Here we describe an updated parameterization for prescribing stratospheric aerosol in the National Center for Atmospheric Research (NCAR) Community Earth System Model (CESM1). The need for a new parameterization is motivated by the poor response of the CESM1 (formerly referred to as the Community Climate System Model, version 4, CCSM4) simulations contributed to the Coupled Model Intercomparison Project 5 (CMIP5) to colossal volcanic perturbations to the stratospheric aerosol layer (such as the 1991 Pinatubo eruption or the 1883 Krakatau eruption) in comparison to observations. In particular, the scheme used in the CMIP5 simulations by CESM1 simulated a global mean surface temperature decrease that was inconsistent with the GISS Surface Temperature Analysis (GISTEMP), NOAA's National Climatic Data Center, and the Hadley Centre of the UK Met Office (HADCRUT4). The new parameterization takes advantage of recent improvements in historical stratospheric aerosol databases to allow for variations in both the mass loading and size of the prescribed aerosol. An ensemble of simulations utilizing the old and new schemes shows CESM1's improved response to the 1991 Pinatubo eruption. Most significantly, the new scheme more accurately simulates the temperature response of the stratosphere due to local aerosol heating. Results also indicate that the new scheme decreases the global mean temperature response to the 1991 Pinatubo eruption by half of the observed temperature change, and modelled climate variability precludes statements as to the significance of this change.
Volcanic perturbations to the stratospheric aerosol layer are an often ill-represented forcing in the climate model simulations (Solomon et al., 2011; Driscoll et al., 2012; Knutson et al., 2013; Zanchettin et al., 2015; Kremser et al., 2016). Earth's climate system has been perturbed by several colossal (volcanic explosivity index (VEI) of 5 or greater) volcanic eruptions since 1960 (see Fig. 1) (Newhall and Self, 1982). The impact each of these eruptions has had on the global mean surface temperature anomaly is shown in Fig. 1.
Figure 1 compares the Coupled Model Inter-comparison Project 5 (CMIP5) multi-model global mean surface temperature anomaly to three different observationally based data sets (Taylor et al., 2012). The vertical dashed grey lines note the date of colossal volcanic perturbations accounted for in most of the forcing files utilized in the CMIP5 simulations. Figure 1 shows the response of the National Center for Atmospheric Research's (NCAR) Community Climate System Model, version 4 (CCSM4) (now referred to as the NCAR Community Earth System Model, CESM1), highlighted in red, to volcanic forcing, as well as the response of most other models submitted to CMIP5.
Global annual mean surface temperature anomalies from 1950 to 2013 referenced to the mean taken from 1961 to 1990. Light grey lines represent the 108 model members that contributed to the RCP4.5 scenario of CMIP5. The black line represents the multi-model ensemble mean. The members contributed by NCAR's CCSM4/CESM1 are highlighted in red. Three observationally based data sets have been included for comparison: the GISS Surface Temperature Analysis (GISTEMP) in purple (Hansen et al., 2010; GISTEMP Team, 2015), NOAA's National Climatic Data Center's global surface temperature anomalies in teal (Jones et al., 1999; Smith et al., 2008), and global anomalies from the Hadley Centre of the UK Met Office (HADCRUT4) in blue (Morice et al., 2012).
Stratospheric aerosol is prescribed in several ways with various levels of complexity in global climate models. Most models contributing to CMIP5, including CCSM4/CESM1 (Meehl et al., 2012), prescribed a zonal mean, monthly mean stratospheric aerosol mass or stratospheric aerosol optical depth (SAOD) as well as the surface area density (SAD) of the aerosol (using data sets such as Sato et al., 1993, Stenchikov et al., 1998, or Ammann et al., 2003). Typically this specification of aerosol is ingested within the model's (1) radiative transfer parameterization and (2) chemistry parameterization using several underlying assumptions about the size distribution and composition of the aerosol. Though adequate, these methods leave much to be desired for accurately simulating the evolution of the perturbation to the stratospheric aerosol layer after these eruptions.
