It all started with the weather – and the bomb. When the first precursors of today’s computers started to appear in research laboratories in the late 1940s – the so-called Von Neumann calculators – it was about better weather forecasts on the one hand, and, on the other, about calculating neutron scattering which provided the basis for the construction of atomic bombs. In the 1960s, the physicist Syukuro Manabe programmed the first climate model considered realistic by researchers, with which he prepared his prediction of man-made climate change, published in 1967: In the event of a doubling of CO₂concentration in the atmosphere, Manabe forecast that temperatures would increase by about 2.5 degrees Celsius. The Japanese was awarded the Nobel Prize in Physics for his work in 2021.

The University of Bern also played an important role in the change of era that marked the introduction of computer models in climate research from the 1960s onwards. The Bernese nuclear physicist Hans Oeschger had developed a method of reconstructing the CO₂content in the atmosphere of past millennia from ice cores. “Knowledge about past climate changes was extremely valuable for the young climate models, as they were only able to be calibrated with past data,” writes science historian Dania Achermann (UniPress 179, February 2020).

Simulations – a success story

Today, computer models have become an integral part of climate science: Whether for calculations by the Intergovernmental Panel on Climate Change (IPCC) on the future of the global climate, a study on the particularly high mortality rate during the sweltering summer of 2022 in Switzerland or a study on the future risk of forest fires in the Canton of Bern, computer models or, more precisely, simulations are needed everywhere. The method has established itself not only in climate research, but in most areas of science. “The idea of simulating something with a computer is a success story that began, among other things, with modeling the atmosphere,” explains Claus Beisbart, Professor of Philosophy of Science at the University of Bern and member of the Oeschger Centre for Climate Change Research. “Simulations have opened up new doors, certain things couldn’t be calculated beforehand.”

The basis of all computer simulations is a set of mathematical equations – in the case of climate models, these are the so-called Navier-Stokes equations for describing atmospheric dynamics. This mathematical system was developed in the mid-19th century, but solving its differential equations is so complex that it could only be done efficiently with the help of computers.

Testing hypothesis after hypothesis

Computer simulations have long since been used to describe not only physical phenomena, but also biological phenomena, for instance. Christoph Schwörer, for example, works with a dynamic vegetation model. He is a postdoctoral researcher in the Paleoecology Group of the Oeschger Centre. The model he uses is called LandClim and was developed at ETH Zurich to simulate the importance of climate effects, forest fires and forest management on the dynamics of forests. Schwörer investigates how climate change and human activities affected vegetation in the past.

For this purpose, he analyzes pollen, macrofossils and charcoal particles found in the sediments of lakes. In this way, past changes can be reconstructed, but why they came about often remains unclear. “With our computer model,” explains Schwörer, “we can test different hypotheses.” For this purpose, only one of the possible influencing variables, such as fire, climate or clearing, is changed, while the others remain constant. “When playing through these possibilities, changes in the landscape are revealed that match our data,” says Schwörer, “that speaks for the hypothesis tested.”

Data alone is not enough

Geoscientist Frerk Pöppelmeier is also reliant on computer simulations. “In my field of research, data alone is no longer enough,” he emphasizes. “To get a little closer to the truth, you have to link measurement results with models.” As a postdoctoral researcher in the group “Earth System Modelling: Climate Dynamics” at the Oeschger Centre, Pöppelmeier found the environment he was looking for to advance his research. His theme: Past changes in the Atlantic circulation, which he reconstructs from marine sediment cores using geochemical proxies (indirect indicators of climate from natural archives). On the basis of this data, he then executes simulations using the “Bern3D” model developed at the University of Bern. “Using the model, we are investigating how interfering factors affect Atlantic circulation and trying to reconcile this with the reconstructions.”

Building your own models

Benjamin Stocker not only uses computer models, he also builds them himself and develops existing models further. His aim is to improve models so that they can realistically reflect the impacts of climate change on terrestrial ecosystems. To do so, they must be able to simulate, for example, the effect of nutrients on the global carbon cycle, explains the head of the research group “Geodata and Earth Observation” at the Oeschger Centre. Today’s models do so only very inadequately. Benjamin Stocker and his team are investigating how the effects of climate change – for example, on plant growth – can be expressed in mathematical equations. Observations and experiments show that plants adapt dynamically to benefit from the rising CO₂concentration: The relationship between newly formed leaves and roots, for example, shifts. In order to be able to absorb more nutrients, plants produce more roots.

“This pattern is very important for simulating future developments in the global carbon and nutrient cycle,” explains Benjamin Stocker. “We want to show how coupled Earth system models can be improved in this direction.” By the way: The modular developments of the group “Geodata and Earth Observation” are open access, i.e. they are openly made available to the entire research community, like most climate models.

As the three examples from the Oeschger Centre show, digitalization in climate research is well advanced. Computer simulations have long been an established tool; recently, some researchers have also been using artificial intelligence, especially machine learning. But even if computer models have become an integral part of research, one fundamental question remains: How trustworthy are the research results obtained in this way?

Let’s ask the philosopher of science, who was a physicist in his first career. Claus Beisbart, why can we believe computer models? “The credibility of computer simulations is based on two things. Firstly, the laws of nature or theories, and secondly, the fact that the simulations are tested with the help of data.” According to Beisbart, the more established the theories underlying a simulation are, the more credible they are. And: Simulations must be able to reproduce past climate. They are validated by the fit between the simulation output and data from the past. “The credibility of simulations is based on reflecting mechanisms we understand,” concludes Claus Beisbart.

Digitalization also harbors risks

Have the decades of work with computer simulations also had an impact on scientific practice? Absolutely, says the science philosopher. “New forms of cooperation have emerged.” Simulations are often so extensive that they can no longer be carried out by small groups, which has encouraged cooperation. Parts of program codes are shifted back and forth today. Working alone, as in the past in theoretical physics, is unthinkable in many computer simulations.

Speaking of program codes: Claus Beisbart sees traditional scientific career paths at risk due to this specialization in computer models. Many young researchers are in demand today mainly as programmers, but in this role it is difficult to develop your own profile with publications. Scientific careers cannot be built in this way, he points out.

Scientific creativity remains key

In the philosophy of science, however, much more fundamental thought is given to computer simulations. What can we really learn from this? The outcome of experiments is open, and that’s what makes them interesting, says Beisbart: “Nature’s answer.” In the case of computer simulations, on the other hand, the result is already included in the model assumptions. These can only be as good as the assumptions themselves. From this the physicist and philosopher deduces two things. First of all, computer simulations cannot completely replace the experiment. And secondly, sometimes you have to change the basic assumptions of a simulation. It is not possible to deduce from the simulation how this is to be done. “Developing a new theory requires a form of creativity that goes beyond simulation.”