We describe the construction of a continuous 38-year record of stratospheric
aerosol optical properties. The Global Space-based Stratospheric Aerosol
Climatology, or GloSSAC, provided the input data to the construction of the
Climate Model Intercomparison Project stratospheric aerosol forcing data set
(1979–2014) and we have extended it through 2016 following an identical
process. GloSSAC focuses on the Stratospheric Aerosol and Gas Experiment
(SAGE) series of instruments through mid-2005, and on the Optical
Spectrograph and InfraRed Imager System (OSIRIS) and the Cloud-Aerosol Lidar
and Infrared Pathfinder Satellite Observation (CALIPSO) data thereafter. We
also use data from other space instruments and from ground-based, air, and
balloon borne instruments to fill in key gaps in the data set. The end result
is a global and gap-free data set focused on aerosol extinction coefficient
at 525 and 1020 nm and other parameters on an “as available” basis. For
the primary data sets, we developed a new method for filling the
post-Pinatubo eruption data gap for 1991–1993 based on data from the
Cryogenic Limb Array Etalon Spectrometer. In addition, we developed a new
method for populating wintertime high latitudes during the SAGE period
employing a latitude-equivalent latitude conversion process that greatly
improves the depiction of aerosol at high latitudes compared to earlier
similar efforts. We report data in the troposphere only when and where it is
available. This is primarily during the SAGE II period except for the most
enhanced part of the Pinatubo period. It is likely that the upper troposphere
during Pinatubo was greatly enhanced over non-volcanic periods and that
domain remains substantially under-characterized. We note that aerosol levels
during the OSIRIS/CALIPSO period in the lower stratosphere at mid- and high
latitudes is routinely higher than what we observed during the SAGE II
period. While this period had nearly continuous low-level volcanic activity,
it is possible that the enhancement in part reflects deficiencies in the data
set. We also expended substantial effort to quality assess the data set and
the product is by far the best we have produced. GloSSAC version 1.0 is
available in netCDF format at the NASA Atmospheric Data Center at
Space-based sources for stratospheric aerosol extinction coefficient data and their status in GloSSAC. Two non-space instruments whose data are used in GloSSAC (NASA Airborne Lidar and NASA 48-inch lidar) are also shown.
Since the discovery of the stratospheric aerosol layer, there has been a
continuing interest in the role of stratospheric aerosol in chemistry and
climate. Stratospheric aerosol climatologies derived primarily from
space-based observations of their optical properties have been key elements
of the study of the effects of major volcanic events. Often, data sets
covering the years following the 1991 eruption of Mount Pinatubo were
developed based primarily on observations by the Stratospheric Aerosol and
Gas Experiment (SAGE II) A complete list of acronyms is included in
Appendix A
Herein, we report on a global space-based stratospheric aerosol climatology
(GloSSAC) that we developed to support the Coupled Model Intercomparison Project
phase 6 (CMIP6; Morgenstern et al., 2017). GloSSAC is most closely related
to the Assessment of Stratospheric Aerosol Properties (ASAP; SPARC, 2006)
and CMIP phase 5 data sets (for papers related to this data set see Vernier
et al., 2011, Solomon et al., 2011, and Mills et al., 2016) and follows the
same basic paradigm that produce those versions. We build it primarily using
space-based measurements by a number of instruments including the SAGE
series, the Optical Spectrograph and InfraRed Imager System (OSIRIS; Rieger
et al., 2015), the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite
Observation (CALIPSO; Vernier et al., 2011), Cryogenic Limb Array Etalon
Spectrometer (CLAES; Massie et al., 1996), and the Halogen Occultation
Experiment (HALOE; Thomason, 2012). We compile the data set in monthly
depictions for 80
Among the challenges to the creation of GloSSAC and its predecessors is the general inhomogeneity of the data sets. The source/instrument from which data are derived changes sometimes without overlap from earlier instruments. In addition, the various instruments measure in fundamentally different ways including limb occultation, limb scatter, and lidar backscatter. It is both obvious and important to note that none of the measurements form a complete set of observations of stratospheric aerosol from which any desired aerosol parameter can be derived without significant assumptions about aerosol composition and size distribution (Thomason et al., 2008). During periods in which aerosol extinction coefficient values at 525 and 1020 nm are not available, they are empirically derived from available observations rather than based on inferred size distributions or similar approaches. We identify and make an effort to exclude observations in which we infer the presence of polar stratospheric clouds and clouds near the tropopause (which is particularly important in the tropics) in an instrument specific manner. While cloud presence determination is generally robust, some variations in the aerosol climatology may arise due to differences in how effective these processes are from instrument to instrument that may depend on variations in the aerosol loading itself. While continuity in the data set is a key goal for GloSSAC, maintaining it over 35 years is challenging. We urge caution in using this data set for “off label” applications such as attempting to infer long-term changes in stratospheric aerosol background levels.
