The sixth phase of the Coupled Model Intercomparison Project (CMIP6) is the
latest modeling effort for general circulation models to simulate and
project various aspects of climate change. Many of the general circulation
models (GCMs) participating in CMIP6 provide archived output that can be
used to calculate effective climate sensitivity (ECS) and forecast future
temperature change based on emissions scenarios from several Shared
Socioeconomic Pathways (SSPs). Here we use our multiple linear regression
energy balance model, the Empirical Model of Global Climate (EM-GC), to
simulate and project changes in global mean surface temperature (GMST),
calculate ECS, and compare to results from the CMIP6 multi-model ensemble.
An important aspect of our study is a comprehensive analysis of uncertainties
due to radiative forcing of climate from tropospheric aerosols (AER RF) in
the EM-GC framework. We quantify the attributable anthropogenic warming rate
(AAWR) from the climate record using the EM-GC and use AAWR as a metric to
determine how well CMIP6 GCMs replicate human-driven global warming over the
last 40 years. The CMIP6 multi-model ensemble indicates a median value of
AAWR over 1975–2014 of 0.221
The goals of the Paris Agreement, negotiated in December of 2015, are to
keep global warming below 2.0
Several prior studies have used a multiple linear regression approach to model the GMST anomaly in order to quantify the impact of anthropogenic and natural factors on climate (Foster and Rahmstorf, 2011; Lean and Rind, 2008, 2009; Zhou and Tung, 2013). Typically, total solar irradiance, volcanoes, and the El Niño–Southern Oscillation (ENSO) are the natural components represented in the multiple linear regression. Greenhouse gases and aerosols are the anthropogenic factors. We use multiple linear regression, in connection with a dynamic ocean module that accounts for the export of heat from the atmosphere to the ocean, to represent the natural and anthropogenic components of the climate system. In addition to the typical natural factors listed above, we include the Atlantic Meridional Overturning Circulation (AMOC), Pacific Decadal Oscillation (PDO), and Indian Ocean Dipole (IOD) to provide a robust representation of the natural climate system (Canty et al., 2013; Hope et al., 2017). Our anthropogenic components also include the effect of land-use change (i.e., deforestation) on Earth's albedo and the export of heat from the atmosphere to the ocean as the atmosphere warms.
Our analysis builds on the work of Canty et al. (2013) and Hope et al. (2017) and includes several key updates. One is the extension back in time of our analysis to 1850. The Hadley Centre Climatic Research Unit (Morice et al., 2012, 2021), Berkeley Earth Group (Rohde and Hausfather, 2020), and Cowtan and Way (2014) provide GMST records starting in 1850, which now allows for simulations of GMST that cover 170 years. The second update is the use of the Shared Socioeconomic Pathways (SSPs) (O'Neill et al., 2017) as our climate scenarios for greenhouse gas and aerosol abundances. The third is the adoption of an upper ocean to our model, formulated in a manner that matches the equations of Bony et al. (2006) and Schwartz (2012). A description of the model, the various input parameters used, and the updates listed above is given in Sect. 2. Section 3 shows results of CMIP6 and EM-GC comparisons to the historical climate record, estimations of effective climate sensitivity (ECS) and comparisons of our model and CMIP6 projections of future GMST change. A discussion of these results is provided in Sect. 4, along with concluding remarks.
In this analysis we use the empirical model of global climate (EM-GC), which
provides a multiple linear regression energy balance simulation of GMST. As
detailed in the following paragraphs, the EM-GC solves for ocean heat uptake
efficiency (
We consider several anthropogenic and natural factors to be components of
The term AMOC
The dimensionless parameter
Our model explicitly accounts for the export of heat from the atmosphere to
the world's oceans (i.e., ocean heat export or OHE). The quantity
Measured and modeled GMST anomaly (
We use the reduced chi-squared (
The calculation of
Figure 1 shows the observed (HadCRUT5) and modeled GMST anomaly from
1850–2019 and the various anthropogenic and natural components that
constitute modeled GMST. Figure 1a shows the value of climate feedback, 1.62 W m
Altering the training period of our model has a slight effect on our results (see Figs. S2, S3, and the Supplement for information on various training periods). We project relatively similar results for end-of-century warming for training periods that start in 1850 and end in either 2009 or 1999 compared to results shown throughout the paper for a training period of 1850 to 2019, indicating the stability of our approach. As detailed in the Supplement, we do find some differences from the results shown in the paper upon use of a training period of 1850 to 1989 due to the reduction in the number of years considered from the available OHC records.
