top of page

New era of Climate Modelling and High-Performance Computing. Where did it all begin?

“What’s the use of having developed a science well enough to make predictions if, in the end, all we’re willing to do is stand around and wait for them to come true?”― F. Sherwood Rowland


I think nothing says more than a famous quote by the Nobel Laureate and professor of chemistry Frank Sherwood Rowland. In the World where more than 3.6 billion people are highly vulnerable to the adverse effects of climate change (IPCC, 2023) we too often function as passive observers rather than catalysts for change. Continued emission of greenhouse gases (GHG) will lead to an inevitable increase in atmospheric temperatures. If we want to achieve our goal of keeping the warming below 1.5 C° relative to preindustrial levels we have to start acting now. But how can we decide which strategy will work best and yield the most results? How can we understand what factors drive climate change? 


What is a climate model? 


Development and advance of climate modeling over the last century opened a window into the future for us. Today it is vitally important not only for climate change predictions but also for global economies. It plays a key role in facilitating development of the adaptation and mitigation strategies to ongoing climate change. 

Once a matter of science fiction, it has now become a reality. Predicting tomorrow's weather or foreseeing long–term climate change based on present patterns is no longer out of our reach. Our ability to model future climate shifts we owe to the expertise of a remarkable pair of scientists – the Norwegian geophysicists Vilhelm and Jacob Bjerknes, a father–and–son duo. 




Fig. 1 Left: Vilhelm Bjerknes and Right: Jacob Bjerknes. Images from Wikipedia.org.


Vilhelm Bjerknes revolutionized the classical view of particle motion that was previously believed to be only dependent on pressure changes.  He was the first to propose a link between changes in density and heat, giving birth to physical fluid dynamics (A. Eliassen, 1982). This allowed Vilhelm to formulate a set of hydrothermodynamic equations to understand atmospheric forcing, linking the flow of heat, water, and air. This was the first step towards weather forecasting. He suggested to use real–time observations to describe the initial state of the system and integrate a set of mathematical physical equations to predict future changes (V. Bjerknes, 1904). Following his father's footsteps his son, Jacob, was working on the theory behind the movement of the cold and warm air masses. He formulated the frontal cyclone model, where fronts represent boundary surfaces separating the cold and warm air of the cyclone (J. Bjerknes, 1918). Today, it is widely accepted as a "polar front theory." In the following years, new members joined the team at the University of Bergen, Norwegian and Swedish meteorologists Halvor Solberg and Tor Bergeron (A. Eliassen, 1995). They continued working on Jacob's model. Combined effort resulted in the birth of the first four–dimensional working model of a mid–latitude cyclone life cycle (J. Bjerknes and H. Solberg, 1922). Jacob's frontal cyclone model has proven to be an extremely useful tool in weather forecasting, providing all the necessary information about the winds, temperature, clouds, and precipitation, allowing deduction of the future behavior of the cyclone based on its shape and stage of development (J. Bjerknes, 1919). Years later, Jacob's attention was captured by the new rapidly advancing field of research: the interaction between the atmosphere and the sea. He went on to study sea surface temperature (SST) changes in the North Atlantic Ocean, showing that annual SSTs are explained by air–sea fluxes associated with wind alteration (J. Bjerknes, 1964). Five years later he published another pivotal study about the El Niño phenomenon in the Pacific Ocean, establishing the link between El Niño and the southern oscillation (J. Bjerknes, 1969). Jacob's research was fundamental for the theory of climate. Modern–day climate change models are built on the results of his research (A. Eliassen, 1995).


So how does a climate model work? 


Climate models also known as general circulation models (GCM), are complex computer programs that follow the laws that govern the Earth’s climate. These laws are represented as a set of partial differential equations describing physical, biological, and chemical processes within the atmosphere, ocean, land, and ice-covered regions. In the model, they are translated into several hundred thousand lines of code written in Fortran programming language. Fortran is the common language used in climate modeling developed by John Backus for IBM in 1950 (John Backus, 1981). Models capture the flow of air and water in the atmosphere and the ocean as well as the transfer of heat. Using observational data as an input, scientists analyze interactions of different factors (air temperature, pressure, density, water vapor content, wind magnitude, etc.) with the atmosphere, land, the ocean, and sea ice. Derived observations are then applied to simulate future system responses (R. McSweeney and Z. Hausfather, 2018).