To address the need for a more accurate representation of colossal volcanic eruptions in the atmospheric current climate models, including all the configurations of the Community Atmosphere Model (CAM) and Whole Atmosphere Community Climate Model (WACCM) within the framework of the CESM1 (Lamarque et al., 2012; Neale et al., 2013; Marsh et al., 2013; Meehl et al., 2013), a new data set was derived to force models participating in the Chemistry Climate Model Initiative (CCMI) (Eyring and Lamarque, 2012; Eyring et al., 2013). Here we describe the implementation of this data set in CESM1 with additional updates in preparation for CCMI Phase 1 simulations (Eyring et al., 2013).
In Sects. 2, 3, and 4 we discuss the original prescriptions of stratospheric aerosol in all configurations of CESM1. Section 5 discusses the new stratospheric aerosol prescription for all of CESM1. Table 1 summarizes the main similarities and differences between the old and new parameterizations described in Sects. 2 through to 5. In Sect. 6 we describe the behaviour of CESM1 and the response of the model to the new representation of the 1991 Pinatubo eruption. In Sect. 7 we summarize our work and make suggestions for future use of this new stratospheric aerosol scheme in CESM1.
Summary of the old (italic) and new (bold) stratospheric aerosol prescription parameterizations in CESM1 model configurations.
Prior to CESM1, CESM1(CAM4) was part of CCSM4. Neale et al. (2010) describe the scheme used to specify volcanic eruptions and the stratospheric aerosol layer in CCSM4 (specifically within CAM4.0) and how this interacts with the other parameterizations of the model. For a description of the model's climate and its response to forcings, see Meehl et al. (2012). Here we summarize the main features of the volcanic prescription in CCSM4/CESM1(CAM4) that have been changed significantly in the updated scheme described below so that future studies utilizing CESM1 may account for changes in the model's behaviour compared to simulations conducted for CMIP5.
In CCSM4/CESM1(CAM4), stratospheric aerosol is treated by prescribing a
single zonally averaged species. The prescription consists of a monthly mean
mass (kg m
In CCSM4/CESM1(CAM4) the stratospheric aerosol mass is interpreted by the radiative transfer code via the predefined mass-specific extinctions, single scattering albedos, and asymmetry parameters. These parameters are calculated using the constants defined above and are stored in lookup tables for the short-wave and long-wave radiative transfer schemes separately (the table has a single dimension that varies by spectral band) for use by each of the spectral bands in the Community Atmosphere Model Radiative Transfer (CAMRT) parameterization. The optical property file for CCSM4/CESM1(CAM4) is entitled “sulfuricacid_cam3_c080918.nc” and may be found in the CESM input data repository at “/glade/p/cesm/cseg/inputdata/atm/cam/physprops/”. This information is combined with similar information from other radiatively active species in CCSM4/CESM1(CAM4) as specified by Neale et al. (2010).
Here we summarize the main features of the stratospheric aerosol prescription in CESM1(CAM5) so that differences may be accounted for between future simulations using the new CESM1 stratospheric aerosol scheme and previous simulations conducted for CMIP5. For a discussion of the parameterization used to represent stratospheric aerosol in CESM1(CAM5), please see chap. 4 of Neale et al. (2012).
CESM1(CAM5) specifies the stratospheric aerosol as a mass mixing ratio of wet sulfate aerosol (i.e. a mixture of 75 % sulfuric acid and 25 % water) to dry air as a function of height, latitude, and time. Unlike CCSM4/CESM1(CAM4), CESM1(CAM5) has the ability to include non-zonally symmetric aerosol (i.e. varying by longitude). In the update described below, this ability has been spread to all present configurations of CESM1.
CESM1(CAM5) utilizes the rapid radiative transfer method for GCMs (RRTMG) (Mlawer et al., 1997; Iacono et al., 2008). For each short-wave band calculation, extinction optical depth, single scattering albedo, and asymmetry factors are determined from the aerosol properties according to their size and mass and radius. For each long-wave band, only absorption optical depth is calculated.
