For over 20 years, the Massachusetts Institute of Technology Earth System
Model (MESM) has been used extensively for climate change research. The model
is under continuous development with components being added and updated. To
provide transparency in the model development, we perform a baseline
evaluation by comparing model behavior and properties in the newest version
to the previous model version. In particular, changes resulting from updates
to the land surface model component and the input forcings used in historical
simulations of climate change are investigated. We run an 1800-member
ensemble of MESM historical climate simulations where the model parameters
that set climate sensitivity, the rate of ocean heat uptake, and the net
anthropogenic aerosol forcing are systematically varied. By comparing model
output to observed patterns of surface temperature changes and the linear
trend in the increase in ocean heat content, we derive probability
distributions for the three model parameters. Furthermore, we run a
372-member ensemble of transient climate simulations where all model forcings
are fixed and carbon dioxide concentrations are increased at the rate of
1 % year

Equilibrium climate sensitivity (ECS), the equilibrium global mean surface
temperature change due to a doubling of atmospheric carbon dioxide
concentrations, is a climate system property that has been widely studied and
strongly influences future climate projections. One of the complexities of
ECS is that it is a function of many feedbacks and processes that act on
different spatial and temporal scales. In particular, the lapse rate, water
vapor, cryosphere, and cloud feedbacks play especially critical roles in
determining the climate sensitivity

One class of studies estimates ECS directly from observations using a global
energy budget approach

Transient climate response (TCR) provides a second metric for estimating
future climate change and is defined as the global mean surface temperature
change at the time of carbon dioxide (

One EMIC that has been extensively used in studies estimating ECS and TCR is
the climate component of the Massachusetts Institute of Technology (MIT)
Integrated Global Systems Model

In the past, “IGSM” has been used to reference both the fully integrated
model as well as the standalone Earth system component. We follow this
convention and refer to the older version of the Earth system model as IGSM,
and we refer to the updated version of the model as the MIT Earth System
Model (MESM). In this study, we provide a transparent method of testing and
accounting for how the simulated behavior and probability distribution
functions change in response to the recent model development. We derive a new
joint probability distribution by closely following the methods of

In Sect.

The climate component of the updated MIT Earth System Model

The radiation code takes into account major greenhouse gases (

Three model parameters that impact the climate system response are easily
modified in MESM. These parameters are the equilibrium climate sensitivity
(ECS), the effective ocean diffusivity (

We now highlight two major updates made between the current version of MESM
and its predecessor. The first update was the incorporation of a new land
surface model. The Community Land Model (CLM) version 3.5

Parameter pairings where the models have been run. Points in black are common to both the IGSM and MESM ensembles. Blue points are unique to the IGSM ensemble and red points are unique to the MESM ensemble.

In this section, we present an outline of the methodology used to derive the
joint probability distribution function (PDF) for the model parameters and
highlight the changes implemented between this study and previous studies
using IGSM. We follow closely the methods of

The goodness-of-fit statistics for each pattern used to evaluate the model
are converted to a PDF using the likelihood function described in

We make two changes to the methodology of

As a second change, we reduce the number of diagnostics used to evaluate
model performance. In general, independent temperature patterns should be
used to evaluate model performance because they rule out different regions of
the parameter space for being inconsistent with the observed climate record.
In particular,

Our results are presented as follows. We first identify the changes in the
input forcings used in our historical simulations by comparing the solar and
ozone components used in the IGSM runs with those used in the MESM runs.
Second, we show how the probability distribution functions change when
reducing the number of model diagnostics from three to two through the
omission of the upper-air diagnostic. Third, we derive probability
distributions using the MESM ensemble and directly compare them to those
derived using the IGSM ensemble using the full ensembles and the case where
only runs with

To identify changes in the forcing time series used to drive the model, we
compare the input forcings for the two components for which we have changed
datasets. When comparing the forcing time series, only differences in the
changes relative to 1860 impact the historical simulations. Time-invariant
differences are accounted for in the offline

Annual mean total solar irradiance. The bias between the

Ozone concentration in the old IGSM time series (red) and the

We observe that ozone concentrations estimated from the AC&C/SPARC dataset
differ in both space and time when compared to the previous concentrations
used with IGSM (Fig.

With the input forcings documented, we focus on deriving probability
distributions for the model parameters. We first test the impact of omitting
the upper-air diagnostic. As noted in Sect.

The 90 % confidence intervals for climate sensitivity (ECS) and
net aerosol forcing (

^{1}

^{2}

^{3}

^{4}

^{5}

Starting from the distributions calculated in

The 90 % confidence intervals and means for climate sensitivity
(ECS), ocean diffusivity (

We next evaluate the impacts that changing the model from IGSM to MESM and
updating the forcing suite have on the parameter distributions. We present
the new marginal distributions for each parameter in Fig.

