The challenge of predicting weather accurately
Weather forecasting in the United States and other developed nations is more accurate than in other more remote parts of the world. The Weather Company saw an opportunity to advance accurate weather predictions globally by leveraging the IBM GRAF.
The transformation of accurate weather forecasts
The Weather Company produces more precise and often more accurate weather forecasts by running higher resolution and more computationally intensive weather models on the latest technology that incorporates IBM Power System AC922 equipped with NVIDIA V100 GPUs.
IBM GRAF results
5xfaster running on IBM Power System AC922 with NVIDIA GPUs vs. x86-based servers
15 Hourweather predictions updated every 1 Hour
Increasingforecasting precision from 15km to 3km
Business challenge story
The art of weather forecasting
Weather affects every inhabitant on Earth, every day. It influences what people do, where people travel, what people eat, and even how people feel. More accurate weather forecasts help make more informed daily decisions and keep people out of danger.
In 2016, IBM acquired The Weather Company, a provider of hyperlocal weather forecasts. This acquisition brought together IBM’s powerful cognitive analytics capabilities and The Weather Company’s extremely high-volume data platform that ingests, processes, analyzes and distributes enormous data sets at scale in real time. Today, The Weather Company delivers around 25 billion forecasts daily and personalized weather data and insights accessible at your fingertips on various platforms like The Weather Channel app and weather.com. In addition, The Weather Company helps millions of consumers and businesses make better decisions as the leading provider of weather-driven business solutions to media, aviation, energy and utilities, insurance and government sectors.
The Weather Company, along with most meteorological services in the world, deploys numerical weather prediction (NWP) technology on supercomputers for weather forecasting. The Weather Company realized that with the addition of high-speed GPU technology they could accelerate weather prediction in order to generate global, high-resolution weather forecasts for the next 15 hours, updating those forecasts every hour. To achieve its goals of running more precise, complex and computationally intensive weather models, they required a supercomputer solution with ultimate speed and scale.
As Todd Hutchinson, head of Computational Meteorological Analysis and Prediction at The Weather Company, explained: “The Weather Company is looking to run the latest weather prediction models at high resolution over the entire world. Higher resolution gives more details in the forecast and resolves features that affect everyday weather, like thunderstorms, which are more difficult to forecast. To make this possible, The Weather Company needs computing capabilities that are cost effective from a capital, energy and data center footprint point-of-view to run these models which require significant computational resources.”
Transforming weather insights into actions
Redefining accurate weather predictions with IBM GRAF
The Weather Company was accustomed to running weather models on conventional x86-based clusters. Although, to continue raising the bar for weather predictions, they needed to build upon their 15+ years of regional weather modeling to expand towards global weather models. TWCo decided to implement the Model for Prediction Across Scales (MPAS) atmospheric model.
“Our goal is to improve our weather forecasts by running weather prediction models at higher resolution, more frequently, and globally. To do this, we are utilizing the hardware and software platforms that are available now from IBM and NVIDIA.” – Todd Hutchinson
Historically, weather models were developed on homogenous CPU-only systems, but as the industry embraces heterogeneous systems that take advantage of specialized accelerators to drive performance improvements, existing code needed to be ported over to GPUs to see those gains. These legacy codes and applications can be decades old, which created challenges in enabling them for GPUs. Since MPAS is a more recent model, some of these barriers were eliminated. By using OpenACC directives and other tools from PGI such as PCAST (PGI Compiler Assisted Software Testing) to diagnose and optimize MPAS, the team was able to successfully port all dynamics routines and an entire suite of physics parameterizations to enable acceleration of an entire weather forecast using GPUs.
To accelerate their capabilities, The Weather Company worked with IBM to design and purchase a high performance computing system based on IBM Power System AC922 servers. The Power System AC922 is the world’s best server for enterprise AI training, and the IBM POWER9 processor at the heart of the AC922 includes the industry’s only CPU-to-GPU NVIDIA NVLink interface, allowing the server to get up to 5.6x greater bandwidth between its incorporated NVIDIA V100 Tensor Core GPUs to deliver faster time to insights.
Todd Hutchinson: “Ultimately, there were a few reasons that we selected IBM Power Systems. First, while most weather models were built before the introduction of GPUs within supercomputing, the weather model that we’re working with (MPAS) is relatively new and was written by the National Center for Atmospheric Research (NCAR) using modern software standards. Researchers at the Computational Information Systems Laboratory at NCAR and the University of Wyoming were working to port MPAS to GPU using OpenACC. With additional support, they felt that they could port the entire weather model to GPU and gain a significant speedup as compared to running on CPU. With that effort ongoing between NCAR and The Weather Company and with the capability of GPUs on a cluster that IBM was able to build for us, it enabled us to move forward with IBM Power Systems.”
The results of having better weather data
Weather forecasting of the future
IBM Global High-Resolution Atmospheric Forecasting System (IBM GRAF)
In January 2019, TWCo revealed a new powerful global weather forecasting system would go live later in the year to provide the most precise local weather forecasts worldwide. The new IBM Global High-Resolution Atmospheric Forecasting System (IBM GRAF) will be the first hourly-updating weather system that is able to predict something as small as a thunderstorm anywhere on the planet. Compared to existing models, IBM GRAF will provide a 9x increase in forecast points across the world.
The new weather system, IBM GRAF, will become available Fall 2019. With the IBM Power System AC922 plus NVIDIA V100 GPUs, The Weather Company can make significantly more calculations within the weather model, and thus provide more frequent and accurate weather predictions globally, in locations that typically don’t have access to detailed forecasts.
As Todd Hutchinson mentioned, “While weather forecasting in some parts of the world such as the United States is quite accurate and timely, weather forecasting is often not nearly as precise or accurate in many other parts of the world. In most areas, weather models run at relatively coarse resolutions of 10-15km and update only once every 6 hours. Often, the latest weather forecast is based on information that is up to 10 hours old. With IBM GRAF, forecasts for most areas of the world will be fine-scale (3km resolution) and will be updated with the latest available data every hour. The benefits of IBM GRAF will be seen in the forecast for the coming 15 hours. So, for example in areas such as Africa, South America and much of Asia — in the morning we will have a much better opportunity to determine whether a particular location is likely to be affected by thunderstorms throughout the coming day.”
What’s next? The Weather Company will not only improve weather predictions to help people and communities better plan for upcoming weather conditions but will also help millions of industries such as retail, utility, aviation, insurance, airlines and others. In terms of airlines, IBM GRAF will be able to provide more effective routes around turbulence. As Todd Hutchinson explains: “We will diagnose turbulence using output from GRAF in order to provide forecasts of where turbulence is likely to occur, so that airlines have the opportunity to route around areas of expected turbulence.”
The Weather Company
The Weather Company is an IBM business – bringing together IBM’s advanced AI and cloud capabilities with The Weather Company’s high volume of weather data. This powerful combination helps people, businesses and communities around the world prepare for and mitigate the cost of weather. The company offers the most accurate weather forecasts globally–more than 25 billion per day–with personalized and actionable weather data and insights. The Weather Company is committed to trust and transparency, and The Weather Channel app and weather.com, as well as Weather Underground and wunderground.com are trusted by hundreds of millions of people to provide accurate, timely forecasts that help them make critical decisions every day. For more, visit https://newsroom.ibm.com/the-weather-company
Take the next step with the IBM GRAF weather model
To learn more about IBM Power System AC922 with NVIDIA GPUs, please contact your IBM representative or IBM Business Partner, or visit the following website: https://www.ibm.com/us-en/marketplace/power-systems-ac922. Or, read more client success stories to learn how others have put IBM’s solutions to use and got major results.