As with CCSM4/CESM1(CAM4), to interact with the radiative transfer scheme,
CESM1(CAM5) calculates mass-specific properties over each spectral band of
RRTMG. The calculations assume the size distribution of the aerosol to be a
log-normal distribution with a geometric mean radius
Note that for a log-normal distribution, the geometric mean radius
(
In CESM1(CAM5) the mass-specific aerosol extinction, scattering, and
asymmetry factor are defined as
Similarly to CCSM4/CESM1(CAM4), the standard configuration of CESM1(CAM5)
uses the stratospheric aerosol forcing data set over the period 1850 to 2010
from Ammann et al. (2003). This data set does not take advantage of the
parameterization in CESM1(CAM5), as described above, to modulate the changes
in stratospheric size distribution (i.e. variations in
In CESM1(WACCM4) and CESM1(CAM4-chem), the prescription of stratospheric aerosol differs from CCSM4/CESM1(CAM4) and CESM1(CAM5) due to the need to specify SAD for use in the heterogeneous stratospheric chemistry parameterization. Marsh et al. (2013), building upon Tilmes et al. (2009), describe the CESM1(WACCM4) scheme. For details about CESM1(CAM4-chem), see Lamarque et al. (2012). In both model configurations, the SAD is prescribed from a monthly zonal-mean time series derived from observations and is identical to that specified in the CCMVal2 REF-B1 simulations (Eyring et al., 2010; SPARC CCMVal, 2010). The standard SAD input file is “/glade/p/cesm/cseg/inputdata/atm/waccm/sulf/SAD_SULF _1849-2100_1.9x2.5_c090817.nc”.
The mass of aerosol to be used by CAMRT (which is the standard radiative
transfer model used in both model configurations) is derived from the
specified SAD by determining a volume density of sulfate aerosol by assuming
a log-normal size distribution with fixed size (
In this work, we have unified the stratospheric aerosol parameterization for
CESM1(CAM4) and CESM1(CAM4-chem) (both found within NCAR's CESM1 code
repository under tag cesm1_1_1_ccmi23), CESM1(WACCM4), and CESM1(CAM5)
(both of the latter configurations found within the CESM1 repository under
tag cesm1_1_1_ccmi30) to take advantage of the new forcing data prepared
for the CCMI simulations (Eyring et al., 2013). The new forcing file is
derived from the SAGE 4
Here we only describe the changes made to the CESM1's configurations. For the more detailed documentation of CAMRT (the radiation scheme in CESM1(CAM4), CESM1(CAM4-chem), and CESM1(WACCM4)) and RRTMG (utilized in CESM1(CAM5)), which were not modified here, please see Neale et al. (2010, 2012) as noted above. In summary, three main changes occurred: (1) the forcing input file (this has the main advantage of updating the stratospheric aerosol masses to reflect the most current observational and modelling studies as well as providing a coherent data set of aerosol mass, surface area density, and radius), (2) CAMRT has been modified to allow for variations in the effective radius of the aerosol distribution with time as provided by the new forcing file, and (3) the optical lookup tables for both CAMRT and RRTMG were updated with new Mie calculations for use in all model configurations.
For the new implementation of the stratospheric aerosol forcing in CESM1 we
utilize the new stratospheric aerosol data set derived to force models
participating in CCMI (Eyring et al., 2013). The CCMI stratospheric aerosol
forcing file (the data and more detailed description are available from
The original CCMI stratospheric aerosol forcing file provides the mass loading, SAD, and size of aerosol from 1960 to 2012. The original file was modified slightly to form the new standard input file for CESM1 for a period ranging from 1950 to 2012. The current CCMI forcing file is entitled “CESM_1949_2100_sad_V2_c130627.nc” and can be found in the CESM input data repository. The main difference between this file and the original file is that the monthly mean values from the minimum in stratospheric aerosol observed in 1998 and 1999 have been used to fill in the years from 1949 to 1959 and into the future from 2012 to 2100. This was done in accordance with the assumptions and scenarios laid out by the CCMI specification (Eyring et al., 2013).
To enable simulations prior to 1960, an additional forcing file is available entitled “CESM_1849_2100_sad_V3_c160211.nc”. This file is identical to the “CESM_1949_2100_sad_V2_c130627.nc” from 1960 to 2100. Prior to this period (i.e. from 1849 to 1960) we have added the impact of colossal volcanic eruptions (VEI 5 and larger) and a representation of the background stratospheric aerosol layer. For this period, we have included the following seven colossal volcanic perturbations to the background stratospheric aerosol layer: (1) Sheveluch in February 1854, (2) Krakatau in May 1883, (3) Okataina in June 1886, (4) Santa Maria in October 1902, (5) Ksudach in March 1907, (6) Novarupta in June 1912, and (7) Bezymianny in October 1955. In between the eruptions, background levels of stratospheric aerosol are based on the monthly mean mass and size from the minimum in stratospheric aerosol observed in 1998 and 1999 (as done for the 1949 to 2100 period described above).