We use seven global mean surface temperature anomaly records. These records include the Hadley Centre Climatic Research Unit version 4 (HadCRUT4; Morice et al., 2012) and version 5 (HadCRUT5; Morice et al., 2021) from 1850–2019, National Centers for Environmental Information NOAAGlobalTemp v5 (NOAAGT; Smith et al., 2008; Zhang et al., 2019) from 1880–2019, NASA Goddard Institute of Space Studies Surface Temperature Analysis v4 (GISTEMP; Hansen et al., 2010) from 1880–2019, Berkeley Earth Group (BEG; Rohde and Hausfather, 2020) from 1850–2019, Cowtan and Way (2014) (CW14) from 1850–2019, and the Japanese Meteorological Agency (JMA; Ishihara, 2006) from 1891–2019. We use the uncertainty time series from HadCRUT4 for all GMST records because the HadCRUT4 uncertainty provides a realistic description of the variation in GMST among the seven records (see the Supplement, Figs. S4 and S5, and Table S1 for more information). Our analysis primarily uses the HadCRUT5 GMST data set, but in some sections, results are shown for the other data sets. All temperature anomalies are with respect to a pre-industrial baseline (1850–1900). To alter each data record so that the temperature anomaly is relative to the same pre-industrial baseline, we adjust all data sets relative to the HadCRUT5 baseline of 1961–1990. We then adjust each data set by the same amount to the HadCRUT5 pre-industrial baseline as described in the Supplement.
For this analysis, we use the estimates of the future abundances of greenhouse gases and aerosols provided by the SSPs. There are 26 scenarios, five baseline pathways, and 21 mitigation scenarios. The baseline pathways follow specific narratives for factors such as population, education, economic growth, and technological developments of sources of renewable energy (Calvin et al., 2017; Fricko et al., 2017; Fujimori et al., 2017; Kriegler et al., 2017; van Vuuren et al., 2017) to represent several possible futures encompassing different challenges for adaptation to and mitigation of climate change as illustrated in Fig. 1 of O'Neill et al. (2014). The 21 mitigation scenarios follow one of the baseline pathways but include specific climate policy to reach a designated radiative forcing at the end of the century.
As part of CMIP6, the ScenarioMIP experiment (O'Neill et al., 2016) includes eight SSPs
(SSP1-1.9, SSP1-2.6, SSP4-3.4, SSP2-4.5, SSP4-6.0, SSP3-7.0, SSP5-8.5, and
SSP5-3.4-OS) that GCMs use to project future GMST. The first number is the
reference pathway that the scenario follows (i.e., SSP1 follows the first
SSP narrative), and the numbers after the dash are the target radiative
forcing at the end of the century (i.e., SSP1-2.6 reaches around 2.6 W m
Observed and projected greenhouse gas mixing ratios.
The historical values of GHG mixing ratios were provided by Meinshausen
et al. (2017b) from 1850–2014. We used the equations from Myhre (1998) to calculate the change in RF due to carbon
dioxide (CO
The value of the change in total aerosol radiative forcing (direct and
indirect) in 2011 relative to pre-industrial (AER RF
Measured (HadCRUT5) and EM-GC simulated GMST anomaly
(
We use the total aerosol RF time series provided by the SSP database for
each SSP scenario. The database provides AER RF from 2005–2100, with values
for all SSPs nearly identical until about 2010 (Riahi
et al., 2017; Rogelj et al., 2018). In the EM-GC, we calculate temperature
projections over the entire observational period beginning in 1850. We
create AER RF time series that begin in 1850 and span the range of
uncertainty given by Chapter 8 of IPCC 2013. We use historical estimates of
AER RF from 1850–2014 for the four RCPs provided by the Potsdam Institute
for Climate Research (Meinshausen
et al., 2011). The AER RF value in 2014 from the appropriate
historical estimate (i.e., RCP4.5 is used for SSP2-4.5) is scaled by a
constant factor such that the historical RCP value at the end of 2014
matches the SSP time series at the start of 2015. This scaling yields a
continuous time series for the RF of climate due to tropospheric aerosols.