We do not make active use of every potential source of space-based aerosol observations in GloSSAC and we select instruments via a straightforward set of criteria. The CMIP6 stratospheric aerosol data set was finalized in early 2015 and GloSSAC v1.0 is simply an extension of that compilation. Therefore, we have avoided any changes in data sources and process for this release. In general, instruments with long records (many years) are preferred over those with short lifetimes, as are those that have a large latitude domain. Data must have been publicly available during the creation of the CMIP6 data set in late 2014. As a result, we excluded SCIAMACHY, which has since met this criterion (von Savigny et al., 2015). We will consider this data set for use in future versions of GloSSAC. In addition, the data must have a peer-reviewed validation paper for stratospheric aerosol products and this requirement currently excludes OMPS (Gorkavyi et al., 2013), MAESTRO (Kar et al., 2007), and SOFIE (Hervig et al., 2017). We also excluded data sets that do not fill a unique function in the data set particularly due to lifetime or spatial coverage (some of which also present additional use challenges). These include SAGE III/Meteor 3M (Thomason et al., 2007), POAM III (Randall et al., 2001), ACE Imager (Vanhellemont et al., 2008), ILAS I/II (Burton et al., 1999), ISAMS (Lambert et al., 1996), HIRDLS (Massie et al., 2010), and GOMOS (Vanhellemont et al., 2016; Robert et al., 2016). Generally, we have chosen to minimize the number of instruments to simplify the already complex problem of making a homogeneous composite data set and the value we place on some data sets is influenced by timeliness. For instance, it is likely that we would not use data from CLAES, whose lifetime was only 2.5 years, if its mission had taken place in the quiescent late 1990s instead of the crucial 1991 to 1993 period. Figure 1 summarizes significant space-based stratospheric aerosol observations and their status within GloSSAC.
All instruments whose data are used in various GloSSAC eras as a function of latitude.
In the following, we describe the basic construction of GloSSAC highlighting changes relative to previous versions. First, we describe the data set's construction in three primary periods. The core period consists almost exclusively of data from SAGE II (1984 to 2005) while an earlier period (the pre-SAGE II period) spans 1979 into 1984 rests upon SAGE I (1979 to 1981) and a diverse collection of ground and airborne observations. A third period consists of observations from OSIRIS and CALIPSO and spans from the end of the SAGE II mission in 2005 through 2016. Following the description of GloSSAC construction for these periods, we describe the filling processes that produce a gap-free data set for 1979 through 2016. This includes a basic interpolation process that is mostly relevant to the two SAGE periods, a new process for estimating grid values in high-latitude winter (SAGE periods), the production of the data set in the “SAGE-gap” period from late 1981 to late 1984, and gaps in the SAGE II data set between the Pinatubo eruption and mid-1993. The later gaps are due to the extreme opacity of the stratosphere following that event. Instrument use in time and latitude is shown in Table 1. We discuss the process for inferring aerosol extinction at 525 and 1020 nm from CLAES, HALOE, OSIRIS, and CALIPSO. We will describe the extensive effort to quality check the data set to remove data artifacts and the known limitations to the data set. Finally, we discuss the contents of the data set as archived and future plans.
Distribution of SAGE II observations throughout its lifetime with one dot per event. From October 1985 to July 2000, there are about 10 000 events per year. Due to an instrument fault, there are no events from August to October 2000 and the instrument operates on a half duty cycle (a mix of sunrise and sunset events) until the end of its mission in August 2005.
The GloSSAC data set is a zonal data set in 5
Steps involved in the creation of the GloSSAC 1020 nm aerosol
extinction coefficient climatology at 21 km:
The initial step in producing GloSSAC is to produce gridded data sets for
SAGE II at its four wavelengths (386, 453, 525, and 1020 nm) and HALOE and
CLAES aerosol measurements at selected wavelengths. We assign each bin a flag
that indicates its source and preserve both the number of data points used
and the number identified as containing cloud. We show the complete set of
flag values in Appendix B. The data are reported on a 0.5 km vertical grid
from 5.0 to 39.5 km. This is the native SAGE II reporting resolution though
its true vertical resolution is
For a given latitude/month bin, we collect all aerosol extinction coefficient
profiles within 5
The processes that terminate SAGE II profiles control the lower extent of
data and these vary among the four measurement wavelengths. Individual
profiles are terminated by either high molecular extinction (at shorter
wavelengths), optically dense clouds (all wavelengths), encountering the
solid Earth (usually just for 1020 nm extinction profiles), and, during
Pinatubo, very high aerosol extinction levels (all). We also exclude any
observations in which we infer the presence of non-opaque clouds. We identify
these clouds using the method described by Thomason and Vernier (2013) (a
revision of an algorithm developed by Kent et al., 2003) and exclude those
points from the analysis. We infer cloud presence almost exclusively in the
troposphere; however, we occasionally infer the presence of clouds in the
lower tropical stratosphere. In addition, we are able to detect and exclude