The volcanic perturbations were included in the forcing file by scaling the
aerosol mass, size and SAD of the Pinatubo eruption from 1991 to the 1998
eruption according to the ratio of injected mass SO
To implement the use of the new stratospheric input file in CESM1, several modifications were made to the mechanics of how the CESM1 ingests stratospheric aerosol forcing files so that time-varying information about the size of the aerosol could be included within the radiative calculations. This resulting code, entitled “prescribed_strataero.F90”, is located in the chemistry utilities of CESM ({top level directory of model version}/models/atm/cam/src/chemistry/utils/prescribed_ strataero.F90). This code reads the necessary input parameters and transforms them into the units and grid needed by the model configuration. By default, it also masks out any aerosol below the model's tropopause. This is an option that may easily be changed. The code may also be easily modified and adapted to input values from other input files.
It should be noted that CESM1 linearly interpolates the input file in time and space to match the time step and spatial grid of the model configuration. As such, this results in differences between the monthly mean aerosol specified from the ingested forcing file and monthly mean values of the aerosol that the model actually experiences. This is particularly an issue during periods of rapid change in aerosol. Similar issues have been noted for the specification of ozone in Neely et al. (2014). The best method to counteract errors due to this issue is to specify the aerosol values at the highest temporal cadence available.
As in previous versions of the model, here we assume that the stratospheric aerosol is comprised of a mixture of 75 % sulfuric acid and 25 % water and conforms to a log-normal size distribution. Unlike the previous parameterizations, the distribution has a varying effective radius that is specified by the input file.
As described above, CESM1(CAM5) already provided the necessary mechanism to
use the temporally and latitudinally varying aerosol size information from
the input file. For CESM1(CAM4), CESM1(CAM4-chem-CCMI), and
CESM1(WACCM4-CCMI) we adapted the short-wave mechanism of CESM1(CAM5) to use
both mass and
To create the new optical lookup table for CAMRT, a new set of Mie efficiency
terms was determined for a range of wavelengths and size parameters
appropriate for the CAMRT and the new aerosol input file. The index of
refraction used in these calculations is based on the assumption of a 75 to
25 % mixture of sulfuric acid and water at 293 K. Data for this were
compiled from the GEISA spectroscopic database
(
All Mie calculations were done using the “MATLAB Functions for Mie Scattering and Absorption” developed by Mätzler (2002). The code used to create the CAMRT optical properties may be found in Sect. S1 of the Supplement.
A similar method was used to also update the optical properties file for all configurations of CESM1 that utilize RRTMG (i.e. CESM1(CAM5). The new optical properties file for model configurations using RRTMG is entitled “volc_camRRTMG_byradius_sigma1.6_c130724.nc” and is available from CESM's input data repository (/trunk/inputdata/atm/cam/physprops/volc_camRRTMG _byradius_sigma1.6_c130724.nc). This code is attached in Supplement Sect. S2. The main differences between the two versions of the code are the spectral bands of the two radiative transfer schemes. This is a direct consequence of the different bands used by CAMRT vs. RRTMG. In addition, only the short-wave parameters were updated for the CAMRT files, while both the short-wave and long-wave were updated in RRTMG files. The reason for only adjusting the short-wave parameters in CAMRT are purely historical due to the complex entanglement of the different species in the CAMRT long-wave parameterization. It was also thought that little improvement would have been made to the model's response to perturbations to the stratospheric aerosol layer.