This scaled time series has AER RF
We use the TSI time series provided for the CMIP6 models from 1850–2014 (Matthes et al., 2017) and append values from the
Solar Radiation and Climate Experiment (SORCE) (Dudok de Wit et al., 2017) for 2015 to the end of
2019. The values of TSI
The time series for SAOD is a combination of values computed from extinction
coefficients for the CMIP6 GCMs (Arfeuille et al., 2014) from 1850–1978
and the Global Space-based Stratospheric Aerosol Climatology (GloSSAC v2.0) (Thomason et al., 2018) from
1979–2018. Extinction coefficients at 550 nm were integrated from the
tropopause to 39.5 km and averaged over the globe using a cosine of latitude
weighting. The CMIP6 and GloSSAC extinction coefficients span 80
We use the MEI.v2 (Wolter and Timlin, 1993;
Zhang et al., 2019) to characterize the influence of ENSO on GMST. In order
to obtain a time series that spans the entire training period of our model,
1850–2019, we append three time series to create an MEI.v2 over the
full extent of our model training period. The MEI.v2 provides 2-month
averages of empirical orthogonal functions of five different climatic
variables from 1979 to the present (Zhang et al., 2019). To
have the ENSO index extend back to 1850, we compute differences in SST
anomalies over the tropical Pacific basin as defined by the MEI.v2 from
1850–1870 using HadSST3 (Kennedy et al.,
2011). Our internal computation of this surrogate for the MEI is then
appended to the MEI.ext of Wolter
and Timlin (2011), which extends from 1871–1978, and the MEI.v2 of (Zhang et al., 2019) (1979–2019). This full time series
provides a representation of ENSO that covers 1850 to the present.
Consistent with prior regression-based approaches (Foster and Rahmstorf, 2011; Lean and
Rind, 2008), we find that a significant portion of the monthly and at times
annual variation in GMST is well explained by ENSO (Fig. 1d). As for the
other natural terms, we assume ENSO
The Pacific Decadal Oscillation is the leading principal component of North
Pacific monthly SST variability poleward of 20
The Indian Ocean Dipole is based on the difference in the anomalous sea
surface temperature (SST) between the western equatorial Indian Ocean
(50–70
We use the Atlantic multidecadal variability (AMV) index as the area-weighted monthly mean SST from HadSST4 (Kennedy et al., 2019)
between the Equator and 60
Ocean heat content data records from five recent and independent papers are
used in this study. We utilize OHC data from Balmaseda
et al. (2013), Carton et al. (2018), Cheng et al. (2017), Ishii et al. (2017), and Levitus et al. (2012), as well as the average of the records to
model the export of heat (OHE) from the atmosphere to the ocean. Figure S9
shows these five OHC records and the multi-measurement average. While
most of these data sets have a common origin, they differ in how extensive
temporal and spatial gaps in the coverage of ocean temperatures have been
handled, ranging from data assimilation (Carton et al., 2018) to an iterative
radius-of-influence mapping method (Cheng et al., 2017). The five data sets
are all set to zero in 1986, which is the midpoint of the multi-measurement
time series, by applying an offset for visual comparison. Since OHE
(units: W m
As noted above, the calculation of
The choice of OHC record has only a small effect on future projections of
GMST using the EM-GC. Figure 4 illustrates the effect of varying the OHC record
on future temperature. The bottom boxes in each panel show the observed and modeled OHC,
the value of
Measured (HadCRUT5) and EM-GC simulated GMST change
(
The attributable anthropogenic warming rate, or AAWR, is the time rate of change in GMST due to humans from 1975–2014. We use AAWR as a metric in the EM-GC framework to quantify the human influence on global warming over the past few decades and, most importantly, to also assess how well the CMIP6 GCMs can replicate this quantity. This analysis is motivated by the study of Foster and Rahmstorf (2011), who examined the human influence on the time rate of change in GMST from 1979–2010 using a residual method. We extend the end year of our analysis to 2014 because this is the last year of the CMIP6 historical simulation. We pushed the start year back to 1975 so that our analysis covers a 40-year period, over which the effect of human activity on GMST rose nearly linearly with respect to time (Figs. 1b and S10c).
