106
to different realizations of an ensemble simulation, all of which are initialized with
different but equality realistic initial conditions; “i” refers to a completely different
method for initializing a particular GCM simulation; and, “p” denotes some perturbation to GCM model physics. The string r*i1p1 appears in the vast majority of
CMIP5 files; examination of the 112 r*i1p1 runs provides a robust examination of
GCM output.
The first method used to extract AAWR from each GCM run, REG, involves
examination of de-seasonalized, globally averaged, monthly mean values of ΔT
from each run, from 1950 to 2010. Archived model output from the historical and the
future run files has been combined. Both the historical and future runs were designed
to use realistic variations of total solar irradiance (TSI) and stratospheric optical
depth (SOD), the climate relevant proxy for major volcanic eruptions. First, regression coefficients for TSI, SOD, and ΔT
HUMAN
are found. For this first step, observations of TSI and SOD are used in the analysis, and ΔT
HUMAN
is approximated as a
linear function. The regression coefficient for TSI is saved. A second regression is
conducted using ΔT from the GCM, for the 1979–2010 time period. For the second
regression, the saved value for the TSI coefficient is imposed, leading to new values
for the coefficients that modify SOD and ΔT
HUMAN
. A two step method is needed to
properly determine the TSI and SOD coefficients, because the two major volcanic
eruptions that took place over the period of interest, El Chichón and Mount Pinatubo,
occurred at similar phases of the 11 year solar cycle. The initial regression starts in
1950 to allow coverage of enough solar cycles for extraction of the influence of solar
variability on GCM-based ΔT to be found, and also because ΔT
HUMAN
over 1950–
2010 found using EM-GC (i.e., Human Rung on the Figs. 2.4, 2.5, 2.9, and 2.10
ladder plots) is nearly linear over this 60 year time frame. The value of AAWR using
REG is the slope of ΔT
HUMAN
, recorded for each of the 112 GCM runs in Table 2.3.
The second method used to extract AAWR from each GCM run, LIN, involves
analysis of global, annual average values of ΔT from the various GCM runs. As
noted above, these GCM runs were designed to simulate the short-term cooling
caused by volcanic eruptions, such as El Chichón and Mount Pinatubo. The volcanic imprint from most of the GCM runs is obvious upon visual inspection: archived
ΔT tends to be smaller than neighboring years in 1982, 1983, 1991, and 1992. For
LIN, we find the slope of global annual average ΔT from each GCM run using linear regression, excluding archived output for the four years noted in the prior sentence. Values of AAWR found using LIN are also recorded for each of the 112 GCM
runs in Table 2.3.
We are confident AAWR has been properly extracted from the archived GCM
output. Neither of our determinations attempt to discern the influence on GCMbased ΔT of natural variations such as ENSO, PDO, or AMOC. While the CMIP5
GCMs represent ENSO with some fidelity (Bellenger et al. 2014), and changes in
heat storage within the Pacific ocean simulated by GCMs has been linked to variability in ΔT on decadal time scales (Meehl et al. 2011), these effects should appear
as noise that is averaged out of the resulting signal, since our estimates of AAWR
are based on analysis of 112 archived GCM runs. While GCMs might indeed have
internally generated ENSO events or fluctuations in ocean heat storage that affect
2 Forecasting Global Warming
to different realizations of an ensemble simulation, all of which are initialized with
different but equality realistic initial conditions; “i” refers to a completely different
method for initializing a particular GCM simulation; and, “p” denotes some perturbation to GCM model physics. The string r*i1p1 appears in the vast majority of
CMIP5 files; examination of the 112 r*i1p1 runs provides a robust examination of
GCM output.
The first method used to extract AAWR from each GCM run, REG, involves
examination of de-seasonalized, globally averaged, monthly mean values of ΔT
from each run, from 1950 to 2010. Archived model output from the historical and the
future run files has been combined. Both the historical and future runs were designed
to use realistic variations of total solar irradiance (TSI) and stratospheric optical
depth (SOD), the climate relevant proxy for major volcanic eruptions. First, regression coefficients for TSI, SOD, and ΔT
HUMAN
are found. For this first step, observations of TSI and SOD are used in the analysis, and ΔT
HUMAN
is approximated as a
linear function. The regression coefficient for TSI is saved. A second regression is
conducted using ΔT from the GCM, for the 1979–2010 time period. For the second
regression, the saved value for the TSI coefficient is imposed, leading to new values
for the coefficients that modify SOD and ΔT
HUMAN
. A two step method is needed to
properly determine the TSI and SOD coefficients, because the two major volcanic
eruptions that took place over the period of interest, El Chichón and Mount Pinatubo,
occurred at similar phases of the 11 year solar cycle. The initial regression starts in
1950 to allow coverage of enough solar cycles for extraction of the influence of solar
variability on GCM-based ΔT to be found, and also because ΔT
HUMAN
over 1950–
2010 found using EM-GC (i.e., Human Rung on the Figs. 2.4, 2.5, 2.9, and 2.10
ladder plots) is nearly linear over this 60 year time frame. The value of AAWR using
REG is the slope of ΔT
HUMAN
, recorded for each of the 112 GCM runs in Table 2.3.
The second method used to extract AAWR from each GCM run, LIN, involves
analysis of global, annual average values of ΔT from the various GCM runs. As
noted above, these GCM runs were designed to simulate the short-term cooling
caused by volcanic eruptions, such as El Chichón and Mount Pinatubo. The volcanic imprint from most of the GCM runs is obvious upon visual inspection: archived
ΔT tends to be smaller than neighboring years in 1982, 1983, 1991, and 1992. For
LIN, we find the slope of global annual average ΔT from each GCM run using linear regression, excluding archived output for the four years noted in the prior sentence. Values of AAWR found using LIN are also recorded for each of the 112 GCM
runs in Table 2.3.
We are confident AAWR has been properly extracted from the archived GCM
output. Neither of our determinations attempt to discern the influence on GCMbased ΔT of natural variations such as ENSO, PDO, or AMOC. While the CMIP5
GCMs represent ENSO with some fidelity (Bellenger et al. 2014), and changes in
heat storage within the Pacific ocean simulated by GCMs has been linked to variability in ΔT on decadal time scales (Meehl et al. 2011), these effects should appear
as noise that is averaged out of the resulting signal, since our estimates of AAWR
are based on analysis of 112 archived GCM runs. While GCMs might indeed have
internally generated ENSO events or fluctuations in ocean heat storage that affect
2 Forecasting Global Warming