To describe the processes and mechanisms that drive the climate of our planet, Earth is divided into a 3D grid of numerous cells (hundreds or even thousands) stacked on top of each other and extended across the surface of the planet. The application of the grid allows personalized input for each cell that can account for differences in terrain or atmospheric layers, reducing model uncertainties. The smaller the size of the grid cell the higher the model resolution. The higher the spatial resolution, the more accurate the results. However, an increase in resolution also increases the computing time for the same simulation period, requiring more powerful computers. Even an increase by a factor of two in the model resolution will require a 10 times increase in the computing power (R. McSweeney and Z. Hausfather, 2018). 





Fig. 2 The schematic representation of the Global Atmospheric Model. Each box represents a grid cell. The smaller the cell size is, the higher is the model resolution. Source NOAA [https://www.gfdl.noaa.gov/climate-modeling/].


Hampered by limitations in computing power, early model simulations were only capable of looking at one aspect of the Earth’s climate at a time, atmosphere, ocean, or land. Technological advancement in high computing capabilities allowed us to gradually integrate different components of the stand–alone models into one model providing a comprehensive representation of the climate system. Further integration of the biosphere components and their interactions into the climate models allowed simulation of the important elemental cycles (carbon, nitrogen), chemistry of the atmosphere, ocean ecology, and changes in vegetation and land use (R. McSweeney and Z. Hausfather, 2018). This was crucial for our understanding of how the climate responds to humancaused greenhouse gas emissions.  


To model the impact and interplay of population, economic growth, and energy consumption on the physical climate, scientists employ Integrated Assessment Models (IAMs, UNFCCC). IAMs are commonly utilized to forecast future greenhouse gas emissions, predict climate-related consequences, and assess the advantages and drawbacks of potential policy measures designed to address these issues (R. McSweeney and Z. Hausfather, 2018).


However, climate models have limitations. 


Even though the atmosphere is represented by a vast number of grid cells, many physical processes take place on scales smaller than the resolution allows to resolve. To adequately model subscale processes that play a crucial role in climate, they must be simplified or parameterized. This means that their values are defined in the computer code rather than being calculated by the model itself (R. McSweeney and Z. Hausfather, 2018). For example, the emission of greenhouse gases will unavoidably affect atmosphere temperatures altering water vapor content and movement of air masses. This could lead to changes in cloud cover. Clouds themselves can either trap or reflect sunlight, thus warming or cooling the climate. To estimate how the cloud cover is expected to behave in certain situations, scientists have to use other mathematical equations that mimic real–world impacts. Other small–scale processes like the exchange of heat and water vapor with the oceans also require parameterizations (R. McSweeney and Z. Hausfather, 2018). The degree and complexity of parametrization among different models vary greatly as some parameters correspond to well–defined physical and chemical values (forces, reaction rates, etc.), whereas others are less well-defined (mixing rates, cloud erosion rates, critical threshold values). To find the best suitable value that reflects the climate most closely, scientists run tests on the models that are referred to as tuning or calibration (Hargreaves, 2010; Hourdin et al., 2017; Mauritsen, et al. 2012; Schmidt et al., 2017). Therefore, there is often variation among model projections. For this reason, each model's projection is typically referred to as a "plausible future" (R. McSweeney and Z. Hausfather, 2018). Depending on how the atmosphere responds or how human behavior changes, the climate could go down any of the paths that models project. If all models agree on the same changes, that is the most likely outcome. This is the reason why there are disparities between available climate prediction models, and some of them are generally "cooler" (the rate of warming is less) than the others (Hausfather et al., 2019).


To improve the models' output, in 1995 researchers started a Coupled Model Intercomparison Project (CMIP) with the participation of many research Institutes around the World. Using a coupled atmosphere ocean model different modeling groups would run the same various experiment designs to better understand how climate changes and to identify potential uncertainties in order to improve the models [WCRP, 2022]. Consistent experiment designs among all models ensure that any disparities in modeled outcomes came from the model itself helping to identify what caused the difference. More than a thousand of papers all over the globe have been already published on the results of the model intercomparison. Every 5 to 6 years a new generation of experiments is designed.