We calculate AAWR utilizing the EM-GC by computing a linear fit to the
The determination of AAWR from historical CMIP6 near-surface air temperature
output involves conducting a regression of deseasonalized, globally
averaged, monthly
We also use a second method to extract the value of AAWR from the CMIP6
multi-model ensemble. This method, termed LIN, involves a linear regression
of global, annual average values of GMST from the CMIP6 multi-model ensemble (Hope et al., 2017). For LIN, we exclude the years of
obvious volcanic influence on the rise in GMST from the CMIP6 multi-model
ensemble historical simulations: i.e., data for 1982 and 1983 (following the
eruption of El Chichón) and 1991 and 1992 (following the eruption of
Mount Pinatubo) are excluded. Archived global, annual average values of GMST
covering 1975–2014, excluding these 4 years, are fit using linear
regression, with the AAWR set equal to the slope of the fit. Values of AAWR
for 1975–2014 found using LIN are also shown in Table S4 for each GCM.
Analysis of AAWR for these 50 GCMs of LIN versus REG (see Fig. S14) results
in a correlation coefficient (
The CMIP6 multi-model ensemble provides simulations of near-surface air
temperature (TAS), which we use to calculate AAWR. The EM-GC uses blended
near-surface air temperature to determine values of AAWR. Cowtan et al. (2015) provide a method to create blended near-surface air temperature
output from the GCMs. The CMIP6 multi-model ensemble contains archived
fields of TAS and the temperature at the interface of the atmosphere and the
upper boundary of the ocean (TOS) (Griffies
et al., 2016), whereas only a subset of GCM groups provide the archived land
fraction needed to calculate blended near-surface air temperature using the
Cowtan et al. (2015) method. Cowtan et al. (2015) compared the modeled and measured trend in global temperature over
1975–2014 and found a 4.0 % difference in the trend upon the use of
blended temperature from CMIP5 GCMs rather than global modeled TAS. Their
analysis focused on a comparison of modeled and measured temperature, not
just the anthropogenic component. We have used the method of Cowtan et al. (2015) to create blended CMIP6
temperature output for the CMIP6 GCMs that provide TAS, TOS, and the land
fraction. Upon our use of blended CMIP6 temperature output for these GCMs
and calculation of AAWR for 1975–2014, we find that AAWR based on blended
CMIP6 temperature is 3.5 % lower than AAWR found when using only TAS. Tokarska et al. (2020b) estimate an effect of
0.013
The equilibrium climate sensitivity represents the warming that would occur
after the climate equilibrated with atmospheric CO
For the estimate of climate sensitivity from the CMIP6 multi-model ensemble,
we use the method described by Gregory et al. (2004) (see the
Supplement and Fig. S15 for more information). The Gregory et al. (2004)
method also estimates effective climate sensitivity from the CMIP6 GCMs (Gregory
et al., 2004; Sherwood et al., 2020; Zelinka et al., 2020) because it
assumes that the feedbacks inferred from the first 150 years of the abrupt
The estimates of climate sensitivity from Eq. (10) and those found using the
Gregory et al. (2004) method are termed “effective” because they assume
that climate feedback inferred from either the historical climate record or the
abrupt
Probabilistic forecasts of the future rise in GMST for various SSPs are an
important part of our analysis. Probabilities of AAWR and ECS are computed
by considering the uncertainty in AER RF
Aerosol weighting method.
Figure 5b shows the value of AAWR (
An important measure of any climate model is the ability to accurately simulate the human influence on the global mean surface temperature (GMST) anomaly. We use the attributable anthropogenic warming rate (AAWR) found by our highly constrained Empirical Model of Global Climate (EM-GC) to quantify how well the CMIP6 multi-model ensemble (see Table S7 for a list of CMIP6 GCMs analyzed in this study) is able to simulate the human influence on global warming over the past several decades.
Figure 6 compares values of AAWR from 1975–2014 computed using our EM-GC
with AAWR found utilizing archived output from the CMIP6 multi-model
ensemble. Seven GMST data sets and five OHC records can be used to estimate
AAWR with the EM-GC. For each choice, AAWR exhibits sensitivity to the
variation of the time series of radiative forcing due to tropospheric
aerosols. Each box-and-whisker plot found using our EM-GC shows, for a
particular choice of GMST and OHC data record, the 25th, 50th, and
75th percentiles of AAWR (box) and the 5th and 95th percentiles
(whiskers) found using the aerosol weighting method described in Sect. 2.5.
