In Figure 4a we compare ACE and Aura MLS which
were making contemporaneous observations between
September 2004 and the present. In Figure 4b we compare
HALOE and Aura MLS which were making contemporane-
ous observations between September 2004 and November
2005.
[17] In the ideal case where we have perfect agreement
between two instruments, the slope of the scatter diagram
would be 1 and the intercept would be 0. In the case of
ACE and Aura, we see there is a slope of 1.08, and for
HALOE Aura there is a slope of 0.91. It can be seen that
solely using the slopes does not do justice to the differences.
For example, Figure 4a shows a much better agreement
than Figure 4b, but the slopes themselves do not reflect this.
Figure 4b shows an offset of order 9% near 2 ppbv, whereas
Figure 4a shows maybe 1 or 2% for values near 2 ppbv. The
mean absolute value of the differences seems a good
indicator of the fits. We also note that in the case of Aura
MLS and HALOE, the scatter diagrams do not have a
constant slope over the entire range of HCl values, several
‘‘wiggles’’ are present. This means that the interinstrument
biases are spatially and temporally dependent. Neural net-
works are multivariate, nonparametric, ‘‘learning’’ algo-
rithms that are ideally suited to learning, and correcting
for, such interinstrument biases.
[18] We have used a neural network with three inputs and
one output. The inputs are equivalent PV latitude, potential
temperature, and HCl from instrument A. The output is HCl
from instrument B. Potential temperature and equivalent
latitude are used because they are good markers of the large-
scale flow pattern [Lary and Mussa, 2004]. When we do the
training we randomly split our training data set into three
portions of 80%, 10% and 10%. The 80% is used to train
the neural network weights. This training is iterative and on
each iteration we evaluate the current RMS error of the
neural network. The RMS error is calculated by using the
second 10% of the data that was not used in the training. We
use the RMS error and the way it changes with training
iteration (epoch) to determine the convergence of our
training. When the training is complete, we use the final
10% as a validation data set. This 10% of the data was
randomly chosen and not used in either the training or RMS
evaluation. We only use the neural network if the validation
scatter diagram, which plots the actual data from validation
portion against the neural network estimate, yields a straight
line graph with a slope of 1. This is a stringent and
independent validation. The validation is global as the data
was randomly selected over all temporal and spatial data
points available. Several training strategies were examined,
the one described included the most species over the longest
time period. The neural network algorithm used was a feed-
forward back-propagation network with 20 hidden nodes.
The training was done by the Levenberg-Marquardt back-
propagation algorithm provided by the Matlab neural net-
work toolbox (available at http://www.mathworks.com/
products/neuralnet/).
[19] Figures 5a and 5c show the results of such a neural
network training to learn interinstrument biases between
ACE v2.2, Aura MLS v1 and HALOE v19 HCl. Figures 5b
and 5d show an independent validation of the training using
a randomly chosen, totally independent, data sample not
used in training the neural network. In each case the x axis
shows the actual ACE v2.2 HCl (the target). In Figures 5a
and 5b the y axis is the neural network estimate of ACE
v2.2 HCl based on Aura MLS v01 HCl. Figure 5a shows the
results using the training data, Figure 5b shows the results
of the independent validation. In Figures 5c and 5d the
y axis is the neural network estimate of ACE v2.2 HCl
based on HALOE v19 HCl. Figure 5c shows the results
using the training data, and Figure 5d shows the results of
the independent validation. The mapping has removed the
bias between the measurements and has also straightened
out the ‘‘wiggles’’ seen in Figure 4.
Figure 4. (a) and (b) Scatterplots of all contemporaneous observations of HCl made by HALOE, ACE,
and Aura MLS. Each point plotted is the median value of a PDF of observations made for a Lagrangian
region over the period of a month. It can be seen that to adequately using the slopes to describe the
differences does not do justice to the differences. For example, Figure 4a shows a much better agreement
than Figure 4b, but the slopes themselves do not reflect this. Figure 4b shows an offset of order 9% near
2 ppbv, whereas Figure 4a shows maybe 1 or 2% for values near 2 ppbv. The mean absolute value of the
differences seems a good indicator of the fits.
