In the pantheon of HPC grand challenges, weather forecasting and long term climate simulation rank right up there with the most complex and computationally demanding problems in astrophysics, aeronautics, fusion power, exotic materials, and earthquake prediction, to name just a few. This special report looks at how HPC takes on the challenge of global weather forecasting and climate research.
Modern weather prediction requires cooperation in the collection of observed data and sharing of forecasts output between all nations, a collaboration that has been going on for decades. This data is used to simulate effects on a range of scales – from events, such as the path of tornadoes, that change from minute to minute and move over distances measured in meters, to turnover of water layers in the ocean, a process that is measured in decades or even hundred of years, and spans thousands of miles.
The amount of data collected is staggering. Hundreds of thousands of surface stations including airborne radiosondes, ships and buoys, aircraft and dozens of weather satellites, are streaming terabytes of information every day. This data is interpolated to fit the three-dimensional grid that approximates the globe and over which the simulation is run to produce a forecast. The importance of this information cannot be understated, especially when it comes to anticipating, understanding and coping with weather and climate disasters. For example, according the National Centers for Environmental Informatics, there were eight U.S. weather and climate events in 2014 with losses exceeding $1 billion each, including droughts, tornadoes, flooding, and severe storms including a major winter storm event.
Because of the complexity involved, the length of the simulation period, and the amounts of data generated, weather prediction and climate modeling on a global basis requires some of the most powerful computers in the world. The models incorporate topography, winds, temperatures, radiation, gas emission, cloud forming, land and sea ice, vegetation, and more. However, although weather prediction and climate modeling make use of a common numerical methods, the items they compute differ.
Weather forecasting deals with short-term phenomena – will it rain tomorrow? From a compute standpoint, this requires very reliable systems that can produce focused results on-time, day after day. Climate modeling takes the long view; much of the work is done on massive systems in academic and research settings.
The governing equations of numerical weather prediction are what mathematicians call non-linear. This means that small changes in input data can result in big changes in the resulting prediction. For that reason some scientists resort to running “ensembles” – multiple runs that differ slightly in their input data, then weigh-average the results. Climate modeling uses multiple models to get an “average” prediction. This is one of the reasons why tomorrow’s forecast can be way off; yet the climate prediction – the average for the season – can be spot on.
Model ensemble forecasts have been used since 1992 to help define forecast uncertainty and extend the window in which numerical weather forecasting is viable into the future.
This insideHPC Special Report explores HPC’s impact on:
You can download the complete report from the insideHPC White Paper Library courtesy of SGI and Intel.