The star symbol indicates the minimum and maximum values of AAWR for each
value of GMST data set and OHC record. The choice of OHC record and GMST
data set has a slight effect on AAWR, as shown by the colored EM-GC symbols
in Fig. 6. The averages of the five 25th, 50th, and 75th
percentiles of AAWR found using the HadCRUT5 data set for GMST are 0.138,
0.157, and 0.176
AAWR from the EM-GC and CMIP6 multi-model ensemble for 1975–2014. Seven temperature data sets and five ocean heat content records are used to compare values of AAWR computed from the EM-GC. The box represents the 25th, 50th, and 75th percentiles, the whiskers denote the 5th and 95th percentiles, and the stars show the minimum and maximum values of AAWR from the EM-GC based on the aerosol weighting method described in Sect. 2.5. The red box labeled “CMIP6” shows the 25th, 50th, and 75th percentiles, the whiskers represent the 5th and 95th percentiles, and the stars denote the minimum and maximum values of AAWR from the 50-member CMIP6 multi-model ensemble.
The box-and-whisker symbol labeled CMIP6 in Fig. 6 shows the 5th,
25th, 50th, 75th, and 95th percentiles of AAWR
calculated from 50 GCMs, also from 1975–2014, as described in Sect. 2.3. The
stars denote the minimum and maximum values of AAWR from the GCMs. Two CMIP6
models exhibit values of AAWR similar to the median values we infer from the
HadCRUT4, CW14, NOAAGT, BEG, GISTEMP, and HadCRUT5 data records using the
EM-GC. In particular, INM-CM5-0 (Volodin and
Gritsun, 2018) yields 0.147
Our determination that the rate of global warming from the CMIP6 multi-model ensemble over the time period 1975–2014 significantly exceeds the rise in GMST attributed to human activity is aligned with a similar finding highlighted in Fig. 11.25b from Chapter 11 of the IPCC 2013 report that CMIP5 models tend to warm too quickly compared to the actual climate system over the time period 1975–2014 (Kirtman et al., 2013). The values of AAWR from the CMIP6 multi-model ensemble from our analysis present a similar finding as Tokarska et al. (2020b) and CONSTRAIN (2020) that some of the CMIP6 models overestimate recent warming trends. Tokarska et al. (2020b) examine the trend in the human component of GMST from 1981–2014. We arrive at a similar conclusion as these studies that CMIP6 GCMs overestimate the rate of global warming for the 1982–2014 time period of AAWR as shown in Tables S2 and S3. Our results, the finding by the IPCC 2013 report, Tokarska et al. (2020b), and CONSTRAIN (2020) appear to be quite different than the conclusion of Hausfather et al. (2020) that past climate models have matched recent temperature observations quite well. The Hausfather et al. (2020) study does not examine CMIP5 GCMs, let alone CMIP6 GCMs, and the last two rows of their Table 1 indicate that the skill of climate models forecasting the change in GMST over time decreased considerably between the Third Assessment Report (TAR) and the Fourth Assessment Report (AR4). The change in temperature over time for the TAR and AR4 only spans 17 and 10 years, respectively (Hausfather et al., 2020). In Fig. 6, we examine the ability of the GCMs to simulate the rise in GMST attributed to humans over a 40-year time period, which provides a better measure of how well the models simulate the observations than the shorter time period. The temperature change over time for the TAR and AR4 examined by Hausfather et al. (2020) ends in 2017, which was right after a very strong ENSO, so their analysis may be influenced by the 2015 to 2016 ENSO event. In contrast, our analysis of AAWR is not influenced by natural variability such as ENSO because we examine the human component of global warming after explicitly accounting for and removing the influence of ENSO on GMST. Consequently, our determination of AAWR from observations (Table S2) and GCMs (Table S3) depends only to a small extent on the specification of the start year (for values ranging from 1970 to 1984) and end year (2004 to 2018). Our analysis shows that upon quantification of the human driver of global warming within both the data record and climate models, the CMIP6 GCMs warm faster than observed GMST over the past 4 decades regardless of precise specification of the start and end year.
Climate sensitivity is a metric often used to compare the sensitivity of
warming among GCMs and with warming inferred from the historical
climate record. Figure 7 shows values of effective climate sensitivity (ECS)
inferred from the climate record using our EM-GC, seven GMST data sets, and
five OHC records. As for AAWR, the largest variation in ECS is driven by
uncertainty in AER RF
ECS from the EM-GC and the CMIP6 multi-model ensemble. Seven GMST
data sets and five ocean heat content records are used to compare values of
ECS computed from the EM-GC. The box represents the 25th, 50th,
and 75th percentiles, the whiskers denote the 5th and 95th
percentiles, and the stars indicate the minimum and maximum values of ECS
using the EM-GC based on the weighting method described in Sect. 2.5. The
circles denote the value of ECS associated with the best estimate of AER
RF
The box-and-whisker symbol labeled CMIP6 in Fig. 7 shows the 25th,
50th, 75th, and 5th and 95th percentiles of ECS
calculated from the output of 28 CMIP6 models, as described in Sect. 2.4.