D15S04 LARY AND AULOV: HCl INTERCOMPARISON AND HISTORICAL CONTEXT
6 of 10
D15S04

Figure 5. (left) Result of training a neural network to learn the interinstrument biases. (right) An
independent validation of this training using a randomly chosen, totally independent, data sample not
used in training the neural network. In each case, the x axis shows the actual ACE v2.2 HCl (the target).
(a) and (b) The y axis is the neural network estimate of ACE v2.2 HCl based on Aura MLS v01 HCl.
Figure 5a is the result using the training data, and Figure 5b is the result of the independent validation.
(c) and (d) The y axis is the neural network estimate of ACE v2.2 HCl based on HALOE v19 HCl.
Figure 5c is the result using the training data, and Figure 5d is the result of the independent validation. In
each case, this is a global training for all contemporaneous observations between each pair of instruments.
The training points are the median values of a PDF of observations made during a given month for a
given equivalent PV latitude–potential temperature bin. The width of the cloud of points in each of these
scatter diagrams is a good measure of the uncertainty associated with the neural network fit.
D15S04 LARY AND AULOV: HCl INTERCOMPARISON AND HISTORICAL CONTEXT
7 of 10
D15S04

2.5. Evaluating the Goodness of Fit
[20] We have objectively evaluated the goodness of the
neural network fit by using the reduced cr2 metric, which is
really just the root mean square deviation normalized to
measurement errors,
c2r ¼
1
n
XN
j¼1
yNN  yj
 2
s2j
ð2Þ
where yNN is the neural network estimate of the benchmark
observation, y is the benchmark observation, and s is the
observation uncertainty, n = N  n is the degrees of
freedom, N is the total number of benchmark observations,
and n is the number of parameters used in the fit.
[21] Table 2 shows the reduced cr2 statistics for a variety
of HCl products. In order to be more critical the HCl
validation products were only used if the fractional obser-
vational uncertainty of the benchmark observation was
less than 0.1. For each benchmark instrument the results
in Table 2 are sorted in descending order of cr2, so the last
product in each group is the product that agreed most
closely with the benchmark instrument. In each case the
product with the largest (worst) cr2 was HALOE HCl, and
applying a neural network recalibration considerably im-
proved the agreement with the benchmark instruments
(reduced cr2).
3. HCl Time Series
[22] Now that we have completely characterized the
interinstrument biases and been able to correct for them
we can connect Aura MLS HCl observations to the heritage
of HALOE. This allows us to produce an HCl time
series from the launch of UARS in 1991 up-to the present.
Figure 6 shows HCl time series for six different locations
(on the 525 K  21 km, 800 K  30 km, 1300 K  41 km,
and 2100 K  50 km isentropic surfaces) from the launch of
UARS to the present with HCl observations from HALOE,
ATMOS, ACE, MkIV and Aura MLS.
[23] Figure 6 (left) uses the original v19 HALOE data,
and the low bias of HALOE HCl relative to all other
instruments is evident. Figure 6 (right) uses the HALOE
v19 data recalibrated with a neural network to agree with
ACE v2.2 HCl. If we compare Figures 6 (left) and 6 (right)
we see that as expected, the recalibration brings the HALOE
data into good agreement with ACE and Aura MLS data,
and the independent ATMOS HCl data.
[24] We also performed a recalibration of the ACE and
MLS data to agree with HALOE v19. These two HCl
recalibrations have been used by Lary et al. [2007] to form
a long Cly time series and associated uncertainty estimate
(typically 0.4 ppbv at 800 K). The uncertainty in the Cly
estimate is primarily due to the discrepancy between the
different observations of HCl, i.e., the HALOE, Aura MLS,
and ACE interinstrument biases.
[25] We have attempted to characterize the biases as
‘‘exactly’’ as possible for each space region. What we do
not do, due to a lack of temporal overlap, is have the bias
varying with time. Artifacts of this do appear to be evident
in some of the neural network adjustments, for example, in
the upper stratosphere, the adjustment of HALOE to MLS
HCl seems to yield a time series which is too flat, this is true
to a lesser extent for the HALOE to ACE HCl adjustment.