Marginal probability distribution functions and TCR cumulative
distribution functions (CDFs) derived from
MESM simulations using the HadCRUT2, HadCRUT3, NCDC, GISTEMP 250, and GISTEMP
1200 surface temperature datasets as observations:

Marginal probability distribution functions derived from the full
IGSM (dashed) and MESM (solid) ensembles using the HadCRUT2, HadCRUT3, NCDC,
GISTEMP 250, and GISTEMP 1200 surface temperature datasets as observations:

To test whether the differences observed in the parameter estimates were due
to the model update, rather than the increased density of model runs, we
subsampled each ensemble at the 480

Observed and simulated global mean surface temperature anomalies. The observed time series (red) are derived from each of the five surface temperature datasets used in the surface temperature diagnostic. Also shown are the time series for each MESM simulation (black). Runs with parameter settings closest to the median values from each distribution are highlighted (blue). All anomalies are calculated with respect to the 1906–1995 climatology used in the surface diagnostic.

Histogram of linear trends in the 0–3 km global mean ocean heat content estimated from each MESM ensemble member. The observed trend (red) and trends estimated from the MESM simulations with parameter values closest to the medians from each distribution (blue) are shown as vertical lines.

To further demonstrate the total effect of changes to the model, forcings,
and ensemble design, we compare the marginal distributions derived from the
full IGSM and MESM ensembles using each surface temperature dataset
(Fig.

Because the parameters are estimated jointly, identifying the causes for
specific changes in the marginal distributions is not always
straightforward. With this caveat, we now present reasons for the observed
changes in the parameter distributions. We begin with

An explanation similar to that used for the aerosol distribution can be
applied to explaining the observed shifts in the climate sensitivity
distribution. In its most basic sense, climate sensitivity is a temperature
change per unit forcing. When holding the temperature patterns fixed, the
change in temperature is a constant. When explaining the aerosol distribution
above, we implicitly fixed the climate sensitivity, requiring the aerosol
forcing to be less negative to keep the net forcing constant. However, if we
fix

In practice, the model parameters are not independent of each other and can change simultaneously. Many combinations of higher climate sensitivity and weaker aerosol forcing lead to similar agreement with the observed temperature record. This suggests a correlation between these two parameters and highlights a strength of estimating the joint PDF for the model parameters: the identification of relationships between the model parameters. However, these relationships also highlight the challenge in attributing changes in a single parameter to a specific cause.

Unlike the climate sensitivity and aerosol forcing distributions, a clear
physical explanation for the observed changes in the

To evaluate how well the model captures the observed record and demonstrate
the wide range of climate states simulated by the MESM ensemble, we compare
the model output to the observed climate record (Figs.

Model response surfaces for

For both the surface temperature and ocean heat content trends, we have
sampled many climate states on the colder and warmer sides of the observed
values. We note here that the negative ocean heat content trends are the
result of simulations with strong cooling that lie well outside the
acceptable range of the parameter space. All simulations with this negative
trend have

To estimate TCR in MESM, we run a 372-member ensemble where all forcings are
held fixed and carbon dioxide concentrations are increased by
1 % year

We fit a third-order polynomial in ECS and

We use the response surface to derive probability distributions for TCR. From
each of the joint probability distributions derived using the subsampled MESM
ensemble, we draw a 1000-member Latin hypercube sample

In this study, we have provided an open, transparent means of testing the changes in model response and parameter estimation to changes in the MIT Integrated Global Systems Model framework. Not only does this systematic accounting of the impacts give a reference point moving forward for studies involving MESM, it proposes a template for assessing the impact that changes in other simplified climate models have on the calibration of their own model parameters. We hope that this study motivates other modeling groups to perform similar investigations that provide documented accounts of model updates, leading to a more robust understanding of the impacts that the changes have on parameter estimation and model behavior.

By updating the model and its input forcings, we identify the impact that the
switch from the MIT Integrated Global Systems Model to the MIT Earth System
Model has on the probability distributions of model parameters. The decreases
in radiative forcing due to the change in radiative forcing code, the new
solar radiation data, and the new ozone concentrations used to estimate the
ozone forcing lead to a net energy deficit when compared to the replaced
forcings. This drives an upward shift in our estimates of the 90 %
confidence interval for climate sensitivity from between 1.2 and
5.3

Because TCR is independent of the input forcings, the only difference between
the IGSM and MESM configurations in the transient simulations is the land
surface model. By showing that the transient climate response surfaces
derived from the two models differ only slightly, we provide evidence that
the switch to CLM3.5 does not greatly impact the temperature evolution in the
model. We have drawn Latin hypercube samples from the parameter distributions
to provide estimates of TCR from the new response surface. Due to the shift
towards higher climate sensitivity and slightly weaker ocean diffusivity, we
observe an increase in our 90 % confidence interval of transient climate
response from 0.87–2.31

The source code of MESM will become publicly available
for non-commercial research and educational purposes as soon as a software
license that is being prepared by the MIT Technology Licensing Office is
complete. For further information, contact mesm-request@mit.edu. A working
paper describing and evaluating the MESM is available at

AGL and APS carried out the MESM simulations. APS wrote the codes for extracting model output. AGL performed the analysis and prepared the original manuscript. AGL and CEF developed the model ensemble and experimental design. AGL, CEF, APS, and EM all contributed to interpreting the analysis and synthesizing the findings.

The authors declare that they have no conflict of interest.

This work was supported by U.S. Department of Energy (DOE), Office of
Science, under award DE-FG02-94ER61937 and other government, industry and
foundation sponsors of the MIT Joint Program on the Science and Policy of
Global Change. For a complete list of sponsors and U.S. government funding
sources, see