Minimum and maximum values are represented by stars. The values of ECS
from the CMIP6 multi-model ensemble are larger than the majority of values
inferred from the climate record using the EM-GC. The height of the box for
the CMIP6 multi-model ensemble estimate of ECS is larger than the height of
the boxes for ECS inferred from the climate record using the EM-GC,
indicating that the GCMs exhibit a wide range of ECS values. The 25th
and 75th percentiles of ECS from the CMIP6 multi-model ensemble are
2.84 and 4.93
Figure 8 summarizes values of ECS found utilizing the analysis of the
century-and-a-half-long climate record using our EM-GC, our examination of a
28-member CMIP6 GCM ensemble, and 13 other recent studies. The studies are
divided into three categories: those that estimated ECS based on
observations (historical analysis), others that used GCM output but
constrained the output in some way (Constrained GCM output), and studies
that examined raw GCM output (GCM output). We obtain a best estimate for ECS
of 2.33
Values of ECS from the EM-GC (black) trained using the
HadCRUT5 GMST record, our analysis of the CMIP6 multi-model ensemble
(black), and 13 other studies grouped by type of analysis. The studies are
listed by lead author (first initial of their first name and first initial
of their last name) and the year of publication unless there are only two
authors, in which case the initials of both authors are listed. The historical
analysis includes Lewis and
Grünwald (2018) NL
Recent studies have shown that the CMIP6 multi-model ensemble exhibits
higher values of ECS than the CMIP5 models because of larger, positive cloud
feedbacks within the latest models (Gettelman
et al., 2019; Meehl et al., 2020; Sherwood et al., 2020; Zelinka et al.,
2020). The IPCC 2013 report gives a likely range of 1.5
We obtain a wide range of ECS values from our EM-GC simulations of the
climate record due to consideration of the uncertainty in the radiative
forcing of climate from tropospheric aerosols (Figs. 5c and 7). However,
under one circumstance, we find values of ECS using the EM-GC that are
similar to the maximum value of ECS from the CMIP6 multi-model ensemble. Our
large estimate of ECS occurs if we assume that anthropogenic aerosols have
exhibited strong cooling and offset a large amount of greenhouse gas
warming such that the observed GMST record can only be well simulated under
the condition of large climate feedback (i.e., values of
Five empirical determinations of ECS (our study plus Lewis and
Grünwald, 2018; Otto et al., 2013; Skeie et al., 2018; and Tokarska et al., 2020a) and the
CMIP5- or CMIP6-constrained estimates of Cox
et al. (2018), Dessler et al. (2018), and Nijsse et al. (2020) are in slight
contrast with the 2.3–4.7
In our model framework, the largest uncertainty in ECS is driven by
imprecise knowledge of the radiative forcing of climate by tropospheric
aerosols. As shown in Fig. 5c, a wide range of ECS values can be inferred
from the century-and-a-half-long climate record. We stress that each value
of ECS shown in Fig. 5c is based on a simulation for which
We conclude this section by commenting on the relationship between ECS and
AAWR in our model framework. Eight of the CMIP6 GCMs (GFDL-ESM4,
GISS-E2-1-G, INM-CM5-0, INM-CM4-8, MIROC6, MIROC-ES2L, NorESM2-LM, and
NorESM2-MM) exhibit values of ECS and AAWR consistent with the minimum and
maximum estimates based on our EM-GC constrained by the HadCRUT5 GMST record
(Table S5 and Fig. S17). An analysis of the relationship between AAWR and
ECS from the CMIP6 GCMs illustrates that 78 % of the variance in ECS among
the 28 CMIP6 GCMs that provide both quantities is explained by AAWR (see
Fig. S17). This result indicates that CMIP6 GCMs that accurately simulate the
rise in observed
The CMIP6 multi-model archive provides future projections of the GMST
anomaly relative to pre-industrial (
Historical simulations and future projections of GMST
from the CMIP6 multi-model ensemble for several SSP scenarios.