ice polar stratospheric clouds (PSCs) but it is likely that saturated ternary
solution (STS) and nitric acid trihydrate (NAT) PSCs slip by the cloud
identification process. This occurs because the methodology relies on the
dominating presence of “large” aerosol particles that are mostly lacking
for these types of PSCs. Away from Pinatubo, 525 and 1020 nm extinction
coefficients are available throughout the stratosphere. Profiles at 453 and
386 nm are available down to about 12 and 16 km, respectively. It should be
noted that the aerosol data at 386 nm are biased low below 20 km and above
the main aerosol layer and are at best of limited quality under all
conditions and altitudes. Following Thomason et al. (2010), while the data
are included in the data set, we recommend caution using SAGE II 386 nm
data. Finally, we exclude any data below the highest altitude at which
1020 nm aerosol extinction coefficient exceeds 0.01 km
Both CLAES and HALOE flew aboard the Upper Atmosphere Research Satellite
(UARS) and all UARS data are reported at pressure levels rather than altitude
like SAGE II. Median-based
extinction profiles on the native pressure grid are derived following rules
similar to those used with SAGE II (profiles are terminated at the bottom if
less than 5 data points are available or less than 50 % of the available
profiles in that time/latitude are available at that altitude). We
interpolate the profiles to the standard altitude grid using altitude-log
pressure from the Modern-Era Retrospective Analysis for Research and
Applications (MERRA; Rienecker et al., 2011) data that is used in SAGE II
data processing. CLAES (October 1991 to April 1993) aerosol extinction data
are used at 1257 cm
We use HALOE (October 1991 to 2005) data at 3.40
The distribution of CLAES 780 to 1257 cm
One of the goals of GloSSAC is to have a continuous “gap-free” data set for 1979 through 2016 at both 525 and 1020 nm. The former is comparable to other long-term data sets like the GISS stratospheric aerosol optical depth record, while the latter is the most robust aerosol measurement available from the SAM/SAGE series and available wavelength for most of the “SAGE” era from 1979 to 2005. However, there are important gaps in the lower stratosphere from the eruption of Pinatubo in June 1991 well into 1993. In addition, a number of SAGE II profiles are compromised by short duration events. These events mostly occur in 1993 and are primarily sunrise events where measurement-taking was terminated before sufficient exoatmospheric data were taken for robust normalization to transmission. The events were temporarily shortened during a period in which the spacecraft batteries were rapidly degrading. Event durations were returned to their normal length in early 1994. As a result, complete 525 and 1020 nm records require the use of non-SAGE II data during this period.
All HALOE NO
Following the general GloSSAC data use paradigm, while other data sets are
available, CLAES and HALOE are shown below to be reasonably well-behaved and
sufficient for the filling process. CLAES and HALOE data offer similar
near-global coverage through most of this period (October 1991 and onwards)
if the data can be transferred from the measurement wavelengths to SAGE II
wavelengths in a robust manner. Figures 6 and 7 show the observed
relationship between SAGE II extinction coefficient measurements and CLAES at
1257 cm
Relationship between the gridded and cloud-cleared CLAES
1257 cm
Relationship between corrected HALOE 3.40
Observed relationship between gridded and cloud-cleared SAGE II observations (below 30 km) as a function of SAGE II 1020 nm observations. The blue line is the conversion relationship used for 1020–525 aerosol extinction conversion throughout GloSSAC development.
The summer of 1991 presents special problems for the reconstruction while at
the same time being a crucial period for the evaluation of the performance of
chemistry-climate models. SAGE II data are missing at
altitudes as high as 25 km after the eruption and UARS data are only
available starting in October. An additional issue for this period is that
there are no SAGE II observations (and no truly tropical data at all) in
June 1991. In previous versions, SAGE II data were interpolated between May
and July 1991 producing values with no observational basis in this month. For
GloSSAC, we have replicated the missing data between 20
Component contributors to GloSSAC or previous versions all shown in
aerosol extinction coefficient at 1020 nm. They include:
Aerosol backscatter ratio from the July 1991 airborne lidar mission aboard the NASA DC-8. Contours are at 1.2 (dark blue), 1.4, 1.7, 2, 4, 8, 16, 32, 64 (orange) with a highest observed value at 82 (yellow). Above 17 km, all of the contours are associated with relatively fresh Pinatubo aerosol.
For July to September 1991, we make use of the tropical reconstruction
created for the ASAP analysis, which is a combination of data from the lidar
station operated by the Centro Meteorológico de Camagüey in Cuba
(23
With the addition of CLAES observations, midlatitudes no longer need patching
by non-space-based data sources as in previous versions since there is little
or no loss of data in mid- and high latitudes between the eruption of
Pinatubo and the start of the CLAES mission. In previous versions, the
primary method to fill missing data in the mid- and high-latitude lower
stratosphere between June 1991 and mid-1993 were data from the NASA Langley
48-inch lidar facility (Osborn et al., 1995) and data from the University of
Wyoming backscatter sonde (Rosen and Kjome, 1991; Rosen et al., 1997) deployed from the
NIWA Lauder (New Zealand) facility. We show these data sets in Fig. 9b and c.
