Getting the Beast to behave
Written by Arne Vollertsen
Improving the accuracy of Nordic climate predictions
Climate models, the hyper-complex calculators climate scientists use to investigate past, present and future climates, are good at showing the big picture. But we also need regional and local models with the highest possible accuracy, for researchers to work on and for decision-makers to use. That is what the NICEST2 project wants to accomplish for the Nordic region – and that involves unleashing The Beast and then getting it to behave.
To predict the long-term effects of climate change, researchers have developed digital twins of Earth’s climate. These Earth System Models, or ESMs, many of them developed over decades, numerically solve sophisticated equations. The computing power required to run these models is immense, and so large supercomputers are the only practical computational resources for ESM research. ESMs are able to represent human influence on climate, for instance by deforestation and greenhouse gas emissions, and they are crucial tools for climate scientists. By simulating the interactions between winds, clouds, oceans, land surface, ice etc. ESMs create What-If scenarios of the future:
What will happen if the world’s energy generation sector switches focus to renewables and low fossil carbon energy systems? What will happen if we carry on with business as usual?
These projections of different futures serve as an important basis for political decision-making, internationally as well as nationally and locally. However, seen from a Nordic point of view, the most dominant Earth System Models do have limitations. In short, they only care about the big picture. They have a tendency to be very large scale, and while being very effective at providing global climate change information they fall short, when you ask them to detect impacts on the Nordic countries at regional and local scale.
For instance, measurements show that the Arctic is warming faster than current climate model scenarios predict. Also, the dominant ESMs are not particularly well suited to predict the effects of decreasing permafrost, or how vegetation is affected by the increasing shifts in sun and cold spring weather in the region.
Tweaking the models
Nordic climate scientists need to tweak the ESMs to better simulate the Nordic region, and this is where the NICEST2 project comes in. NICEST is an acronym for Nordic Collaboration on e-Infrastructures for Earth System Modelling, and the project is continuing previous efforts by NeIC to facilitate access to computing power, infrastructure and services for Nordic climate scientists. The long-term goal of the initiative is to strengthen the Nordic position within climate modelling and to create better and more precise predictions for the region.
Looking inside an ESM
To demonstrate the shortcomings of current ESMs, as seen from a Nordic point of view, and how to remedy them, we need to look under the hood of an ESM.
An Earth System Model is a digital simulation of Earth’s climate, based upon actual observations and measurements taken over a long period of time. It includes not only a simulation of the physical processes regarding winds, clouds, oceans etc. It also simulates the global carbon cycle, atmospheric chemistry and other factors, enabling the ESM to predict how a changing climate will affect the concentration of greenhouse gases and aerosols – which will in turn feedback on the climate.
Many models in one
Actually, an Earth System Model is not one model, but a set of models stitched together. Climate scientists are highly specialized, and they work independently on creating the best possible model of one specific component of the climate system, for instance the atmosphere, the ocean, land (hydrology, vegetation), sea ice etc. At some point these components are merged together with a piece of software called a Coupler, handling the interaction between the model components.
This is where you can get in trouble. When you try to run the model you could be “unleashing the Beast” as some climate scientists prefer to put it.
The individual components may work very well, as they are very constrained. But put together they create a Beast that can produce results that are completely off the chart. This is because tiny anomalies in one corner of the model can reverberate through the entire simulation, just like the proverbial butterfly flapping its wings and creating a tornado on the other side of the globe.
Tuning is needed
To get the Beast to behave, so to speak, you have to do some tuning. First you feed the model with real-life measurements from the past. Then you tune it, adjusting parameters and turning knobs to achieve the best agreement between the observations and the output of the model.
As the ESMs are developed for other regions, for instance the US, they are tuned for those regions, and not for the Nordics. That means they may produce poor predictions for the Nordic region.
At this point it is important to state that an ESM will never be able to deliver 100 percent precision. Even the most advanced ESMs produce climate predictions for different scenarios with some level of uncertainty, mainly for two reasons. Firstly, there is much we do not know about Earth’s climate, for instance how clouds develop at high altitude. Secondly, because it would be too costly to run otherwise, the resolution of ESMs must be limited, which in turn requires the use of a number of assumptions or parameterisations.
Better Nordic fit
However, much can be done to improve the accuracy of Nordic climate predictions, and that is one of NICEST2’s goals. The project is helping the Nordic climate research community develop tools to tune ESMs to better fit the Nordics, enabling researchers to deliver deeper insights into climate change impacts on Nordic environment and society. This Nordic tuning will increase the accuracy of the models in regards to for instance the small, intense low-pressure systems which form in the winter over Arctic waters, called Polar Lows, and the Greenland Block, a high-pressure zone that has grown more powerful in recent years.
Making HPC access easier
Also, NICEST2 wants to make access to supercomputing resources easier for researchers.
High performance computing is essential for running ESMs, some of them run for months on the most powerful machines available, just to produce a single simulation. Large research groups in the US and the EU have access to all the resources, technical support and specialist knowledge they need to efficiently run their models on big machines. Researchers in smaller countries are less fortunate and left to their own devices, having to do more of the technical work themselves. The NICEST2 project aims to lift that workload and to provide technical support, optimize workflows and make services developed for e.g. high-energy physics, available for the Nordic climate research community.
That also includes preparing for the future, helping climate researchers adapt their ESMs to the new generation of European supercomputers that will be available soon. Just as ESMs continue to develop, so do supercomputers. In Europe, the EuroHPC initiative is building a European infrastructure of next-generation pre-exascale supercomputers consisting of eight HPC sites. One of them is the LUMI supercomputer in Kajaani, Finland, and the NICEST2 technicians will support Nordic researchers who want to run their models on LUMI. And when that work is done there will be new challenges ahead. Beyond LUMI, exascale systems are coming, and NICEST2 wants to position the Nordic climate science community to utilize these machines as well.