The red trapezoid in Fig. 9 labeled as the IPCC 2013 likely range is the
same trapezoid as that displayed in Fig. 11.25b from Chapter 11 of the
IPCC 2013 report (Kirtman et al.,
2013). The recent observations of
Figure 9 illustrates that there are two groups of CMIP6 multi-model projections
of
The EM-GC is also used to project future changes in
The large range of
Time series of future projections of
Probabilistic forecasts of the future rise in
Figure 12 compares probability distribution functions (PDFs) for the
projection of
Probability density functions (PDFs) for
Numerical values of probabilities for staying at or below the Paris
Agreement target for SSP1-2.6 or upper limit for SSP4-3.4 are given for the
seven GMST records using the EM-GC and CMIP6 multi-model ensemble in Table 1. Projections of
An analysis by Tokarska et al. (2020b) supports
our finding of a higher likelihood of attaining the goals of the Paris
Agreement than suggested by the CMIP6 multi-model ensemble. Tokarska et al. (2020b) filter the CMIP6 multi-model output on the level of agreement with
observations to show that the SSP1-2.6 scenario has a likely range of
warming of 1.33–1.99
The transient climate response to cumulative emissions (TCRE) relates the
rise in
Probability of achieving the Paris Agreement target (SSP1-2.6) or upper limit (SSP4-3.4) for seven GMST records using the EM-GC and the CMIP6 multi-model ensemble. The probabilities using the EM-GC are computed using the aerosol weighting method. The probabilities using the CMIP6 models are computed by calculating how many of the models for that scenario are below the temperature limits compared to the total number of models.
Total cumulative and future carbon emissions that will
lead to crossing the Paris temperature thresholds based on the EM-GC trained
using the HadCRUT5
To obtain a 66 % likelihood of limiting the rise in future
An analysis by van Vuuren et al. (2020)
assesses remaining carbon budgets based on cumulative emissions after 2010.
Their analysis indicates that only 228 Gt C can be released after 2010 to have a
66 % probability of achieving the Paris Agreement target of limiting the
rise in
Atmospheric abundances of methane will likely continue to increase as society expands natural gas production and agriculture, making it important to analyze the impact of various methane scenarios on the rise in GMST. It is unlikely that future atmospheric methane abundances will progress as indicated by SSP1-2.6 (see Fig. 2), a low radiative forcing scenario. Current observations shown in Fig. 2 illustrate that the methane mixing ratio is following SSP2-4.5 and has missed the initial decline needed to follow the SSP1-2.6 pathway. To analyze the effect varying future methane abundance pathways will have on GMST, we have generated linear interpolations of the SSP1-2.6 and SSP3-7.0 methane abundances and created four alternate scenarios (see Fig. S22), which we call blended methane scenarios. We can substitute one of the blended methane scenarios into the EM-GC instead of using the projection of methane specified by the SSP database to quantify the sensitivity of future warming to various evolutions of methane in terms of the rise in GMST.
Figure 13 shows the probability of staying at or below the Paris Agreement
target (gold) or upper limit (purple) for SSP1-2.6 (solid) and
SSP4-3.4 (dotted) as a function of the methane mixing ratio in 2100. The
lowest atmospheric methane mixing ratio value in 2100 of 1.15 ppm is from
the SSP1-2.6 methane pathway, and the highest mixing ratio in 2100 of 3.20 ppm is from the SSP3-7.0 methane pathway. The four in between are the blended
methane scenarios. As the atmospheric methane abundance increases, the
likelihood of achieving the goals in the Paris Agreement decreases. For
SSP1-2.6, the probability of limiting the rise in GMST below the
1.5
Probability of staying at or below the Paris Agreement target and
upper limit for SSP1-2.6 and SSP4-3.4 as a function of varying methane
scenarios using the EM-GC trained with the HadCRUT5
In Sect. 3.3.3, we showed that if all GHGs follow the SSP4-3.4 scenario
there would be a 64 % probability of limiting warming to 2.0
Reducing the future anthropogenic emissions of methane might be more
challenging than controlling future emissions of carbon dioxide because
methane has such a wide variety of sources related to energy, agriculture,
and ruminants (Kirschke et al.,
2013). Given the current widespread use of methane as a source of energy in
the United States and parts of Europe (Saunois et al.,
2020), combined with the continued growth in the global number of ruminants (Wolf et al., 2017), it seems
unrealistic for atmospheric methane to follow the peak and sharp decline
starting in 2025 of the SSP1-2.6 pathway (Fig. 3b). Our analysis suggests that
failure to limit methane to the SSP1-2.6 trajectory will have a larger
impact on the achievement of the 1.5
In our analysis above, we have assumed that the value of
