Today’s climate models trace their lineage to the global circulation models of the 1950s. The core equations are the same, but the algorithms that implement them have evolved, and scientists have taken advantage of each new generation of faster computers to improve their models. The models I’ve studied often weigh in at more than a million lines of code, contributed over many years by hundreds of scientists. And they keep evolving; every climate model is a work in progress. Even the beating heart of a climate model – its “dynamical core” – gets replaced every once in a while. In this chapter, we’ll examine one model in particular, the UK Met Office’s Unified Model, and explore its dynamical core and the design decisions that shaped it.

How certain can we be about projections of future climate change from computer models? In 1979, President Jimmy Carter asked the US National Academy of Science to address this question, and the quest for an answer laid the foundation for a new way of comparing and assessing computational models of climate change. My own work on climate models began with a similar question, and led me to investigate how climate scientists build and test their models. My research took me to climate modelling labs in five different countries, where I interviewed dozens of scientists. In this chapter, we will examine the motivating questions for that work, and explore the original benchmark experiment for climate models – known as Charney sensitivity – developed in response to President Carter’s question.

Climate models are often presented as tools to predict future climate change. But that’s more a reflection of the questions that politicians and the general public ask of the science, rather than what the science does. Climate scientists prefer to use their models to improve our understanding of the past and present, where more definitive answers are possible. Predicting the future is notoriously hard. It requires some careful thinking about what can be predicted and what cannot. On this question, early experiments with climate models led to one of the most profound scientific discoveries of the twentieth century – chaos theory – which gave us a new understanding of the limits of predictability of complex systems. The so-called butterfly effect of chaos theory helps explain why a computer model can predict the weather only for a few days in advance, while the same model can simulate a changing climate over decades and even millennia. To find out why, read on!

Climate and weather are intimately connected. Weather describes what we experience day-to-day, while climate describes what we expect over the longer term. So it’s not surprising the models used to understand weather and climate share much of the same history. While Arrhenius’s model ignored weather altogether, focusing instead on the energy balance of the planet, modern climate models grew out of the early work on numerical weather forecasting – the basic equations for how winds and ocean currents move energy around, under the influence of the Earth’s rotation and gravity. The equations for these circulation patterns were first worked out by Arrhenius’s colleague, Vilhelm Bjerknes, in 1904, but it wasn’t until the invention of the electronic computer that John von Neumann put them to work forecasting the weather. The approach developed by von Neumann’s group now forms the core of today’s weather forecasting models.

Our confidence in climate science depends to some extent on our confidence that the models are valid. But a computational model can never be perfect, because a model cannot capture everything. What do we mean by “valid”? In this chapter, we will examine how climate modellers test their models, and what they do when they find errors. One surprising result is that climate models appear to be less buggy than almost any other software ever produced. More importantly, climate modellers have adapted the tools of science – hypothesis testing, peer review, and scientific replication – in remarkable new ways to overcome the weaknesses in any individual model, to ensure their scientific conclusions are sound. A study of the Max Planck Institute for Meteorology, in Hamburg, Germany will show how they do this.

Scientists have been building computational models of the climate and studying the consequences of our use of fossil fuels for more than a century. In the twenty-first century, these consequences are all around us, and the need for urgent action has become clear. In this chapter, we show how experiments with climate models give us a clear picture of the choices we face, and how the climate system will respond to those choices. We’ll show how advice from climate models shapes policy targets, such as the 2°C limit and goal of reaching net-zero emissions. In the political arena, scientific advice has to compete with many other sources of information and misinformation, which has slowed meaningful action, so we’ll also examine the political processes by which we collectively make decisions, and the role each of us plays in those processes. Ultimately, climate models can guide us on how to tackle climate change, but only if we find the wisdom to understand and act on that guidance.

The first computational climate model was built by a Swedish scientist in 1895, 50 years before the invention of programmable electronic computers. All the calculations had to be done by hand. Despite this, the model shares many similarities with today’s computer-intensive climate models, and is a good introduction to how modern models work. The model also provided the first ever prediction of how our use of fossil fuels would lead to global warming, but that wasn’t why it was built. So what was it built for, and were its predictions any good?

Climate models provide a “virtual laboratory” for experiments that cannot be done on the real planet. A large community of scientists runs their own experiments on these models, and collects observations from the real world to see if these match what happens in the experiments. In climate science, the sharing of computer models helps build such a community by making it easier for scientists to develop their own experiments. Shared models also remove any mystery around an experiment, because the program code in the model is a precise, unambiguous statement of what the experiment does. The development of such a community was a crucial step forward for climate science, and, as we shall see, the National Centre of Atmospheric Research, in Boulder, Colorado played a central role in the development of this community.

To fully understand climate change – its causes and consequences – you need a grasp of many different fields of science. Bringing together multiple experts can be hard because researchers are increasingly specialized, don’t understand each other’s jargon, and aren’t encouraged to explore how their knowledge inter-relates. But in climate science, computational models overcome these barriers. Today’s climate models are assembled from many pieces, built by different research groups, each capturing a different aspect of the overall climate system. This isn’t easy – like a jigsaw puzzle where the pieces weren’t designed to fit together. But once the pieces are assembled, the models support a new kind of collaboration. They allow scientists from very different fields to combine their knowledge to answer big questions, and work together on shared experiments. In this chapter, we’ll explore this process of coupling climate models, find out why it’s so challenging, and meet another of our case studies, the Institut Pierre Simon Laplace (IPSL) in Paris, France.