Swiss First to Tap GPUs to Improve National Weather Forecasts

by Roy Kim

Ten years ago, Hurricane Katrina devastated New Orleans. Three years ago, Hurricane Sandy battered New York City. Hundreds lost their lives. Damages were in the billions.

Wherever you live, predicting the weather is a high-stakes game.

Now, thanks to GPU-accelerated computing, the Swiss have made significant advancements in their ability to predict storms and other weather hazards with higher levels of accuracy.

The Swiss Federal Office of Meteorology and Climatology, MeteoSwiss, is the first major national weather service to deploy a GPU-accelerated supercomputer to improve its daily weather forecasts.

Our forecast: They won’t be the last. Accurate weather forecasting requires vast amounts of computing performance. Unleashing that computing horsepower—without sucking power grids dry—is an ideal fit for GPUs.

A typical COSMO simulation, showing cloud patterns over Switzerland and surrounding areas (Image by NVIDIA / MeteoSwiss)

The new system provides 40 times higher performance than the CPU-based system it’s replacing. It allows MeteoSwiss to develop weather models with more than two times higher resolution—and at three times higher energy efficiency.

Higher Resolution Weather Models = More Accurate Forecasts

One key to improving weather forecasting is increasing the resolution of computer-generated weather prediction models.

A weather model samples the state of the atmosphere at a given time, and uses fluid motion and thermodynamics equations to predict the state of the atmosphere at some time in the future.

The model divides a forecast region into a grid, and the equations are solved within each grid cell with interactions between the neighboring cells to compute a prediction. The closer grid points are to one another, the higher the overall model resolution, which leads to increased realism in the final forecast.

Running weather models configured at 1km resolution is a milestone weather services worldwide hope to achieve, and the new MeteoSwiss system does just that.

Enabling the agency to increase weather model resolution for short-range (24 hour) forecasts from 2.2km to 1.1km resolution, the powerful new system lets MeteoSwiss forecast the amount, duration and location of rain and snow in more detail, and provide early warning forecasts for severe weather events.

The new system also lets MeteoSwiss increase resolution of its medium-range forecasts three times higher, and extend the range from three to five days.

The new MeteoSwiss system is based on the Cray CS-Storm cluster supercomputer, one of the highest density GPU-accelerated systems available today. Expected to be fully operational by next summer, the MeteoSwiss system is housed at the Swiss National Supercomputing Centre in Lugano, Switzerland. It’s powered by NVIDIA Tesla K80 GPU accelerators — the flagship of our Tesla Accelerated Computing Platform.

Widely Used COSMO Weather Model Now GPU Accelerated

Daily weather forecasts are performed on high-performance supercomputers using numerical weather prediction models.

MeteoSwiss will be the first to deploy the GPU-accelerated version of the COSMO model, which is also extensively used by other national weather services in Germany, Italy, Greece, Poland, Romania and Russia, and for regional climate studies at more than 70 research institutes.

The COSMO consortium is working to make the GPU-accelerated version available to these weather services and other consortium members.

So pencil this into your forecast: within a few years, a GPU-accelerated supercomputer may be predicting your weather, too.