Of weather forecasts, climate models, why supercomputers are stretched to their limits and the consequences of these limitations: A meeting with Thomas Schulthess, Director of the Swiss National Supercomputing Centre (CSCS)

I was supposed to meet Thomas Schulthess at the CSCS in Lugano this past March. But then the COVID-19 pandemic happened, the CSCS staff withdrew to working from home, and the Centre was no longer accessible for visitors. Now, three months later, Thomas Schulthess and I sit at a table in an adjoining room of a restaurant at his place of residence, Wädenswil, near Zurich. He and his team are still working from home, even though the lockdown was annulled some time ago. Obviously, we start off our conversation by talking about the coronavirus crisis.

«All in all, we have passed well through this crisis,» Schulthess tells me. «We had introduced hygiene and distance rules already at the end of February, and operation continued without constraints, as we are able to conduct most of our work from home. All 120 members of staff are well, and no one got infected.»

What turned out to be a much greater challenge than coronavirus, he explains, was the cyberattack in May, which affected almost all supercomputers in Europe. «It was aimed at the heart of the computer centre. Apart from an interruption of operation of our supercomputer ‹Piz Daint›, however, the attack did not cause any damage; but it showed us the weaknesses in our systems, and we have learned,» he adds, before we switch to the actual subject of our meeting: The role of supercomputing in the field of weather and climate forecasts and the development of supercomputers.

Weather forecasts with an accurate resolution of one kilometre

For more than ten years, weather forecasts for Switzerland have been calculated by a dedicated supercomputer at the CSCS. The Swiss national weather and climate service, MeteoSwiss, sends the data from its measuring stations to Lugano, where they are mathematically processed in a model and entered into the computer. According to the specifications of the weather model software, this computer calculates the development of weather data for the next three to five days, from which the meteorologists from MeteoSwiss subsequently deduce their forecasts.

However, the collaboration between the CSCS and MeteoSwiss comprises much more than mere computing services. In mixed teams consisting of meteorologists, information scientists and computer specialists, the two federal institutions jointly implement the weather models as well as the software and the matching computing infrastructure.

«Together with MeteoSwiss, we implement the models on the computers and we also implement the computers. This means that we do not buy the computers off the shelf, but rather build the components in collaboration with the manufacturer in such a way that they are perfectly tailored to the model calculations and work processes that we carry out for MeteoSwiss,» Schulthess explains. Moreover, the European weather services and the supercomputing centres also collaborate closely for the development of these weather models, as they are updated every three to four years or replaced by new, more powerful ones.

I ask Schulthess where he sees today’s great challenges in terms of weather computation. He names two: The growing demand for increasingly accurate, local weather forecasts and the rising costs of the supercomputers.

«An example: Zurich Airport asks for more accurate fog forecasts. Today, MeteoSwiss is able to predict the weather in a model resolution of one kilometre, which is sufficient for thunderclouds in the Alpine region, but not for fog,» Schulthess says. «Here, we need a resolution of 100 metres, and in order to get it, the computer must perform a thousand times better. This means that our computer has to either calculate for a longer time or we have to buy computers that are a thousand times faster. We do not have the time for extended calculations, as Zurich Airport needs to know today about the fog condition’s development over the next hours.»

Climate predictions for the next 10 years

In addition to the meteorologists of MeteoSwiss, the climate researchers of ETH Zurich, too, rely on the supercomputers of the CSCS to calculate their simulations. Especially since the climate change debate became so prevalent, climate scientists are under great pressure. Society and politics are asking with increasing urgency for reliable predictions about regional and local climate changes in order to react with appropriate measures.

«The climate models today are able to reliably predict the means of climate warming in 10 or 100 years. However, for laypersons, it is rather difficult to understand when experts state that the temperature will rise globally by one or two degrees on average. But if my climate colleagues tell me that Central Europe will look like North Africa in a couple of decades, and that the Alps will resemble the Atlas Mountains, then I can imagine the results,» Schulthess says. «Unfortunately, we cannot make such predications today. This is why ETH Zurich’s climate scientists, such as Christoph Schär and others, work very hard to enhance their climate models so that their climate simulations reach a resolution of one kilometre, as it is already possible with today’s weather models.»

Actually, the climate researchers use the same, although somewhat adapted, models as the meteorologists. But unlike the weather forecasters, whose time horizon comprises a few days, climate research deals with climate simulation over a period of 10 to 100 years. Like the meteorologists, the climate researchers collaborate closely with the CSCS specialists to develop the models and software, and also to build the supercomputers. A team of climate scientists is currently working with the CSCS on defining the requirements concerning the new supercomputer that will one day replace the current high-performance computer «Piz Daint», on which the climate simulations are presently running. This new computer will ideally be able to simulate the climate conditions in 10 years for every location on the planet with a resolution of one kilometre. The Director of the CSCS is confident that they will succeed.