Globally averaged stratospheric AOD at 550 nm integrated from
15 km and above. The red line represents the AOD as simulated by the
original CCSM4/CESM1 simulation configurations. The green line represents the
new AOD determined from the SAGE
In Fig. 2 we document the resulting global SAOD between 1960 and 2000 produced by the new prescribed stratospheric aerosol parameterization utilizing forcing entitled “CESM_1949_2100_sad_V2_c130627.nc” (referred to as the new CESM1 AOD). This is in comparison to the SAOD resulting from the parameterization used by the original CCSM4/CESM1 and the latest version of the observationally based Sato et al. (1993) data set. Several differences are apparent in the comparison. In general, the peak global mean SAOD after each major eruption is reduced in the new CESM1 compared to both the original CCSM4/CESM1 specification and the AOD of the Sato et al. (1993) data set. The one exception to this is the 1963 Agung eruption in which the Sato et al. (1993) results show an even more reduced, though broader, peak than both the original CCSM4/CESM1 and new CESM1. Between the Agung eruption in 1963 and the 1974 Fuego eruption, there are many significant differences between the three SAOD time series. Notably, the CESM1 SAOD does not peak in 1968 as the other two data do and Sato et al. (1993) show higher levels of aerosol throughout the period. The reasons for these differences are due to the underlying assumptions about the eruptions included in the creation of the forcing file. Though several moderate eruptions (VEI 4) are known to have occurred in this period (Stothers, 2001; Bauer, 1979; Langmann, 2013; Sato et al., 1993), measurements are sparse and, without further investigation, the correct representation of these perturbations to the stratospheric aerosol burden is highly uncertain. After Fuego, outside of periods perturbed by volcanic eruptions, Sato et al. (1993) and new CESM1 display similar levels of background SAOD, while the CCSM4/CESM1 does not account for background stratospheric aerosol (the impact of this exclusion of background stratospheric aerosol is discussed in Solomon et al., 2011).
Figure 3 examines the differences between the new and old prescribed stratospheric aerosol schemes in more detail. Figure 3 shows a comparison of the resulting monthly mean and zonal mean SAOD after the 1991 Mt. Pinatubo eruption from old and new schemes in CESM1(CAM4) (Fig. 3a and b) and CESM1(CAM5) (Fig. 3c and d). Figure 3 shows that, to first order, the most significant change in the new scheme is the distribution of mass used in the forcing file. For further examination of the impact of the individual changes to the radiation code and forcing file on CESM1(CAM4) and CESM1(CAM5), see Sect. S3 of the Supplement. Results for CESM1(WACCM4) and CESM1(CAM4-Chem) are not shown as the new stratospheric aerosols are identical to those utilized in the new CESM1(CAM4) prescription.
Monthly times series comparison of the zonal mean SAOD at 550 nm
after the 1991 Mt. Pinatubo eruption for the old
Global, monthly, ensemble, and mean change in the top of atmosphere
radiative flux due to the simulated Mt. Pinatubo eruption in June of 1991.
Each original and new volcanic ensemble member is differenced from a set of
simulations (not shown) conducted with identical initial conditions but with
no stratospheric AOD forcing. Shaded regions represent the
To examine the impact of the new stratospheric forcing on CESM1's simulated climate response, we performed an experiment that compared five ensemble members of CESM1(CAM5) with the new stratospheric aerosol parameterization vs. five members using the original parameterization over the period influenced most strongly by the 1991 Mt. Pinatubo eruption. Each of the five members in the respective ensembles used different initial ocean states and atmospheric initial conditions that were derived from the original five CESM1(CAM5) CMIP5 simulations. The differences between the two ensembles show the possible improvement the new scheme has on CESM1's ability to simulate the climate response to a colossal volcanic eruption.
In Fig. 4 we show the impact on the simulated global monthly mean top of atmospheric net radiative flux. A reduction is seen at the peak of the stratospheric aerosol perturbation in late 1991. Notably, outside the period of highest aerosol loading after the eruption (i.e. the second half of 1991), there is very little difference in the net radiative flux between the two ensembles.
In Fig. 5, the global annual mean temperature (i.e. the response to the
differences in the simulated forcings in Fig. 4) is shown for each of the
two ensembles in comparison to observations from the GISS Surface Temperature
Analysis (GISTEMP) (Hansen et al., 2010; GISTEMP Team, 2015). For the
original CCSM4/CESM1 forcing parameterization, the difference between the
model and analysis record is similar to Fig. 1, while the new
parameterization simulates a trend that more closely follows the observed
record within the variability of the model runs and error estimate of the
observations. The most significant improvement is observed in the 1992 global
annual temperature. As in Fig. 1, the original CCSM4 parameterization causes
the simulated ensemble-mean global average temperature anomaly to drop
Global, annual, ensemble, and mean temperature anomaly due to the
observed (GISTEMP) and simulated Mt. Pinatubo eruption in June of 1991.