We note that the size of the training data set plays a role, the
overlap between HALOE and MLS is rather short and the
result of this can be seen in Figure 6e the corrected time
series using this short overlap (blue line) has the wrong
shape. The the overlap between HALOE and ACE is longer
and the red line in Figure 6e does not suffer from this
artifact. We are currently working on trying to improve this.
4. Summary
[26] We have used PDFs to characterize the interinstru-
ment biases between the HCl products provided by Aura
MLS v01, ACE v2.2, and HALOE v19. These biases are
presented in a number of ways, including global equivalent
latitude potential temperature cross sections for every month
of overlap between the instruments (available from http://
www.pdfcentral.info/). The bias between HALOE and Aura
MLS is greatest above the 525 K (
21 km) isentropic
surface. The global average mean bias between Aura and
ACE for January 2005 was 2% and between Aura and
HALOE was 14%. A scatter diagram compares all the
overlaps globally for all the months observed by a pair of
instruments.
[27] The width of the PDFs are a measure of the spatial
variability. The Aura MLS HCl PDFs are consistently larger
than those for ACE and HALOE, this reflects the retrieval
uncertainties. The median observation uncertainty for Aura
MLS HCl is 12%, which is 50% larger than the median
ACE v2.2 uncertainty of 8%. The instruments generally
have a similar spatial distribution, e.g., both Aura MLS and
ACE have wide PDFs in the lower stratosphere vortex, and
narrow PDFs in the upper stratosphere.
[28] We used neural networks to correct for interinstru-
ment biases and produce a consistent time series of HCl
from 1991 to the present. Such an HCl time series is of use
in estimating a time series of Cly .
Table 2. Evaluating Goodness of Fita
Product cr2 Number of Bins
ACE v2.2
HALOE v19 24 5480
HALOE v19 NN calibrated as MLS 15.3 5480
HALOE v19 NN calibrated as ACE 12.7 5480
Aura MLS v1 11.4 8895
ATMOS
HALOE v19 9.47 561
HALOE v19 NN calibrated as MLS 3.68 561
HALOE v19 NN calibrated as ACE 3.51 561
Aircraft ASUR
HALOE v19 63.8 477
HALOE v19 NN calibrated as ACE 39.6 477
HALOE v19 NN calibrated as MLS 36.9 477
aThe cr2 statistics for a variety of HCl products. In order to be more
critical, HCl validation products were only used if the fractional
observational uncertainty for the benchmark observation was less than
0.1. For each benchmark instrument the results are sorted in descending
order of cr2, so the last product in each group is the product that agreed most
closely with the benchmark instrument, the cr2 for this instrument is shown
in bold. The number of contemporaneous bins which were available for
each comparison is shown.
D15S04 LARY AND AULOV: HCl INTERCOMPARISON AND HISTORICAL CONTEXT
8 of 10
D15S04

[29] Acknowledgments. It is a pleasure to acknowledge NASA for
research funding, Lucien Froidevaux, and the Aura MLS team for their
data; the ACE team, Peter Bernath, Chris Boone, and Kaley Walker for their
data; the HALOE team and Ellis Remsberg for their data; the ATMOS team
for their data; the MkIV team for their data; and Bill Read, Anne Douglass,
Rich Stolarski, Paul Newman, Ross Salawitch, and anonymous reviewers
for numerous useful conversations. The ACE mission is funded primarily
by the Canadian Space Agency.
References
Bernath, P. F., et al. (2005), Atmospheric Chemistry Experiment (ACE):
Mission overview, Geophys. Res. Lett., 32, L15S01, doi:10.1029/
2005GL022386.
Douglass, A. R., R. B. Rood, R. S. Stolarski, M. R. Schoeberl, M. H.
Proffitt, J. J. Margitan, M. Loewenstein, J. R. Podolske, and S. E. Strahan
(1990), Global three-dimensional constituent fields derived from profile
data, Geophys. Res. Lett., 17(4), 525–528.
