• Subscribe

Measuring the storm

Speed read
  • Weather forecasting is especially difficult for mountainous regions like Switzerland
  • Accurately modeling entire weather systems demands large computing resources
  • MeteoSwiss’s COSMO forecasting model is advancing prediction capabilities

Weather prediction is notoriously unreliable. Switzerland is a prime example of why studying atmospheric conditions can be so hard. The country’s many high mountains and deep valleys can quickly change the course of weather patterns. It was out of this hardship that the Consortium for Small-scale Modeling (COSMO) was born.

Accelerated weather. By changing from CPUs to GPUs, MeteoSwiss scientists were able to increase the resolution of the COSMO weather prediction model and provide more accurate forecasts. Courtesy NVIDIA.

MeteoSwiss, a part of the Swiss Federal Department of Home Affairs, along with several national weather services, developed COSMO which is a numerical forecasting model that uses high-performance computing (HPC) resources to simulate weather systems. To learn more about COSMO, we interviewed Xavier Lapillonne, a senior scientist at MeteoSwiss who presented a talk on the system at ISC High Performance 2018.

Making a change

Initially founded in 1998, COSMO was developed as a non-hydrostatic atmospheric model, which means it doesn’t filter out certain atmospheric conditions. However, technological advances have allowed this system’s creators to go far beyond that. A big change came with the decision to switch to GPUs.

“We’d been working in CPUs, and we decided to investigate how we could use GPUs,” Lapillonne says.

<strong>High mountains and deep valleys.</strong> Switzerland’s extreme topography means sudden changes in weather patterns. It takes an HPC numerical forecasting model to keep up. Courtesy Kabelleger/David Gubler/(banbilder.ch). <a href='https://creativecommons.org/licenses/by-sa/3.0/deed.en'> (CC BY-SA 3.0)</a>“The model is rather large at about 350,000 lines of code of Fortran 90, so it’s not so easy at first to adapt for GPUs. On the one hand, we used compiler directives for a large part of the model, but since there are some limitations of types of preference, we also used another approach for the most time-consuming part. Then we developed domain-specific language (DSL) which we used to run the dynamic.”

Improving COSMO’s HPC infrastructure like this is essential for increasing the model’s resolution. Weather prediction systems need to be able to map the local area as accurately as possible, which is a huge challenge in mountainous regions like Switzerland.

“We still are running a rather coarse grid, so the highest we have is one kilometer, and it is already considered very high resolution,” Lapillonne says.

“Some models are running two or four kilometers and, even at this resolution, there are some effects of the dynamics which cannot be resolved. The turbulence is not completely resolved, and for this we need to use ad-hoc models. This is called physical parameterization and this introduces more uncertainty into the model.”

The high-resolution model that Lapillonne is referring to is COSMO-1. With a resolution of 1.1 kilometers, COSMO-1 is the most precise forecasting model in the Alpine region. It also has a ceiling of 4,268 meters above sea level for a full view of some of the highest atmospheric conditions, which demands a rendering of 80 vertical layers.

“By going to a higher resolution, we get closer to the physical equation,” Lapillonne says. “The other important aspect is the resolution of the orography. Mountains have a big impact on the weather, so if you capture the mountain and the valleys, the wind system is better resolved.”

COSMO-1 is able to predict the weather at 10-second intervals, which enables consistently accurate data. What’s more, the public has access to real time forecasts on the MeteoSwiss website. COSMO-1 gives forecasts for the current and next day.

Early warnings

In a mountainous region with a temperamental climate, staying on top of weather patterns helps keep people safe. You also need a little more advance warning than COSMO-1 can offer.

Enter COSMO-E, an ensemble model that calculates the probability of a weather event occurring within the next five days. With a resolution of 2.2 kilometers, COSMO-E is less precise than COSMO-1, but this decreased clarity can still help citizens prepare for dangerous weather. 

<strong>Off the rails.</strong> MeteoSwiss alerted populations about heavy rainfall the day before a landslide that affected Alpine railways and injured 11 people. Without the warning, damage may have been worse. Courtesy Medienbericht der Kantonspolizei Graubünden.“One of the success stories of what we got [when] going to GPUs is that we could put emphasis on ensemble forecasting systems,” says Lapillonne. “This enabled us to give probabilistic information for when we alert the population or the authorities about extreme events. For a landslide in August 2014, it was possible to alert the population about strong rainfalls.”

MeteoSwiss used COSMO-E to put out a level 3 (significant hazard) warning the day before the landslide, which ultimately affected Alpine railways and injured 11 people.

While MeteoSwiss provides forecasting information for Switzerland, the COSMO model continues to be developed in collaboration with other countries. Germany, Greece, Israel, Italy, Poland, Romania, and Russia are all coming together under COSMO in order to improve forecasting efforts for the future.

“For us, it’s exciting and challenging because, with exascale, we don’t know exactly what type of chip we are going to have, or how we are going to adapt to a model,” says Lapillonne.

“We think we will have to face more and more parallelism. We will have to do even more adaptation of the programming model to be able to catch the trends. So far it’s really challenging to work on this and adapt the model for this next generation of systems.”

Forecasting systems may not be perfect right now, but with models like COSMO, the world’s weather is quickly becoming a little more predictable.

Read more:


Join the conversation

Do you have story ideas or something to contribute? Let us know!

Copyright © 2023 Science Node ™  |  Privacy Notice  |  Sitemap

Disclaimer: While Science Node ™ does its best to provide complete and up-to-date information, it does not warrant that the information is error-free and disclaims all liability with respect to results from the use of the information.


We encourage you to republish this article online and in print, it’s free under our creative commons attribution license, but please follow some simple guidelines:
  1. You have to credit our authors.
  2. You have to credit ScienceNode.org — where possible include our logo with a link back to the original article.
  3. You can simply run the first few lines of the article and then add: “Read the full article on ScienceNode.org” containing a link back to the original article.
  4. The easiest way to get the article on your site is to embed the code below.