Recently, we have recovered and archived data from NASA Langley airborne
missions in July 1991 and May 1992 at the NASA Atmospheric Sciences Data
Center,
At this point, there are still substantial gaps throughout the data set, mostly because of the spatial sampling pattern of a mid-inclination solar occultation instrument. Gaps are filled using linear interpolation in time but not in altitude or latitude. While we could interpolate and completely fill the grid, in practice, interpolation is limited to gaps of no more than 2 consecutive months. This works well in mid- and low latitudes except in late 2000 where SAGE II was off for several months due to an instrument error. In this case alone, interpolation is permitted to 4 months since it is a relatively benign period and there are few data available to provide alternative guidance. We do not believe that this seriously compromises the analysis. The most significant issue in this period is a poor depiction of the Antarctic polar vortex in austral spring, where it is effectively missing entirely. With the allowable degree of interpolation, the GloSSAC 1020 nm grid at 21 km is now filled except at high latitudes in winter as shown in Fig. 3c.
At high latitudes, the 2-month requirement leaves substantial gaps in the
winter hemisphere at latitudes as low as 60 Equivalent latitude is tied to all SAGE II events using MERRA data and
available to all users of that data set (Manney et al.,
2007).
GloSSAC analysis prior to using the equivalent latitude filling
process
Distribution of equivalent latitude (per degree) for August at
70
During the SAGE lifetime (January 1979 to November 1981), the 1000 nm
aerosol extinction coefficient measurements form the basis of the overall
analysis. We did not use the SAGE 450 nm measurements in this analysis since
they are poor quality and not usable at all below 20 km (Thomason et al.,
1997a). The SAGE data are supplemented by 1000 nm extinction measurements by
the Stratospheric Aerosol Measurement (SAM II; 1978–1993), which provide
data only at high latitudes (
The SAGE “gap” period from December 1981 to September 1984 is of critical
interest since it encompasses the El Chichón eruption (March/April 1982).
However, with very limited space-based measurements available There
is the potential for very valuable aerosol data for the El Chichón period
from the Solar Mesospheric Explorer (October 1981 to April 1989; e.g.,
Eparvier et al., 1994). However, the current (non-released) aerosol product
has a significant seasonal/latitudinal bias due to issues related to a very
difficult accommodation for viewing geometry. Perhaps future efforts will
yield a useful product from this instrument.
In the Northern Hemisphere, the 1000 nm extinction record is filled with
SAM II (shown in Fig. 9d and e) between 80 and 65
At low latitudes and southern midlatitudes, virtually no data are available
except from airborne lidar missions conducted by NASA between 1982 and 1984.
Five airborne lidar missions were flown in July 1982 (13 to 40
After the end of the SAGE II mission in August 2005, the stratospheric aerosol extinction coefficient climatology becomes solely dependent on aerosol measured by OSIRIS and CALIPSO. This represents not only a change in instrument but also the way in which aerosol is measured. OSIRIS measures limb scatter radiance from which aerosol extinction coefficients at 750 nm (and other parameters) are inferred. CALIOP is the CALIPSO platform's nadir-viewing lidar that produces a stratospheric backscatter aerosol coefficient product primarily at 532 nm. While these changes represent some challenges to the continuity of the overall climatology, they both produce near-global coverage on a daily basis. Though we have not exploited the potential for higher temporal resolution for GloSSAC v1.0, we are considering how to exploit the higher temporal resolution data for future versions.
For GloSSAC, we used the OSIRIS aerosol extinction climatology as produced in
(Rieger et al., 2015). This climatology provides monthly latitude- and
altitude-resolved extinction converted to 525 nm and bias corrected to SAGE II.
Although this climatology removes much of the bias between the two
instruments, the methods used in Rieger et al. (2015) are slightly
different than those used to create the SAGE II climatologies in this
paper. For instance, the latitude bins are 5
The primary issues associated with the use of backscatter data from CALIPSO
are measurement calibration and noise. The noise can be reduced by averaging
millions of profiles to obtain zonally averaged data in the stratosphere on
a monthly basis. Rogers et al. (2011) showed that aircraft high spectral
resolution lidar measurements and CALIOP data agreed within
2.7 %
We initially calculate mean total attenuated backscatter at 532 nm every
1
We are fortunate to have roughly 4 years of overlap in the data from OSIRIS and SAGE II. This period is critical for understanding not only how OSIRIS and CALIPSO interrelate but also to use OSIRIS to infer indirectly how SAGE II relates to CALIPSO. Since clouds in the upper troposphere may have a deleterious impact on the measurement of aerosol extinction in the lower stratosphere (characteristic of limb measurements in general), we exclude all OSIRIS data in the lowest 2 km of the stratosphere. For the overlap period, we show, in Fig. 13a, the relationship between OSIRIS inferred 525 nm aerosol extinction coefficient and SAGE II measurements at that wavelength. Overall, the comparison is favorable; SAGE II and OSIRIS are well correlated with OSIRIS tending to be 10 to 20 % less than SAGE II (median 0.88) in a period that has the lowest aerosol loading observed between 1979 and 2016. If we use OSIRIS “as is” or scaled by the median ratio value between OSIRIS and SAGE II data sets, we observe a discontinuity at the August 2005 (SAGE II) and September 2005 (OSIRIS) boundaries. While in retrospect it may not have been the most satisfactory solution, we scaled OSIRIS to minimize an obvious discontinuity using a factor of 0.8. The switch from SAGE II to OSIRIS occurs a few months following the eruption of Manam (January 2005) that effectively signaled the end of the volcanically quiescent period that began in the late 1990s. The degree to which this event creates the discontinuity is not clear and further work on melding the SAGE II and OSIRIS records is necessary. Since OSIRIS is the only source of space-based observations between September 2005 and April 2006 we use it alone through this period. Some interpolation at mid- and high latitudes is required and we follow the interpolation method used for SAGE II observations to fill these gaps. In addition, the 2 km exclusion in the lower stratosphere leaves gaps that are only partly filled by temporal interpolation. Where bins remain unfilled, the lowest measured value in a latitude/time column is replicated down to the tropopause. This is rare and rarely for more than 1 or 2 altitude bins. As with the other periods, data in the OSIRIS-only period are flagged to indicate how we derived the value at each grid point.