«There is plenty of work to be done,» Schulthess admits, «both in numerical form and in terms of climate science. However, if we understood it all properly and if everything goes well, we will be able to calculate a model trajectory (the course of the global climate over 10 years) within a reasonable amount of time by the mid-2020s here in Lugano. Yet, climate researchers require about 50 trajectories to correctly interpret the development of the climate statistically. They therefore need 50 supercomputers, and we cannot cover this all by ourselves. Fortunately, we conduct these calculations together with our European partners.»

Calculating for Europe

In December 2019, the President of the European Commission, Ursula von der Leyen, launched the «Green Deal». The initiative’s overarching aim is to reduce the EU’s greenhouse gas emissions to net zero by 2050, transforming Europe’s economy and society in a climate-neutral and sustainable manner. In order to reach this objective, the Commission fully relies on digitisation and Big Data, meaning all globally available data relevant to climate is to be digitised and made accessible for analyses and predictions about climate development. The European supercomputing centres such as CSCS play a key role in this ambitious plan, as they possess knowhow and experience in configuring, building and operating the supercomputers necessary to calculate those mega simulations.

Currently, institutions such as national weather services and top researchers of big universities are the primary users of the supercomputers at European computation centers. It is the European Commission’s aim that, in the future, more researchers from smaller institutions are introduced to and have access to the possibilities and services of the new generation of supercomputers. This is the only way to obtain as many new findings as possible in so little time from the flood of digitised data. To that end, the Commission launched, among others, the ESiWACE Project (Excellence in Simulation of Weather and Climate in Europe). The current ESiWACE2 programme brings together 20 climate and weather research institutions, weather services and supercomputing centres from all over Europe, among them MeteoSwiss and the CSCS.

«We are contributing our knowhow and our software products so that the climate researchers and weather people of our European partners can use this software as well,» Schulthess says. «And then we will soon have these new climate models that enable simulations with a resolution of one kilometre. In order to calculate them, we need the new exascale supercomputers. We know how to build and operate them, and we will help the researchers to use them in the future.»

According to Schulthess, the project directly benefits the CSCS. «Thanks to ESiWACE,» he explains, «We can introduce researchers of local universities to the new technologies and our infrastructure. We thus promote young talents and broaden the base of researchers who are able to use our supercomputers.»

However, Schulthess also firmly believes that the benefits are shared on all sides. «The CSCS is one of the most competent supercomputing centres in Europe,» he says. «By participating in ESiWACE, we contribute substantially to the understanding of climate simulation development. The EU benefits from the CSCS, both in technological and in scientific terms, but we also wish to introduce the findings won to the wider public. The political debate addressing the challenges of climate change is conducted within a European context, and the EU hereby plays a key role. Climate does not know about borders, and Switzerland is not an island. This, we have also just been taught by coronavirus.»

The future of supercomputing

Towards the end of our meeting, Schulthess talks about how the digitisation of our society affects supercomputing. «We will need more high-performance computers. And as Moore’s Law is no longer valid, we will need to generate more electricity from renewable energy resources.» What this means in practice, however, is complicated.

For example, he says, «If I need one gigawatt of electricity for a supercomputing centre in the densely populated Central Europe, I take this gigawatt away from society, and it will be costly and politically unsustainable. This is why we must build the computers where electricity can be produced in a cheap and environmentally friendly way. Today, this is only still possible in Northern Europe, which is sparsely populated and where hydropower can be used for electricity generation and cooling. We will have to bring the computers and the data to this power, not the other way around. This is why the CSCS is already collaborating in a consortium with the experts from Finland, Norway and Iceland.»

Looking into the future, the Director of the CSCS envisions an increasingly collaborative role for the centre.

«Currently, we possess a high-performance infrastructure with twelve gigawatts, and we are able to extend it maximally by a factor of two through intelligent investments. But after this, we will have reached our limit,» Schulthess says. «This means that the CSCS can no longer fulfil its task alone. It must consider itself as part of a European group and cultivate specialties within this network. I can see us collaborating with our colleagues from ETH Zurich Computer Science and Electrical Engineering Departments for the development and building of prototypes and pilot systems that are capable of, for example, calculating a model trajectory in a climate simulation. The reproduction, scaling and building we will then realise together with our European partners on computers somewhere in Northern Europe,» Thomas Schulthess replies. Then he is called to the next video conference, back in his office at home.