Datanami
EnterpriseAI
HPCwire Japan
QCwire
HPC & AI Wall Street
It’s hard to avoid news of global warming and climate change. At approximately 415 parts-per-million (ppm), many scientists believe that earth’s carbon dioxide (CO2) levels are as high as they’ve been since the Miocene era approximately 15 million years ago1. At this time, there were trees in Antarctica, equatorial regions were uninhabitable, and our earliest ancestors were not yet walking upright.
The earth’s changing climate may be the cause of severe local weather conditions in North America. In just the past weeks we’ve seen record flooding in Iowa, Illinois, Ontario and Quebec, a national disaster in Nebraska, a bomb cyclone in South Dakota, and a multi-day weather event that saw 52 tornados touch down across seven states putting 50 million people at risk. All this and it’s only May – hurricane season hasn’t even started yet.
Accurate forecasting becomes essential
With more frequent and severe weather events, accurate short- and medium-term weather forecasting will be critical. Aside from weather-related emergency alerts, airlines will need better forecasting to plan flight routes to avoid damaging turbulence and costs due to more frequent weather disruptions. Farmers will need to know when to plant, fertilize, and harvest to optimize crop yields. Utility companies will need better forecasts to help them pre-position assets to address increasingly severe storm damage to critical infrastructure.
Read: Weather Data Gives Red Bull Racing a Competitive Edge
How weather modeling works
Weather modeling is essentially a computational fluid dynamics (CFD) problem. Much like simulating airflow over an airplane wing, the basic idea behind weather forecasting is to know the current state of atmospheric conditions at a point in time. The atmosphere is divided into cells for calculation purposes, and complex equations governing fluid dynamics and thermodynamics are applied to each cell and their neighbors, stepping forward in time and predict the future state of atmospheric conditions. The better the model, and our understanding of initial atmospheric conditions, and the more computing power we can throw at the problem, the more accurate the forecast.
Advances in HPC dramatically improve forecasting
Starting in 2011, the National Oceanic and Atmospheric Administration (NOAA) massively increased investments in HPC, partnering with IBM to develop the Weather and Climate Operational Supercomputing System (WCOSS2). WCOSS is operated by the National Centers for Environmental Prediction (NCEP) Reston, VA and Orlando, FL. The HPC environment is comprised of two IBM Spectrum LSF clusters with over 5,000 nodes and 14 PB of storage managed by IBM Spectrum Scale. These clusters support over 500 users and process 3.5 billion observations and 15 million computer simulations daily.
Connect with users and showcase your expertise in the new IBM Spectrum LSF User Community.
The Global Forecast Model (GFM) created by NOAA continually runs forecast models 384 hours (16 days) in the future, running simulations at 13-kilometer cell resolution. Since the deployment of WCOSS, compute power at the National Weather Service has increased by a factor of 80 to over 8.4 PFLOPS today. While WCOSS couldn’t prevent the historic damage from Florence, Michael and six other named hurricanes in 2018 (resulting in damage estimated at $50B US), the increased forecast accuracy unquestionably saved many lives.
GRAF – IBM’s Global High-Resolution Atmospheric Forecasting System
In addition to these public-sector efforts, IBM has been investing in new weather forecasting technologies as well. In January of this year, IBM and The Weather Company (an IBM subsidiary) announced the IBM Global High-Resolution Atmospheric Forecasting System (GRAF). GRAF is comprised of 84 IBM Power System™ Accelerated Compute Server (AC922) nodes each with four NVIDIA V100 GPUs. Like other large-scale HPC clusters used in weather modeling, GRAF runs IBM Spectrum LSF and IBM Spectrum Scale.
Rather than modeling weather at 12-15km cell resolution, GRAF models weather over land at just 3km, a 200% improvement in forecast resolution. This allows GRAF to forecast localized events such as thunderstorms often missed by lower resolution forecasts. Also, GRAF can generate new forecasts hourly, providing consumers with more accurate and up to date forecasts of weather events. GRAF is also the first forecasting system able to tap sensor readings from aircraft in flight and crowd-sourced barometric readings from mobile device users who opt-in to sharing data from The Weather Network mobile app.
In addition to IBM’s R&D investments, these advances in weather forecasting are made possible by The Weather Company’s open-source collaboration with the National Center for Atmospheric Research (NCAR). GRAF incorporates the latest-generation global weather model – the Model for Prediction Across Scales (MPAS) developed by NCAR and Los Alamos National Labs.
As the effects of global warming become more pronounced in the coming years, innovations like GRAF enabled by leading-edge HPC systems will play an important role in helping protect lives and property and minimizing impacts from severe weather.
Learning more
To learn more about IBM weather solutions and how advanced weather-related insights from GRAF and other innovations can help your business visit https://www.ibm.com/weather.
References:
3: https://www.nextplatform.com/2018/01/11/noaa-weather-forecasts-stick-cpus-keep-eye-gpus/