It is particularly disappointing that SAGE II and CALIPSO data sets do not
overlap. CALIPSO observations are always at the lower end of aerosol loading
observed during the SAGE II lifetime. Fortunately, the OSIRIS/SAGE II overlap
period is also primarily at low aerosol loading and permit the use of OSIRIS
as a transfer medium for understanding the CALIPSO backscatter to SAGE-II-like
extinction coefficient conversion. We show the ratio of CALIPSO 532 nm
backscatter coefficient to scaled OSIRIS 525 nm extinction coefficient as a
function of OSIRIS extinction in Fig. 13b. Nominally, we might expect some
dependence on the ratio to extinction value due to a correlation between
extinction magnitude and aerosol size. In fact, we do see a tail towards
lower extinction-to-backscatter ratio with lower extinction values but the
vast bulk of the data exists in an amorphous blob and the confidence in the
observed relationship is low. Part of the lack of confidence is due to the
relatively high noise exhibited by the CALIPSO data relative to the other
instruments and the potential for bias associated with the normalization
process used for all lidar instruments. As a result, we use the median value
of this distribution (53 sr) as the sole extinction
to backscatter ratio conversion factor. This value is well within expected
values for extinction-to-backscatter ratio (roughly between 30 and
60 sr) and effectively maps CALIPSO observations to
OSIRIS. If a conversion suggested by distribution shown in Fig. 13b were
used, large extinction coefficients would tend to increase while smaller
extinction coefficient values (
Following April 2006, CALIPSO and OSIRIS are both available to the end to the record and beyond. Since we only use nighttime data from CALIPSO, and OSIRIS only acquires data in daytime, the data sets span the entire range of latitudes during all seasons, whereas one or the other would have high latitude gaps similar to those of SAGE II. Since we have forced considerable consistency into the OSIRIS and CALIPSO 525 nm extinction data sets, we mix these sets such that where both exist, we report the average of the two. When only one exists, we report that value. Overall, we do not observe discontinuities or other issues in this mixing process and the overall data set is pleasing. In Fig. 3f–h, we show the entire data set with OSIRIS only (panel f), CALIPSO only (panel g), and the two combined (panel h). With both data sets, the need for interpolation is mostly limited to only winters where the PSC clearing process for CALIPSO leaves some holes in the data set that we interpolate through as in other periods. While an argument can be made for whether STS is in fact simply a special case of aerosol which should be retained, at this time, GloSSAC attempts to remove all PSC effects as well as possible. Extinction at 1020 nm is estimated using the relationship shown in Fig. 8 (in reverse to its previous application). With these additions, the GloSSAC data set is complete from 1979 through 2016 (Fig. 3h).
Final GloSSAC distribution for 525 nm extinction at 21 km using the same contouring intervals and coloring as in Fig. 2.
While the OSIRIS/CALIPSO segment of the data set is generally in good shape, we make two observations that users should consider. One is that, unlike the SAGE-only versions of this data set, the conversion of OSIRIS and CALIPSO data are strongly tied to 525 nm rather than 1020 nm (SAGE II's most robust channel). As this part of the data set is effectively a single channel data set, users should primarily make use of 525 nm data (shown in Fig. 14) after the end of the SAGE II mission in August 2005. This is critical because the post-SAGE II period is dominated by a series of small eruptions whereas the SAGE II record is dominated by the recovery from large eruptions by El Chichón and Ruiz/Nyamuragira (late 1985 and early 1986, respectively), and Pinatubo. The SAGE-based conversion between 1020 and 525 nm extinction coefficients (and vice versa) is dominated by large volcanic events. These characteristically correlate aerosol size and extinction magnitude such that large extinctions exhibit a 525 to 1020 nm extinction ratio as low as 1.0 (indicating extinction dominated by large particle sizes) and low extinctions show a ratio from 3 to 6 in the main aerosol layer (indicating extinction dominated by smaller aerosol). Even in the SAGE II record, we observe exceptions to this scenario following small eruptions by Kelut (1990), Ruang (2002), and Manam (2005). Figure 15 shows the 525 to 1020 nm extinction ratios from the tropics between 2000 and 2016. Prior to September 2005, the plot is based primarily on SAGE II measurements and we can see the impacts of the Ruang and Manam eruptions increasing the extinction ratio while extinction itself was also increased. This suggests that these eruptions effectively reduced the dominating particle size possibly by introducing new small aerosol that do not coagulate quickly. After August 2005, the plot is based on aerosol extinction at 525 nm inferred from OSIRIS/CALIPSO and the empirical relationship shown in Fig. 8. Between August and September 2005, there is a discontinuity in the extinction ratio indicating that the climatological conversion process does not capture the Manam event well. With a number of small volcanic events scattered throughout the OSIRIS/CALIPSO period, we believe it is likely that this disconnect with the SAGE II part of the record is a regular feature after 2005 and use of the 1020 nm data should be avoided. In future versions, we may be able to leverage some sizing information from a second OSIRIS channel, the CALIPSO/OSIRIS pairing, or by contributions from other instruments like SCIAMACHY to manage this issue in a more robust manner.