Froidevaux, L., et al. (2006a), Early validation analyses of atmospheric
profiles from EOS MLS on the aura satellite, IEEE Trans. Geosci. Remote
Sens., 44(5), 1106–1121.
Froidevaux, L., et al. (2006b), Temporal decrease in upper atmospheric
chlorine, Geophys. Res. Lett., 33, L23812, doi:10.1029/2006GL027600.
Gao, R. S., et al. (2002), Role of NOy as a diagnostic of small-scale mixing
in a denitrified polar vortex, J. Geophys. Res., 107(D24), 4794,
doi:10.1029/2002JD002332.
Hsu, J., M. J. Prather, O. Wild, J. K. Sundet, I. S. A. Isaksen, E. V. Browell,
M. A. Avery, and G. W. Sachse (2004), Are the TRACE-P measurements
representative of the western Pacific during March 2001?, J. Geophys.
Res., 109, D02314, doi:10.1029/2003JD004002.
Hu, Y., and R. T. Pierrehumbert (2001), The advection-diffusion problem
for stratospheric flow. part I: Concentration probability distribution
function, J. Atmos. Sol. Terr. Phys., 58(12), 1493–1510.
Johnson, D. R., A. J. Lenzen, T. H. Zapotocny, and T. K. Schaack (2002),
Numerical uncertainties in simulation of reversible isentropic processes
and entropy conservation: part II, J. Clim., 15(14), 1777–1804.
Lait, L., et al. (1990), Reconstruction of O3 and N2O fields from ER-2,
DC-8, and balloon observations, Geophys. Res. Lett., 17(4), 521–524.
Lary, D. (2003), Representativeness uncertainty in chemical data assimila-
tion highlight mixing barriers, Atmos. Sci. Lett., 5(1–4), 35–41.
Lary, D. J., and H. Y. Mussa (2004), Using an extended kalman filter
learning algorithm for feed-forward neural networks to describe tracer
correlations, Atmos. Chem. Phys. Disc., 4, 3653–3667.
Lary, D. J., M. P. Chipperfield, J. A. Pyle, W. A. Norton, and L. P.
Riishojgaard (1995), 3-dimensional tracer initialization and general diag-
Figure 6. Selected HCl time series from the launch of UARS to the present with HCl observations from
HALOE, ATMOS, ACE and Aura. (a) For 2100 K (
50 km) at 85
S, (b) and (c) for 1300 K (
41 km) at
55
N and 55
S, and (d) and (e) for 55
N at 800 K (
30 km) and 525 K (
21 km). In each case the green
line and shading is for the original HALOE v19 data, and the red line and shading is for HALOE data
remapped with a neural network to agree with Aura MLS v1 HCl. The black line is the ACE v2.2 data
with the grey shading representing the associated uncertainty. The d in the legend which labels the
shading refers to the total uncertainty (observational, representativeness, and where relevant, neural
network adjustment). The remapping of HALOE generally brings the HALOE data into better agreement
with the ATMOS (squares) data as can also be seen from the cr2 values shown in Table 2.
D15S04 LARY AND AULOV: HCl INTERCOMPARISON AND HISTORICAL CONTEXT
9 of 10
D15S04

nostics using equivalent PV latitude-potential-temperature coordinates,
Q. J. R. Meteorol. Soc., 121(521A), 187–210.
Lary, D. J., D. W. Waugh, A. R. Douglass, R. S. Stolarski, P. A. Newman,
and H. Mussa (2007), Variations in stratospheric inorganic chlorine be-
tween 1991 and 2006, Geophys. Res. Lett., 34, L21811, doi:10.1029/
2007GL030053.
McHugh, M., B. Magill, K. A. Walker, C. D. Boone, P. F. Bernath, and J. M.
Russell III (2005), Comparison of atmospheric retrievals from ACE and
HALOE, Geophys. Res. Lett., 32, L15S10, doi:10.1029/2005GL022403.
Mees, J., S. Crewell, H. Nett, G. Delange, H. Vandestadt, J. J. Kuipers, and
R. A. Panhuyzen (1995), Asur—An airborne sis receiver for atmospheric
measurements of trace gases at 625 to 760 ghz, IEEE Trans. Microwave
Theory Tech., 43(11), 2543–2548.