Change in GMST from 1850–2019 for observations from HadCRUT5
(black) and 1850–2100 modeled (red) using SSP2-4.5 and a value of AER
RF
Figure 14 shows the change in observed and modeled GMST for an EM-GC
simulation training with the HadCRUT5 GMST record and using an AER RF time
series with a value of AER RF
In Figs. 14 and S23 we also analyze an RF scenario termed SSP2-4.5' that
serves as a doubled CO
We fit the climate record over the past 170 years (
Several other studies have investigated the amount of change in
Allowing
The assumption of constant feedback within the EM-GC framework used in the
rest of the paper is reasonable because there is no strong evidence
from the climate record for a noticeable increase in
In this paper we use a multiple linear regression energy balance model
(EM-GC) to analyze and project changes in the future rise in global mean
surface temperature (GMST), calculate the attributable anthropogenic warming
rate (AAWR, the component of the rise in GMST caused by human activities)
over the past 4 decades, and compute the effective climate sensitivity
(ECS, the rise in GMST that would occur with atmospheric CO
The median value of AAWR from 1975–2014 computed using our EM-GC constrained
by the century-and-a-half-long record for GMST provided by HadCRUT5 is
0.157
In our model framework, the best estimate of ECS is 2.33
We also examined the probability of limiting the future rise in GMST below
the Paris Agreement target of 1.5
We also quantify the sensitivity of the probability of achieving the Paris
Agreement target (1.5
Finally, we have also quantified in the EM-GC framework the remaining
budgets of carbon (i.e., CO
We conclude by noting that the CMIP6 multi-model ensemble provides many
useful parameters, such as sea level rise, sea ice decline, and precipitation
changes, that provide a great societal understanding of the impact of
climate change. We do not mean to undermine the importance of the CMIP6 GCMs
with this analysis. Rather, we hope that studies such as this, along with
other recent evaluations of CMIP6 multi-model output like Nijsse et al. (2020) and Tokarska et
al. (2020b), will lead to improved use of the CMIP6 multi-model ensemble for
policy decisions. Our EM-GC was built to specifically simulate and project
changes in GMST; we do not examine numerous other components of the climate
system that affect society. We emphasize that our projections show that
unless society can implement steep reductions in the emissions of carbon and methane in the next 10 years, the 1.5
All data used as inputs into the EM-GC are available from resources on the
web. We have provided the links to the resources below. The data are also
available along with the EM-GC output used in this analysis at IOD: the COBE SST data are provided by the NOAA ESRL physical sciences
division from their website at Tropospheric ozone RF: MEI.v2 and MEI.ext: PDO: SAOD: TSI: OHC records:
Balmaseda: Carton: Cheng: Ishii: Levitus: SSP database: all information for the SSPs obtained from the SSP database is
at CMIP6 input data:
CMIP6 model output archive:
The supplement related to this article is available online at:
LAM, APH, and TPC developed the model code used in this analysis. LAM, APH, and BFB collected data. RJS supervised, administrated, and developed the project. LAM wrote the original draft, and RJS, APH, BFB, TPC, and WRT participated in the review and editing of the paper.
The authors declare that they have no conflict of interest.
We would like to acknowledge the World Climate Research Programme for coordinating and promoting CMIP6 through its Working Group on Coupled Modelling. We thank the climate modeling groups participating in CMIP6 for producing and making their model results available, the Earth System Grid Federation (ESGF) for archiving the data and providing access, and the several funding agencies who support ESGF and CMIP6. This project could not have occurred without the results from CMIP6. We thank the NASA Climate Indicators and Data Products for Future National Climate Assessments program and the NOAA Cooperative Institute for Satellite Earth System Studies for their financial support of this research. We thank University of Maryland undergraduate Lauren Borgia for participating in extensive, in-depth discussions of recent papers on cloud feedback and climate sensitivity. Finally, we thank both reviewers for very careful reads of the original paper that led to substantial improvements, as well as Martin Stolpe for contacting us privately while the paper was in discussion regarding an erroneous description of the effect of creating blended near-surface air temperature that had appeared in the submitted paper.
This research has been supported by the National Aeronautics and Space Administration (grant no. NNX16AG34G) and the National Oceanic and Atmospheric Administration (grant nos. NA14NES4320003, NA19NES4320002).
This paper was edited by Christian Franzke and reviewed by two anonymous referees.