Ratio of GloSSAC 525 to 1020 nm extinction
coefficient at 2.5
The second issue we observe in the OSIRIS/CALIPSO period is that aerosol extinction is higher in the lower stratosphere (below 20 km) in mid- and high latitudes of both hemispheres than typically observed in the similar SAGE II period leading up to that segment. It appears to be associated with data from both OSIRIS and CALIPSO and may simply be the outcome of regular volcanic events throughout this period. The primary sink for aerosol is through polar latitudes and enhancements in extinction are even expected following low latitude eruptions. However, the elevated levels appear to persist into less active periods and are manifested fairly equally in both hemispheres, while volcanic activity occurred mostly in the Northern Hemisphere. It is possible that the GloSSAC depiction is correct; however, an unexpected disconnect between SAGE II and OSIRIS/CALIPSO data is of concern and users should be aware of some issues in this time and region. For CALIPSO backscatter data, it is possible that improving the backscatter coefficient to extinction coefficient conversion may reduce the apparent discrepancy. In addition, with the beginning of SAGE III's mission aboard the ISS in 2017, we hope to use those new data to understand this issue. We also plan to examine SCIAMACHY and/or OMPS as contributors to this issue as well as to GloSSAC in general.
As a data product intended for use by the climate modeling community, it is critical to deal with as many issues in the data set as possible and not leave those issues for the users to discover on their own. While the data used in GloSSAC are generally robust, it is still common for occasional bad individual values or entire profiles to occur and have a deleterious impact on the data set if accepted as truth. As a result, we have implemented a quality assurance process to identify and remove low quality data from GloSSAC. Although we considered a number of automated schemes to identify “bad” data, the most effective means was a month-by-month visual examination of the data. In this case, we identify bad data points/profiles using our best scientific judgment and remove them from the data set. We only remove data when the impact is obvious and we apply it only to the final 1020 and 525 nm data products. While issues typically appear in both wavelengths, they occasionally occur at only one wavelength and we deal with these individually. The extinction products consist of a little more than 1 million individual values, and in quality assurance we identify less than 5000 bad data point or less than 0.5 % of all data values (roughly the equivalent of 2 months in 38 years). In the SAGE II period, these data points tend to occur at high latitude where we have noted (albeit rarely) data quality issues in the past. Once the bad data points are removed, we interpolate the data across any new gaps using the same approach used in other processes. Data replaced in this manner are flagged.
We have created a SAGE-II-based monthly climatology for altitudes above 30 km to replace OSIRIS and CALIPSO data. In general, neither of these data sets is consistent with SAGE II above that altitude (where extinction is very low), whereas SAGE II is generally robust to higher altitudes. In this climatology, we average all SAGE II data for each month except the years 1991 to 1994, where Pinatubo effects were obvious above 30 km. Any OSIRIS or CALIPSO data above 30 km at 525 and 1020 nm is replaced with the climatology and flagged.
A nominal stratospheric background is included as a part of the GloSSAC data set. It consists of the average of 1999, 2000, 2001, 2003, and 2004; we excluded 2002 because of the eruption of Ruang in September of that year. The year 2000 is the lowest aerosol extinction in the entire record and it could be used as a background level. However, there is a notable effect of the quasi-biennial oscillation (QBO) on aerosol extinction above 20 km (Thomason et al., 1997a) and using 2000 as the background year in a repeating series has discontinuities as large as a factor of 2 every January. The 5-year average, while generally slightly larger than 2000 levels, effectively removes most if not all QBO-related discontinuities.
The focus of the GloSSAC data set is aerosol optical measurements; however, there is substantial interest in aerosol properties that are derivable from these properties, particularly aerosol surface area density and effective radius. As a result, we have included values for these parameters where SAGE II measurements are available. They are derived using the method described in Thomason et al. (2008) and consistent with the same properties included in the native SAGE II version 7 data set and designed to bracket the potential range in these parameters. These data are included for informational purposes and they should not be interpreted as canonical estimates for them.
