Verne Global

HPC | Industry |

14 July 2017

How HPC is Helping the Future of Weather Prediction

Written by Nick Dale

Nick is Senior Director at Verne Global and leads our work across HPC and specifically its implementation within advanced manufacturing and meteorological research.

Weather plays a critical role in the function of the economy. In a recent white paper published by the Meteo Project, an insurance brokerage specialising in weather risk, weather variations in the US alone have a monetary impact to the economy measuring around $500 billion dollars per year, while in Europe the effect is around €400 billion euros annually. Another estimate by meteorological data firm Weather Analytics concludes that 33% of the world’s GDP is affected by the weather.

At the extreme end of this effect are catastrophic weather emergencies. When Hurricane Katrina hit the U.S. city of New Orleans, it caused over $100 billion dollar in damage and rendered 300,000 homes uninhabitable, making it one of the most well-known natural disasters in the country’s history. The effects of weather on the economy are felt in all regions though, even those that don’t experience frequent natural disaster. In snowy climates weather causes ten to fifteen percent of all catastrophic auto, home, and business insurance losses annually. In coastal areas, storms and natural disasters threaten human life and billions of dollars in property. How weather affects the economy has other, more complex dimensions as well, as the same snowstorm that increases heating and transportation costs may also bring skiers and tourist dollars. What is clear to both meteorologists and economists is that stronger predictive capabilities helps business and populations make more intelligent decisions and plans and about the weather.

The best method we have of predicting the weather, and protecting populations against its effects is numerical weather prediction (NWP). The process of NWP involves collecting massive amounts of input data from satellites, weather balloons, radar, and a variety of other sources, then using modeling software to apply algorithms to this data and predict the formation, intensity and movement of weather systems. Like other modeling applications NWP applications require the capability to process large datasets in as short a time as possible, and because of this they’ve benefitted greatly from the development and proliferation of HPC systems.

Hurricane Katrina’s devastating effect on New Orleans is a prime example of how HPC-enabled NWP can help to protect against extreme weather. European weather models accurately predicted the storm’s left turn toward the Mississippi River Delta and the city of New Orleans, while the American meteorological systems, which were modeling the storm at a much lower resolution, failed to make this key prediction. Many within the field of meteorology directly implicate the fateful lack of HPC infrastructure, which had deteriorated to one-tenth the power of counterpart systems in European, for the lack of preparedness on behalf of the federal government.

However, just as with other modeling advanced applications, NWP has a thirst for compute power that is hard to quench. Processing a single seven-day forecast currently requires approximately 2000 time-steps, with each time-step requiring trillions of computer operations. This means weather forecasting with current climate models consumes quadrillions of operations for a single forecast. As higher-resolution models continue to emerge that account for more variables, the computational requirements to run them will continue to increase.

This never-ending desire for better models means that meteorological organizations require the computational power of HPC systems that are both costly to build and operate. For example, the latest HPC system at the UK’s Met Office, one of the leading centers of meteorological study in the world, cost £97 million pounds. Similar systems at the National Oceanic and Atmospheric Administration in the U.S. have been built to recover capability and reputation after the embarrassment of Hurricanes Katrina and Sandy. Furthermore, these large NWP systems have power needs that measure in the megawatts, which further challenge the timely and efficient processing of meteorological data.

The costs of better numerical weather prediction are easily justified, though, as they not only safeguard human life, property, but can also help make businesses in a wide variety of industries more efficient and profitable. In the aviation industry, for example, it’s estimated that weather modeling applications save between 10 and 20 million tons of CO2 emission every year by helping planes use global wind currents to increase fuel efficiency. In the energy industry, U.S. utilities are able to save more than $150 million per year by using numerical weather prediction to more efficiently meet electricity demands. In agriculture, which is highly dependent on weather, the benefits of NWP could have a transformative effect across the developing world, where a poor growing season can lead to famine, mass migration, and negative economic growth.

Verne Global is dedicated to helping companies and organisations that do numerical weather forecasting ramp up their HPC operations to keep pace with the increasing need for NWP applications. By bringing more cost-effective, renewable sources of energy to our clients doing advanced weather analysis and prediction, Verne Global is able to help keep energy costs in check while ensuring that this vitally important work delivers consistent results and value to our clients and the greater population.


Sign up for the Verne Global newsletter

Opinion, thought leadership and news delivered directly to your inbox once a month.