To validate the models scientists perform historical runs, where the models are run over the historical time period for which the observation data exist. Instead of forcing the model to fit the observed data, scientists let the model derive the outcome that further is compared with the recorded data. If the model output closely predicts the observed data (hindcast), it provides more confidence in the model’s ability to predict future changes. A recent comparison of the existing models shows that they predicted the observed warming trends reasonably well (Hausfather et al., 2019). 


What does HPC tell us about climate change? 


In the modern world where we are confronted by adverse impacts of climate change, models offer essential insights into the potential consequences for the planet's future. Applying the observational data about such external factors (forcings) like GHG, aerosol particle concentration emitted when burning fossil fuels, forest fires, or volcanic eruptions we can predict how it will affect the amount of heat absorbed by the earth’s atmosphere (R. McSweeney and Z. Hausfather, 2018). 


Running different greenhouse gas emission scenarios, scientists can simulate changes in future temperature and humidity within the atmosphere layers, and predict potential changes in precipitation patterns (rainfall and snowfall). Models can provide more information on how climate change can affect ocean conditions, like temperature, salinity, pH, and extent of ice cover. Using gained information as input for other models we can predict how these changes may affect natural resources like clean water, forests, crop productivity, or biodiversity (R. McSweeney and Z. Hausfather, 2018).


Presently, the majority of climate change model predictions rely on the "Representative Concentration Pathways" (RCPs) that incorporate socioeconomic and emission scenarios detailing the growth and development of global society (ICCP, 2014). These scenarios serve as the base for the assessment of negative human impacts and associated costs. Used as an input into models they help us devise viable mitigation and adaptation strategies. Each representative concentration pathway covers a nearly full range of emission scenarios available in the current scientific literature. In these pathways concentration, rather than emission is used as an input into the model. Four pathways were devised that each predicted a different temperature increase in the 21st century: RCP2.6 (0.9-2.3 Co), RCP4.5 (1.7 – 3.2 Co), RCP6.0 (2.0-3.7 Co) and RCP8.5 (3.2-5.4 Co) (van Vuuren et al., 2011).





Fig. 3. (a) Representative Concentration Pathways and corresponding annual carbon dioxide emissions (coloured areas show a 5 to 95% range). The emission scenarios underlined by the WGIII categories are based on the most recent scientific literature and are defined based on CO2-eq concentration levels (in ppm) in 2100. (b) Depict the increase in global mean surface temperature as a function of the given cumulative total global CO2 emissions. Coloured plume shows the spread of the data projections. Ellipses depict total human–induced warming reached by the end of 2100 versus cumulative CO2 emissions from 1870 to 2100 under the WGIII scenario categories. The filled black ellipse depicts emissions recorded up to 2005 and temperatures recorded between 2000–2009 with associated uncertainties. Figure source IPCC Report for Policy Makers, (2014).


These pathways predict potential climate futures from the present into the year 2100 and beyond. Different RCP scenarios from high emission scenario to low emission scenario are expected to provide invaluable information for the socioeconomics variables and mitigation options as well as adverse effects of climate change.


HPC and Machine Learning


Despite significant advancements in modern climate models in recent years, they remain constrained by computing power, with only very large supercomputers capable of handling high–resolution global–scale climate modeling. To further improve model accuracy and resolution more observational data are needed. Certain regions of the planet are still insufficiently studied, posing challenges in interpreting model output due to a lack of comparable data (R. McSweeney and Z. Hausfather, 2018).


Future improvements in computing power will bolster the development of higher–resolution models with finer details, allowing for the improvement of the accuracy of the model projections. The increased computing capacity will reduce computational costs, making model runs more time–efficient and facilitating the provision of up–to–date information to policymakers for the formulation of effective climate change adaptation plans.


In recent years, scientists have harnessed the capabilities of artificial intelligence and machine learning to enhance climate models. These technologies enable more efficient analysis of the data and identification of patterns. The insights gained from these techniques hold the potential to refine climate models even further (R. McSweeney and Z. Hausfather, 2018).


References


Backus J. (1981). The History of Fortran I, Ii, And Iii. History Of Programming Languages, https://www.softwarepreservation.org/projects/FORTRAN/paper/p25-backus.pdf


Bjerknes J. (1918). On the Structure of Moving Cyclones. Publisher Unknown.