Measurement uncertainties are included in the data set only for the SAGE II portion of the data sets. In this regard for each latitude/time/altitude bin, we include the standard deviation of the measurements used (a combination of geophysical variability and measurement noise) and the median reported measurement uncertainty. Generally, SAGE II aerosol extinction coefficients in the lower stratosphere have uncertainties of less than 10 % and, during Pinatubo, often less than 5 %. The same uncertainty parameters for all space-based measurements, including new potential data sources SCIAMACHY and SAGE III/ISS, will be included in the next GloSSAC release. Beyond measurement uncertainty, the conversion of one measured quantity to another adds the potential for significant and variable bias to GloSSAC values at 525 and 1020 nm. The process we used to scale data from OSIRIS, CLAES, HALOE, and CALIPSO to the long-term 525 and 1020 nm data sets nominally eliminate bias from between the data sets and the spread of measurements is more or less the combined measurements uncertainty combined with geophysical variability. However, this is only the case where the data sets overlap. For CLAES/HALOE, their use in GloSSAC coincides with the massively volcanic Pinatubo period and their sole use is for altitudes/latitudes where SAGE II data does not exist. For CALIPSO/OSIRIS, their use is solely for the period after the SAGE II mission ends. A complicating factor for this period is that the overlap period between SAGE II and OSIRIS was volcanically quiescent while much of the period after 2005 consists of a steady drumbeat of small but significant volcanic events. There are few independent robust measurements for either the Pinatubo period (particularly in the tropics) or the post-SAGE II period. As a result, it is difficult to evaluate how well the conversion processes work and there is the potential for bias. We hope the new SAGE III mission in this mildly volcanic period will give insights into the potential for bias with both OSIRIS and CALIPSO as well as suggest mechanisms for migrating any issues. This is a focus for future developments in GloSSAC.
Total GloSSAC stratospheric aerosol optical depth at
525
Stratospheric optical depth at 525 and 1020 nm integrated from a base at the
tropopause upwards can be easily computed using components of the GloSSAC
data set. These are shown in Fig. 16. The GloSSAC minimum 525 nm optical
depth in the tropics (0.0028) occurs in May 2001 as an extended period of
very limited volcanic influence was terminated by the eruption of Ruang
(Indonesia) in October 2002 and subsequent eruptions. The peak optical depth
at 525 nm is 0.22 and occurs in the tropics several months after Pinatubo in
November 1991. Although the delay is not an obvious outcome, several factors
contribute to this feature. Given that the primary injection altitude was
well above 20 km, there would be little loss of aerosol from the
stratosphere in the first months following the eruption. Also, since a
significant fraction of what would become sulfate aerosol entered the
stratosphere as SO
Comparison of total stratospheric aerosol optical depth at northern midlatitudes, the tropics, and southern midlatitudes at 525 (SAGE II/GISS) and 550 nm (AVHRR). AVHRR is shown in black, AVHRR with a 28-year median annual cycle removed is red, GloSSAC is shown in blue, and the GISS data set is shown in green.
Figure 17 shows a comparison of GloSSAC 525 nm optical depth, Version 2.0 of the Advanced Very High Resolution Radiometer (AVHRR) total atmospheric aerosol optical depth (Zhao, 2013; Zhao and Chan, 2014), and the GISS stratospheric aerosol 500 nm optical depth (Sato et al., 1993). AVHRR provides a measurement of total atmospheric aerosol optical depth at 500 nm. Aerosol optical depth is usually dominated by tropospheric aerosol and variability in the stratosphere is not apparent. This is not the case following large volcanic events where the volcanic perturbation can be larger than the tropospheric component. For comparison purposes, we remove the 28-year median annual cycle from the long-term AVHRR record to highlight the impact of the Pinatubo and El Chichón eruptions. In addition, we plot AVHRR data only during the first years after both eruptions as AVHRR is unable to infer stratospheric effects once the stratospheric optical depth is much less than about 0.02. Figure 17 shows the AVHRR total and “stratospheric” optical depth. While there is reasonable agreement between the AVHRR and GloSSAC data products in midlatitudes, the tropical optical depths show a substantial difference for both eruptions. For Pinatubo, it suggests a tropical total optical depth in excess of 0.4 (at 500 nm), which is substantially larger than the corresponding value of 0.22 in the GloSSAC stratospheric optical depth. Some of the difference could be due to loading in the upper troposphere that is not a part of the GloSSAC stratospheric optical depth (integrated from the tropopause upward) but that AVHRR includes. It is also likely that setting a baseline is partly responsible for this issue. On the one hand, if only the few years prior to the Pinatubo eruption are used, the peak optical depth from AVHRR decreases to about 0.3. On the other hand, the optical depth after early 1993 is less than and becomes much less than the background values from the pre-Pinatubo period. At least in the tropics, it is also clear there are some discontinuities in the optical record that appear to be unrelated to geophysical phenomena. In any case, the AVHRR peak Pinatubo optical depth is between 50 and 100 % larger than that from GloSSAC.
