Global climate models
Introduction
A popular book on climate science published in 2021 pointed
to the limitations of computer simulations of the Earth’s climate and suggested
that the faster computers needed to give satisfactory results in reasonable run
times would not be available for two or three decades. This post outlines some
recent advances in climate modelling, preceded by notes on earlier work.
Early simulations of climate
Although climate models using computers existed in the
1950s, the period from 1970 to 2007 described by Hausfather et al. (2019) makes
a useful introduction. The authors analysed the projections of change in global
mean surface temperature (GMST) made by several generations of past models.
Simple energy balance models were made in the early 1970s, based
mainly on CO2 concentrations. They have been “gradually sidelined” in favour of
general circulation models first published in the late 1980s.
The two main factors which influence long-term performance
are the accuracy of the model’s physics and the accuracy of “projected changes
in external forcing due to greenhouse gases and aerosols, as well as natural
forcing such as solar or volcanic forcing.” However, the forcings due to future
emissions depend on human behaviour, technological change, economic and
population growth, and cannot be predicted by the modellers, who may instead project
a “range of forcing trajectories representative of several plausible futures”.
This must be borne in mind when evaluating the projections of climate models.
The forecasts of a new climate model cannot be evaluated until
enough time has passed to provide data for comparison; however, where suitable
historical date is available, the new model can be started from a date in the
past to make a ‘hindcast’ which can be compared with what actually happened. Early
climate models, such as those first used in the 1960s, acquire special interest
because their forecasts can be compared with decades of historical records of
temperature.
The authors “compared observations to climate model
projections over the model projection period using two approaches”. The two sets
of comparisons were for change in temperature versus time and for change in
temperature versus change in radiative forcing. The results need detailed examination,
but the authors conclude that the past climate model projections evaluated in
their analysis were skilful “in predicting subsequent GMST warming in the years
after publication.” Some overestimated and a few underestimated warming but “most
models examined showed warming consistent with observations, particularly when mismatches
between projected and observationally informed estimates of forcing were taken
into account.” While the projections of the relatively simple models in the
study are obsolete, “they may be useful tools for verifying or falsifying
methods used to evaluate state‐of‐the‐art climate models.”
More recent climate models
The need for increasingly sophisticated climate models is
outlined in an article on kilometre-scale modelling of the Earth system (Gettelman
et al., 2023). “Simple energy balance models evolved into coarse resolution
General Circulation Models, then into the coupled Earth System Models (ESMs) we
have today.” While these models can “project broad outlines of the future
climate”, they cannot project local and regional extremes of weather, drought
and inundation. The need for such local information has driven the development
of a new class of regional and global ESMs. They are known as km-scale models and
have horizontal scales of less than 10 kilometres. While such models in some
ways resemble Numerical Weather Prediction models, their longer time scales make
it necessary to model additional features such as ocean circulation, the “evolving
land biosphere, hydrosphere, cryosphere and balanced global energy and carbon
budgets.”
The km-scale models are computationally expensive: at the
time of writing 10 km global models could project for centuries, “3 km global
atmosphere or ocean models alone for a few years, and 1 km global atmosphere
models for only a few days”. The authors predicted that three-km global
simulations at the centennial scale “will be increasingly feasible for many
within the next five years.”
A short article entitled A New Generation of Models for
Kilometer-Scale Climate Projections described advanced climate models under
development at the time of publication (MPI, 2024). While the mains trends in global
climate are described as “unequivocally established”, the likelihood of events
such as extreme rainfall, drought and heatwaves could not be modelled at the
local level. A new generation of high-resolution models uses smaller spatial
scales than those of traditional approaches, potentially allowing climate
science to bring local granularity to climate information globally, resolve yet
unexplained changes due to global warming, perform simulations on the scale of
observations, and represent processes that are not at present simulated. A
short list of projects “building towards high-resolution climate projections” follows;
they are nextGEMS, WarmWorld, DestinE, EERIE, and a set under the SCALEXA
heading.
The Next Generation Earth System Models (nextGEMS) project is
dedicated to developing storm-resolving Earth system models, with a grid
resolution down to 2.5 km – “a dramatic improvement over current models”. The
project overview published in 2025 refers to the aim of achieving “multidecadal
climate simulations at kilometre-scale resolution, representing interactions
between the atmosphere, ocean, and land in far greater detail than conventional
climate models.” (nextGEMs 2026). Computational speed on the Levante supercomputer,
which began operations in 2022, reached about 500 simulated days per day. The
computer can perform 14 quadrillion mathematical operations per second.
