The European Union is finalizing plans for a “digital twin” of planet Earth, which would simulate the atmosphere, ocean, ice and land with unparalleled accuracy, providing forecasts of floods, droughts and fires days and even years in advance.
Destination Earth, as the project is called, will also try to capture human behavior, allowing leaders to see the impact of weather events and climate change on society and assess the effects of different climate policies.
“It’s a really bold mission,” said Ruby Leung, a climate scientist at the Pacific Northwest National Laboratory of the US Department of Energy (DOE).
By rendering the planet’s atmosphere at a scale of only 1 kilometer wide, much more accurate than existing climate models, Destination Earth can base its forecasts on more detailed real-time data than ever before.
The project, which will be described in detail in two workshops later this month, will start next year and will run on one of the three supercomputers that Europe will implement in Finland, Italy and Spain.
What does the Destination Earth project entail?
Destination Earth has risen from the ashes of Extreme Earth, a proposal led by the European Center for Medium-Distance Weather Forecasting (ECMWF) for a flagship billion-dollar research program, which was eventually canceled.
However, interest in such projects remained. Fears that Europe was left behind by China, Japan and the United States in terms of supercomputers led to the European High-Performance Computing Joint Undertaking, an investment of 8 billion euros to lay the foundations for possible “exascale” machines capable of 1 billion calculations per second.
Typical climate models run at resolutions of 50 or 100 kilometers; even the top ones like the “European” model of the ECMWF run at 9 kilometers.
The 1-kilometer resolution of the new model will allow it to directly transform convection – the vertical heat transport critical for the formation of clouds and storms – instead of relying on an algorithmic approximation.
“I call it the third dimension of climate modeling,” says Bjorn Stevens, a scientist at the Max Planck Institute of Meteorology.
The high resolution will allow Destination Earth to base its forecasts on more detailed data. Weather models do not do well at all when it comes to temperature and pressure observations from satellites, weather stations, aircraft and buoys to guide their simulations.
Why such an effort?
The aim is to enable decision-makers to directly assess how climate change will impact society – and how society could change the trajectory of climate change.
For example, the model could predict how climate change will affect agriculture and migration patterns in Brazil – and also how reductions in ethanol subsidies could limit deforestation in the Amazon.
Currently, climate scientists extract regional results from global climate models and pass them on to experts in agriculture or economics to understand the effects on human behavior.
Now, says Erin Coughlan de Perez, a scientist in the field of climate hazards at the Red Crescent Climate Center of the Red Cross, modelers “are moving from forecasts of what the weather will look like to forecasts of what it will produce.”
However, the technology is still behind
Exascale supercomputers rely on both traditional computer chips and graphics processing units (GPUs), which are efficient at managing intensive computations.
GPUs are good for running model components in parallel and for training artificial intelligence algorithms – two techniques that Destination Earth will rely on to increase its performance.
But the old climate modeling code will have to be redone. ECMWF has a beginning: it adapts its forecast model to a GPU-based environment and last year tested it at a resolution of 1 kilometer for 4 months simulated using Summit, the American supercomputer that was the fastest in the world until a car Japanese eclipsed it recently.
The massive amount of data generated by the model will be a problem of its own. When the Japanese team conducted its 1-kilometer-scale experiment, it took half a year to extract something useful from a few days of data.
Destination Earth will solve much of this problem by designing ways to analyze the model’s results in real time.
As an operating system, Destination Earth is likely to run on multiple time scales, says Bauer.
One will be almost daily, targeting extreme weather events weeks or months into the future. It will also run to make long-term predictions, which will involve predictions within decades made at half-year periods.