In order for the AVHRR/GloSSAC difference to be due purely to stratospheric aerosol, the mostly likely GloSSAC-related culprit would be the conversion of CLAES infrared observations to SAGE II wavelengths. The correction from CLAES to SAGE would have to be in error by a factor of about 2. The correlation between SAGE II and CLAES observations (Fig. 6) is well behaved and provide little suggestion that an error on that scale is possible. Sun photometer measurements from sites in American Samoa, Mauna Loa, and other sites (Dutton et al., 1994; Stone et al., 1993; Russell et al., 1996; Dutton and Christy, 1992) suggest a peak mid-visible optical depth between 0.2 and 0.25 and perhaps as large as 0.3. The GloSSAC value is on the low end of these values but the Sun photometer measurements will also include volcanic aerosol in the troposphere. As a result, we believe that GloSSAC stratospheric optical depths for the Pinatubo period are reasonable. The GISS data set after 1979 is based on the data from the same instruments used in GloSSAC and a good level of agreement would be expected. In general, that is observed until at least 1998 (Fig. 17). There are some minor differences most likely related to updates in SAGE data products, changes in cloud clearing, and the filling process. After 1998, however, the GISS optical depth is uniformly about a factor of 2 less than GloSSAC values with an almost immediate transition from reasonable to poor agreement. The large differences between these data sets after 1998, particularly up to the end of the SAGE II period in 2005, are difficult to understand and the GISS values appear to be in error. Overall, we do not recommend the use of AVHRR or GISS for validating CCM estimates for stratospheric column optical depth. On the other hand, users of GloSSAC should be aware that there is almost certainly substantial aerosol in the upper troposphere particularly in the tropics during the several months if not a few years following the Pinatubo eruption. That material is not a part of the stratospheric GloSSAC data set yet may have significant climate influence.
GloSSAC version 1.0 is available in netCDF format at the
NASA Atmospheric Data Center at
Despite some limitations, we believe that this is by far the best data set in this series of data sets (ASAP, CCMI). Compared to previous releases of the data set such as ASAP or the set for CCMI in 2014, we have implement a number of major improvements. These include the handling of the Pinatubo SAGE II saturation period in 1991 to 1993, the way in which missing values at high latitudes are filled during the entire SAGE II period, and how the post-SAGE II period is constructed using OSIRIS and CALIPSO. The data set is focused on providing as close to measured aerosol optical properties as possible. Recognizing the complexity of mixing data from many sources, unmodified source data are preserved in the data set at the GloSSAC resolution.
For users, we recommend the following practices for this data set:
For validation of aerosol properties derived within a chemistry-climate
model, we suggest that the most robust comparisons are with the measurements
directly. As a result, we suggest that they use the data flags to identify
these values in the data set and compare model-derived parameters with those
identified as measured, as opposed to indirectly inferred values. We have not focused on the derivation of bulk aerosol properties within this
data set though it is suitable for that process. Even though values are
reported at 525 and 1020 nm for every grid box, it is critical to recognize
when data are based on a single measurement wavelength. This includes
everything outside the SAGE II period and some data gap periods within the
SAGE II period associated with Pinatubo. Users who wish to use this data set
for developing climatologies of aerosol properties are welcome to do so as
well as distribute any products derived from your effort. We would appreciate
attribution of the source material.
The summary of key issues associated with the data set are the following:
The summer of 1991 in the tropics is poorly resolved due to the loss of
SAGE II in the lower stratosphere and because CLAES data do not become
available until October of that year. In any case, the highly inhomogeneous
state of the stratosphere in the several months following the Pinatubo
eruption makes a monthly depiction of questionable validity. The OSIRIS/CALIPSO period presents two issues. There is clearly an issue
with converting measurements from 525 to 1020 nm and the later data should
be used very cautiously. This is a one-wavelength period where only 525 nm
values should be used. Also, there are high levels of aerosol extinction in
the lower stratosphere throughout this segment of the data set. While we
cannot exclude that it is correct, users should exercise caution with these
data. Data in the troposphere is only reported during the SAGE II period and
only away from the Pinatubo eruption. It is likely that there is considerable
aerosol in the upper troposphere during this period but we have little
ability to produce values based on measurements in this period. While
tropospheric aerosol is not the general area of concern for GloSSAC, it is
likely that volcanic aerosol in the upper tropical troposphere plays a role
in changing climate during the aftermath of the Pinatubo eruption.
We plan to release new versions in about a yearly cycle. Extensions of the data set using the current processing paradigm will be indicated by minor version number changes (ie., 1.0 to 1.1). If new data sources or significant processing changes occur, the version will change the major number (i.e., 1.0 to 2.0). Current plans are to release version 2.0 in 2018 with the addition of at least SAGE III/ISS data at the end of the record. We will also look at other newer data sets particularly the available SCIAMACHY data set but also aerosol products from OMPS and AerGOM. We may look into deriving data at a higher temporal resolution to more fully utilize the data afforded by OSIRIS and CALIPSO. For the SAGE period, we may examine the approach for deriving ozone variability described in Damadeo et al. (2014). Feedback from users will also be useful in updates to the data set.
In the past, this data product was mostly an “in-house” intermediate product not readily available to the science community. This new approach, and this paper, is an effort to make it more transparent and accessible to all potential users.
GloSSAC data flag values and meaning.
The authors declare that they have no conflict of interest.
This article is part of the special issue “Chemistry–Climate Modeling Initiative (CCMI) (ACP/AMT/ESSD/GMD inter-journal SI)”. It is not associated with a conference.
The authors would like to thank Andrea Gettleman, editor, and Matthew Toohey and Chaochao Gao, reviewers, for suggestions and comments that improved the content of this paper. Edited by: Andrew Gettelman Reviewed by: Matthew Toohey and Chaochao Gao