The WarmWorld project aims to address the uncertainty
associated with climate impacts on small scales by advancing the high-resolution
capabilities of the existing Earth System Model ICON. Its 2025 report
highlighted advances in kilometre-scale modelling, an exascale-ready modelling
system (one quintillion calculations per second), improved workflows and data
handling, and enhanced computational efficiency (WarmWorld, 2026).
DestinE is a project to produce a digital twin of the Earth
that “would provide answers to questions we might have about its future”
bringing together “decades of advancements in weather prediction, Earth
observations, high-performance computing and climate science” (DestinE, 2026)
EERIE: Earth-System Models “are developed to capture the
dynamics of certain turbulent flow patterns, so-called ocean eddies” through “simulations
in which ocean eddies are explicitly represented by the laws of physics” using the
biggest super computers in Europe in the most energy efficient way (EERIE, 2026)
SCALEXA covers a range of projects aimed at improving high-performance
computing, artificial intelligence and data analytics. Taken together the
projects “form a comprehensive effort of advancing climate science.” An example
is ADAPTEX, “An open-source software framework for exascale-capable flow
simulations on dynamic adaptive grids and its application in Earth system modelling”
(Adaptex, 2026).
The above projects show that computer science is advancing faster
than was suggested by the pessimistic forecast quoted in the introduction to this
post. However, climate modelling has not solved all its problems. In the
concluding discussion of their paper on Earth System Models, Simpson et al. (2025)
list some of the “potential priorities and future opportunities for the climate
science community over the next few decades.” More signs of anthropogenic influence
are likely to emerge in parts of the climate system and will need to be
accounted for. It will be essential to maintain and improve long- term climate
observational records, in order to monitor the emergence of climate change,
quantify decadal variability, and evaluate models. Climate modelers should “work
closely with observational experts to develop new datasets that would be
particularly useful for climate model evaluation.” Uncertainty estimates should
be documented “with language that is accessible to nonspecialists” and “new
techniques should be explored to take advantage of more existing historical
observations to reconstruct additional climate records back in time to provide
longer- term context”. The model-
observational comparison cycle should be more frequent than the present 7-year
cycle.
Perhaps most importantly,” improved communication is needed
between climate analysts and model developers to ensure that model development
choices are made with a view toward understanding, and ultimately reducing,
known discrepancies in trends between models and the observational record.” New
opportunities will present themselves in areas such as “evolving eco system
demography, the global carbon cycle, ice sheets, and glaciers” which have “large
societal importance, but have traditionally been absent or poorly represented
in models.”
Global coupled climate models
Global coupled climate models simulate the Earth's climate
system by modelling the interactions between different subsystems, such as the
atmosphere, oceans, land, and ice. Comparing the performance of different
models can lead to improvements and “support national and international climate
change assessments” (Wikipedia, 2026). The project aims to improve climate
models and support national and international climate change assessments. The
Coupled Model Intercomparison Project (CMIP) was initiated in 1995 and there
have been several subsequent phases, the current phase being CMIP7.
Brunner et al. (2026) discuss “the evolution of model
performance from the beginning of the Coupled Model Intercomparison Project
(CMIP) in the 1990s to the latest kilometer-scale models today.” The authors
single out the IFS-FESOM model as outperforming “even the best CMIP6 models”
but note “the considerable efforts still needed to realize” the full potential of
kilometer-scale models. (IFS-FESOM: the Integrated Forecasting System coupled
to the Finite volumE Sea Ice-Ocean Model). The value of comparing models across
generations is stressed, and the study draws on “an extensive archive of 176
models, developed over three decades and five model generations”. Methods of
evaluation are described and show a continuous improvement in performance over the
period studied. Limitations and barriers to progress are also discussed in
detail.
The role of emulators
“Researchers at the University of Bristol have developed a
new method which could help scientists perform large-scale climate simulations
at a fraction of the cost and time needed compared to traditional climate
models” (Bristol, 2026). The research described in this news item specifically
relates to modelling climate within the Quaternary period. Despite the speed of
modern computing, modelling hundreds of thousands of years can still take years
of real time. The researchers used a powerful climate model “to train a much
faster statistical emulator that can reproduce the behaviour of the full model.”
The emulator “can simulate climate
changes across the entire Quaternary period in just minutes and run on a
standard laptop.”
The paper on which the news item is based (Williams et al.,
2026) referred to General Circulation Models (GCMs) which “require substantial
computational resources, meaning they are unsuitable for exploring
orbital-scale variability on million-year timescales.” The researchers used a GCM “to calibrate a
faster statistical model, or emulator”. They found a “good agreement between
the emulated climate and proxy data over the last 800,000 years”.