Neu, J. L., L. C. Sparling, and R. A. Plumb (2003), Variability of the
subtropical ‘‘edges’’ in the stratosphere, J. Geophys. Res., 108(D15),
4482, doi:10.1029/2002JD002706.
Pierrehumbert, R. T. (1994), Tracer microstructure in the large-eddy
dominated regime, Chaos Solitons Fractals, 4(6), 1091–1110.
Pierrehumbert, R. T. (2000), Lattice models of advection-diffusion, Chaos,
10(1), 61–74.
Press, W., S. Teukolsky, W. Vetterling, and B. Flannery (1992), Numerical
Recipes in FORTRAN—The Art of Scientific Computing, 2nd ed.,
Cambridge Univ. Press, New York.
Proffitt, M., et al. (1989), In situ ozone measurements within the 1987
Antarctic ozone hole from a high-altitude ER-2 aircraft, J. Geophys.
Res., 94(D14), 16,547–16,555.
Rood, R. B., A. R. Douglass, M. C. Cerniglia, L. C. Sparling, and J. E.
Nielsen (2000), Seasonal variability of middle-latitude ozone in the low-
ermost stratosphere derived from probability distribution functions,
J. Geophys. Res., 105(D14), 17,793–17,805.
Russell, J. M., et al. (1993), The HalogenOccultation Experiment, J. Geophys.
Res., 98(D6), 10,777–10,797.
Russell, J. M., et al. (1996), Validation of hydrogen chloride measurements
made by the Halogen Occultation Experiment from the UARS platform,
J. Geophys. Res., 101(D6), 10,151–10,162.
Schoeberl, M. R., et al. (1989), Reconstruction of the constituent distribu-
tion and trends in the Antarctic polar vortex from ER-2 flight observa-
tions, J. Geophys. Res., 94(D14), 16,815–16,845.
Schoeberl, M. R., L. C. Sparling, C. H. Jackman, and E. L. Fleming (2000),
A Lagrangian view of stratospheric trace gas distributions, J. Geophys.
Res., 105(D1), 1537–1552.
Sparling, L. C. (2000), Statistical perspectives on stratospheric transport,
Rev. Geophys., 38(3), 417–436.
Sparling, L. C., and M. R. Schoeberl (1995), Mixing entropy analysis of
dispersal of aircraft emissions in the lower stratosphere, J. Geophys. Res.,
100(D8), 16,805–16,812.
Strahan, S. (2002), Influence of planetary wave transport on Arctic ozone
as observed by Polar Ozone and Aerosol Measurement (POAM) III,
J. Geophys. Res., 107(D20), 4417, doi:10.1029/2002JD002189.
Toon, G. C., et al. (1999), Comparison of MKIV balloon and ER-2 aircraft
measurements of atmospheric trace gases, J. Geophys. Res., 104(D21),
26,779–26,790.
Waters, J. W., et al. (2006), The Earth Observing System Microwave Limb
Sounder (EOS MLS) on the Aura satellite, IEEE Trans. Geosci. Remote
Sens., 44(5), 1075–1092.
Yang, H. J. (1995), 3-dimensional transport of the ertel potential vorticity
and N2O in the GFDL skyhi model, J. Atmos. Sol. Terr. Phys., 52(9),
1513–1528.
Zander, R., M. R. Gunson, C. B. Farmer, C. P. Rinsland, F. W. Irion, and
E. Mahieu (1992), The 1985 chlorine and fluorine inventories in the
stratosphere based on ATMOS observations at 30-degrees north latitude,
J. Atmos. Chem., 15(2), 171–186.

O. Aulov and D. J. Lary, Goddard Earth Sciences and Technology Center,
University of Maryland Baltimore County, Baltimore, MD 21228, USA.
(david.lary@umbc.edu)
D15S04 LARY AND AULOV: HCl INTERCOMPARISON AND HISTORICAL CONTEXT
10 of 10
D15S04