Bjerknes J. (1919). On The Structure Of Moving Cyclones. Monthly weather review. https://doi.org/10.1175/1520-0493(1919)47<95:OTSOMC>2.0.CO;2


Bjerknes J. and H. Solberg (1922). Life Cycle of Cyclones and the Polar Front Theory of Atmospheric Circulation. Geophysical Publication, vol 3, 1.


Bjerknes J. (1964). Atlantic Air-Sea Interaction. Advances in Geophysics, 10,  1-82.


Bjerknes J. (1969). Atmospheric Teleconnections from the Equatorial Pacific. Monthly Weather Review, 97,3. https://doi.org/10.1175/15200493(1969)097<0163:ATFTEP>2.3.CO;2


Bjerknes V. (1904). Das Problem der Wettervorhersage, betrachtet von Standpunkt der Mechanik und Physik. – Meteorol. Z. 21, 1–7.


Eliassen A. (1982). Vilhelm Bjerknes and his students. Annual Review. Fluid Mechanics, 14:1-11.


Eliassen A. (1995). Jacob  Aall   Bonnevie Bjerknes 1897—1975 A Biographical Memoir by. National Academy of Science.


Hargreaves, J.C. (2010). Skill and uncertainty in climate models. WIREs Climate Change, 1: 556-564, https://doi.org/10.1002/wcc.58


Hourdin, F., Mauritsen, T., Gettelman, A., Golaz, J.-C., Balaji, V., Duan, Q., Folini, D., Klocke, D. J. D., Qian, Y., Rauser, F., Rio, C., Tomassini, L., Watanabe, M., and Williamson, D. (2017). The art and science of climate model tuning, B. Am. Meteorol. Soc., 98, 589–602, https://doi.org/10.1175/BAMS-D-15-00135.1


IPCC (2000). IPCC Special Report on Emissions Scenarios. Prepared by Working Group III of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge, United Kingdom, and New York, NY, USA.


IPCC (2014). Climate Change 2014 Synthesis Report Summary for Policymakers.


IPCC (2023). Summary for Policymakers. In: Climate Change 2023: Synthesis Report. Contribution of Working Groups I, II and III to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change [Core Writing Team, H. Lee and J. Romero (eds.)]. IPCC, Geneva, Switzerland, pp. 1-34, doi: 10.59327/IPCC/AR6-9789291691647.001

Mauritsen, T., et al. (2012). Tuning the climate of a global model, J. Adv. Model. Earth Syst., 4, M00A01, doi:10.1029/2012MS000154


McSweeney R. and Hausfather Z. (2018). How do climate models work? https://www.carbonbrief.org/cmip6-the-next-generation-of-climate-models-explained/

National Oceanic and Atmospheric Administration (NOAA), https://www.gfdl.noaa.gov/climate-modeling/


Schmidt, G. A., Bader, D., Donner, L. J., Elsaesser, G. S., Golaz, J.-C., Hannay, C., Molod, A., Neale, R. B., and Saha, S. (2017). Practice and philosophy of climate model tuning across six US modeling centers, Geosci. Model Dev., 10, 3207–3223, https://doi.org/10.5194/gmd-10-3207-2017.


United Nations Framework Convention on Climate Change (UNFCCC). Integrated Assessment Models (IAMs) and Energy-Environment-Economy (E3) models. unfccc.int


van Vuuren D. P., Edmonds J., Kainuma M., Riahi K., Thomson A., Hibbard K., Hurtt G.C., Kram T.,  Krey V., - Lamarque J.F., Masui T., Meinshausen M.,  Nakicenovic N., Smith S.J., Rose S.K. (2011). The representative concentration pathways: an overview. Climatic Change, 109:5–31.


WCRP Coupled Model Intercomparison Project (CMIP) (2022). www.wcrp-climate.org



Dr. Alexandra Filippova


Dr. Filippova is a marine geochemist with an extensive and diverse scientific background in geoscience, natural management, polar marine research and climate science. Over a decade she studied processes that affect ocean circulation in the past and how they could compare to the modern day situation. One of the key questions of her studies included the role of climate induced melt water inputs in ocean circulation and climate changes on short and long term scale. Beyond academia, she truly enjoys volunteering with Non-Profit Organizations, where she advocates for diversity and inclusion of caregivers in all STEMM fields (Mothers in Science) and works on development of sustainable projects that aim at preserving nature and biodiversity (Viable Community). 



81 views0 comments
bottom of page