Computational models which predict Earth’s future climate typically
address changes over hundreds of years, and so do not face the same run time
issues as those modelling entire geological periods. Nevertheless, it is
interesting to ask whether emulators also have a place in modelling climate in
the shorter term. The post therefore concludes with brief extracts from two more
papers on emulators published this year.
Womack (2026) describes Earth System Models (ESMs) as “our
most comprehensive tools for projecting future climate impacts across the land,
ocean, and atmosphere” but notes that their “extreme computational costs limit
their ability to survey the vast space of potential emissions trajectories.” He
believes that climate emulators “are poised to fill this scenario-assessment
gap” but notes the many remaining questions “around their theoretical
underpinnings, physical consistency, and ultimate utility for areas like impact
assessment”.
Van Katwyk (2026) describes “emulators designed for use as
surrogate models for Earth system components: statistical or machine
learning-based models trained to reproduce the output of climate simulators
given the same inputs.” Such emulators approximate the behaviour of a “limited
set of measures from complex models, but do so millions to billions of times
faster with high accuracy and flexibility.” They enable scientists to quickly
run experiments and are transforming the relationship that scientists have with
climate models: “what was once a static set of hard-won results from a
simulator can now be probed and queried for new scientific insights.” Climate
emulation has evolved over the last decade “from simple approximations … to
complex, machine learning (ML)-based emulators” which are “shifting the
bottleneck of scientific discovery from computation to training
data–potentially drawn from both simulators and observations.”
References
Adaptex, 2026, An open-source software framework for
exascale-capable flow simulations on dynamic adaptive grids and its application
in Earth system modelling, online, accessed 27 August 2026
https://www.dlr.de/de/sc/forschung-transfer/projekte/adaptex
Bristol, 2026, New method to carry out long term climate
simulations at fraction of cost, News and Features, online, accessed 31 August
2026
Brunner, L. et al., 2026, Three decades of simulating global
temperature patterns with coupled global climate models, Nature, Communications
Earth & Environment, online, accessed 22 August 2026
https://www.nature.com/articles/s43247-026-03497-w
DestinE, 2026, Destination Earth (DestinE) - digital model
of the earth, online, accessed 27 August 2026,
https://digital-strategy.ec.europa.eu/en/policies/destination-earth
EERIE, 2026, European Eddy-Rich Earth-System Models, online,
accessed 27 August 2026, https://eerie-project.eu/
Gettelman A., et al, 2023, Kilometre-Scale Modelling of the
Earth System: A New Paradigm for Climate Prediction, World Meteorological
Organisation, online, accessed 22 August 2026
Hausfather, Z. et al., 2019, Evaluating the
Performance of Past Climate Model Projections, Geophysical Research Letters,
online, accessed 22 August 2026
https://agupubs.onlinelibrary.wiley.com/doi/pdfdirect/10.1029/2019GL085378
MPI, 2024, A New Generation of Models for Kilometer-Scale
Climate Projections, Max Planck Institut für Meteorologie, online, accessed 22
August 2026
nextGEMs, 2026, nextGEMS overview paper marks a milestone
for future European climate research, online, accessed 27 August 2026,
Simpson, I.R. et al., 2025, Confronting Earth System Model
trends with observations, Science Advances Review, Climatology, online,
accessed 22 August 2026
https://www.science.org/doi/pdf/10.1126/sciadv.adt8035
Van Katwyk, P., et al., 2026, Rewiring climate modeling with
machine learning emulators, Nature Communications Earth & Environment,
online, accessed 31 August 2026
https://www.nature.com/articles/s43247-026-03238-z
WarmWorld, 2026, WarmWorld at Three Years: Advancing
km-Scale Earth System Modelling, online, accessed 27 August 2026,
https://www.warmworld.de/news/3yr_progress/
Wikipedia, 2026, Coupled Model Intercomparison Project,
online, accessed 29 August 2026
https://en.wikipedia.org/wiki/Coupled_Model_Intercomparison_Project
Williams, C., et al., The relative role of direct orbital
forcing versus CO2 and ice feedbacks on Quaternary climate, Nature
Communications, online, accessed 31 August 2026
https://www.nature.com/articles/s41467-026-70750-3
Womack, C.B. (2026): On the theory and optimal design of
emulators for climate impact assessment. PhD Thesis, MIT Department of
Aeronautics and Astronautics, online, accessed 31 August 2026
Comments
Post a Comment