Anomalies are referenced to the 1990 annual mean in each ensemble member.
Shaded regions represent the
Tropical (20
Note that the observed global mean temperature in 1991 contains a strong El
Niño–Southern Oscillation (ENSO) signal (Thompson et al., 2009; Canty et
al., 2013), which the model ensemble will not accurately reproduce due to its
own inherent variability. This causes significant difficulty in the use of
changes in global mean temperature as a metric for model improvement after
the 1991 Pinatubo eruption. Studies that have attempted to isolate the pure
volcanic surface cooling signal from other sources of variability (including
ENSO) result in estimates of maximum cooling ranging from
In addition to the changes in the global surface temperature response, the new stratospheric aerosol scheme drastically improves the CESM1(CAM5)'s performance in representing stratospheric heating after a colossal volcanic eruption. This is shown in Fig. 6, where we compare the 50 hPa temperature anomaly for the two ensembles against the Radiosonde Innovation Composite Homogenization (RICH) (Haimberger et al., 2008). This is notable as the original stratospheric aerosol scheme in CCSM4/CESM1 caused heating that was over 7 times the observed anomaly and had significant implications for changes in stratospheric dynamics and chemistry. In the new CESM1 scheme, the simulated stratospheric heating is at most double the observed anomaly.
Here we describe the new prescribed stratospheric aerosol parameterization for CESM1. This work represents a significant improvement in the prescribed representation of stratospheric aerosols in CESM1 as it unifies the treatment between the chemical and radiative transfer parameterizations within all atmospheric models under the CESM1 umbrella. We have shown that the new prescription of stratospheric aerosol consistently improves the representation of stratospheric aerosol and resulting model response, especially after colossal volcanic eruptions. Most significantly, the new scheme more accurately simulates the stratospheric temperature response to the 1991 Pinatubo eruption. Results also indicate that the new scheme decreases CESM's global mean temperature response, but observed and modelled climate variability precludes statements as to the significance of this improvement.
This scheme may also be easily adapted to other stratospheric aerosol forcing scenarios, such as those used in geoengineering experiments, by simply changing the masses, radii, and SAD of the input file, as has been done in Xia et al. (2016). For future historical simulations, there are two possible new prescribed stratospheric aerosol data sets being prepared for CMIP6 that the new CESM1 parameterization will be able to utilize. One will be an update to the CCMI data set presented here that covers the period form 1850 to present. The second file will be created using the output of a prognostic stratospheric aerosol scheme within CESM1 (Mills et al., 2016) that simulates the stratospheric aerosol layer from 1850 to the present day and uses only the injections of SO2 from the VolcanEESM (Neely and Schmidt, 2016) database.
Here we have focused on the technical specification of the new implementation of prescribed stratospheric aerosol in CESM1 and the impact this new specification has on the global radiation budget. As mentioned, the implementation also includes improvements to CESM1's specified stratospheric aerosol SAD. The impact the new SAD forcing has on the chemical parameterization of CESM1 is described in Tilmes et al. (2016).
The original stratospheric aerosol data set derived to force models
participating in CCMI, which is the basis of the work presented here, is
available from
Released CESM code is made available through a subversion repository. The
code may be downloaded by following the specific User's Guide for each model
version after registering as a CESM user. For more information, please see
The scripts used to create the optical parameters are attached in the Supplement. All questions about these scripts should be directed to the lead author.
We thank Daniel Marsh, Rolando Garcia, Sean Santos, and Michael Mills for their assistance in developing the new volcano parameterization. CESM is sponsored by the National Science Foundation (NSF) and the U.S. Department of Energy (DOE). Administration of the CESM is maintained by the Climate and Global Dynamics Division (CGD) at the National Center for Atmospheric Research (NCAR). Computing resources were provided by the Climate Simulation Laboratory at NCAR's Computational and Information Systems Laboratory (CISL), sponsored by the National Science Foundation and other agencies. Ryan R. Neely III was supported by the National Center for Atmospheric Research's Advanced Study Program (NCAR ASP) during this work. Edited by: A. Stenke Reviewed by: two anonymous referees