Predicting – not producing – climate change
Interview with Prof Thomas Ludwig – German Climate Computing Centre
Thomas Ludwig is the lord of FLOPS and bytes: The computer scientist heads the Deutsches Klimarechenzentrum (DKRZ) in Hamburg, which is Germany’s central body for researchers who seek to understand how our climate changes in the long term. His supercomputers can compute and store enormous amounts of data – and also deliver detailed simulation results to scientists at the HZG.
"Meanwhile, we have been able to calculate several components in parallel. Experts at the HZG study air currents so they can come to conclusions about the future development of wind and waves." Photo: HZG/ Christian Schmid
Mr Ludwig, Germany is one of the leading nations in climate research. Does science owe this status in part to your Climate Computing Centre, the DKRZ?
In certain way, yes – even if we only provide the infrastructure for this success. In this branch of research, many questions simply cannot be addressed without the immense computing capacity of high performance centres like ours.
Why is that?
Because researchers must take into account highly complex global developments in, for example, extremely diverse regions such as the Arctic and the Sahara. And they must do so over periods of at least thirty years – only then can we talk about climate in a scientifically meaningful way. Many scientists, however, also choose to look at considerably longer periods, even back to the last ice ages. If the calculation results match other data – for example, those from ice cores – this is a clear indication of their robustness.
When the Computing Centre managed to calculate two developments at the same time in one climate model twenty-five years ago – the oceans and the atmosphere – this was considered a decisive step toward the DKRZ’s success. How many factors can models at your centre integrate today?
Considerably more than the two components back then. Depending on the project, the researchers now take into consideration, for example, the growth of algae in a marine region or an area’s agricultural use. Experts at the Helmholtz-Zentrum Geesthacht examine air currents so that they can draw conclusions about future development of wind and waves. Today we can also take into consideration climate change dynamics much better than before. There is hardly a value in our work that remains constant over the years. In some cases, developments are mutually reinforcing: if a forested region, for example, shrinks as a result of a region’s temperature increase, it possesses fewer trees to produce oxygen or absorb CO2. This leads to an even more pronounced change in climate. By now, we can simulate these cycles in our models.
"We rank thirty-four out of the five hundred highest performance computers globally." Photo: HZG/ Christian Schmid
This requires an incredible amount of computing capacity – how has this capacity developed at the DKRZ since its founding thirty years ago.
Worldwide computing capacity has increased by a factor of one thousand every twelve-anda-half years, and the DKRZ is consistent with this trend. Since its founding in 1987, the capacity has increased altogether by approximately fifteen million. The third floor of our centre is home to approximately 100,000 processor cores – today these machines can handle more than three quadrillion computations per second! We rank thirty-four out of the five hundred highest performance computers globally. These components are housed in seventy-nine cabinets, each as large as a phone booth and weighing one ton. Installing the equipment in our building was half the challenge for our structural engineers. What’s more is that our data archive is one of the largest in the world.
It stores as much data as you’d find in 135,000 laptop computers. The data is accessible online to scientists all over the world. In hindsight, you can see how rapidly the computing capacity at our centre has developed: even our first supercomputer ranked as one of the largest in the world. It, however, only had one processor at the time and had the capacity of one of today‘s smartphones.
"Our present computer consumes even less power than its predecessor – though its capacity is twenty times greater." Photo: HZG/ Christian Schmid
Do you expect similar development in the future too?
At least for the coming decades. Even our next computer, which will go into operation in 2020, is expected to increase our computing power tenfold. It is again financed by the Helmholtz Association, and this time by the Max Planck Society and the city of Hamburg as well.
You place great emphasis on the energy efficiency of your computers. As you’re constantly increasing your computing capacity, however, you face an unavoidable dilemma: more capacity requires more energy, right?
Surprisingly, no. Our present computer consumes even less power than its predecessor – though its capacity is twenty times greater. This is attributed to a special cooling system that enables us to dispense with fans in the computer nodes. Water pipes cool the processors instead. The water heats up to fifty degrees and is then directed to the roof of our building. Here it cools down and flows back into the circuitry again. What is special in this system is its high heat tolerance. It even works with warm water at forty degrees, which is why we don’t need to cool the liquid as much as with other computer models. This works for us even in summer without cooling units, though of course the comparably mild weather here in Hamburg also plays a role. We save altogether 300,000 Euros or more per year in energy costs. In addition, we only use green energy – we do, after all, want to predict climate change, not produce it. If, however, we can achieve a similarly good energy balance with the next computer remains to be seen. Computer technology changes so quickly that I can’t say what model we’ll start operating in 2020.
Photo: HZG/ Christian Schmid
As a computer scientist, how were you drawn to climate research?
Scientific computing always fascinated me, those complex models developed to aid science. While they are purely mathematical constructs, they aren’t scientific gimmicks; these simulations mostly contribute to branches of science that are of basic importance to all human beings. In medicine, for example, where I began my career, we used models to calculate the course of illnesses such as cancer.
In climate research, too, computer programs are particularly relevant. On the one hand, they forecast developments that aff ect many people, such as the land becoming uninhabitable or soils no longer yielding crops due to climate change. But on the other hand, scientists in this discipline are heavily under attack. This is why IT must provide particularly reliable data. It’s why I feel challenged! This is, for example, why what are known as “ensemble computations” are standard: the same process is computed twenty times, again and again, with slightly altered initial conditions to ensure that biased values aren’t used. We only consider our outcome as valid when our models provide clear results in those calculations. This is why I can say with full conviction that climate research works with extremely reliable data. I have still yet to see any other science that puts such effort into quality assurance.
Are you able to see the climate research trends in your research applications?
Absolutely, especially the extent to which the topics have now diversified. The range of the several hundred projects here is enormous. On the one hand, scientists look at vast periods of time or huge regions while, on the other, they look at representations of fragile structures such as clouds, the movements of which we only reproduce over a period of a few minutes. We are not only more flexible temporally, but also spatially. Earlier, our computing capacities were only sufficient for grids of five hundred kilometres edge length. We could only take into consideration three measurement points for all of Germany at the time. Research has progressed much further here.
Scientists from the Helmholtz-Zentrum Geesthacht, for example, are closely observing individual regions in Europe and Germany and are precisely reproducing those surface areas with a resolution of up to ten kilometres. This is important because climate change can have a very different eff ect within a country – on the coast, for example, with more storms and flooding while inland the soils are drying up.
You are regularly confronted with some very distressing research results as director of the German Climate Computing Centre. Does that frustrate you sometimes?
Of course I’m not immune to the results; they’re often too alarming. But I seek a professional distance. Sometimes the disaster scenarios can be even impressively beautiful: I recently viewed a tornado simulation that could have come from a Hollywood blockbuster. And then I think to myself that we need to work here with precision and quality standards – and collect evidence of what will happen if we do nothing.
The interview with Prof. Dr. Thomas Ludwig was conducted by science journalist Jenny Niederstadt at the DKRZ.
Published in in2science #4 (June 2017)