Computing giant IBM is getting into the weather modeling business. After buying the Weather Company — a.k.a. the Weather Channel — in 2016, it was only a matter of time before IBM combined supercomputers and weather forecasting.
Specifically, IBM and the Weather Company are collaborating with the National Center for Atmospheric Research to implement the MPAS model, or Model for Prediction Across Scales, on IBM’s new generation of supercomputers. The MPAS has been available for research and forecasting since 2013.
What makes this announcement a little more interesting the National Weather Service had the opportunity to use this model to replace the GFS, though NCAR pulled out of the selection process in the final hours. The selection committee ended up choosing a different model in July of last year — a decision one expert said was “a disaster for U.S. weather prediction.”
The weather community will find out which model is more valuable as both organizations work to implement them for large-scale weather prediction over the next couple of years.
Like its name suggests, the MPAS can be used at different scales. Global models tend to have a lower resolution because they need to cover so much area. Regional models, like the HRRR, have a higher resolution, but they don’t cover the globe. The MPAS can cover the globe at high resolution, particularly over areas of interest like the United States.
“In many ways, the strategy is to bring the value that is represented by the HRRR to the entire globe,” said Peter Neilley, director of weather forecasting sciences and technologies at the Weather Company.
In addition to the new high-resolution satellite observations that are coming online thanks to GOES-16, the group hopes to eventually feed “outside the box” data to the model to improve predictions. Observations from the “Internet of things” — intelligent cars, phones, buildings — are just some of the sources the Weather Company’s model could take advantage of.
Of course, IBM also brings its supercomputing power to the project. The model will be optimized to run on IBM’s POWER9-based system, which is expected to be available at the end of 2017, according to the company’s statement.
“IBM is one of only a few organizations in the world that has the capability to develop a model to run at this global, granular scale,” said Mary Glackin, head of weather science operations at the Weather Company.
That is a goal of the project, according to Neilley, who says typical weather models are inefficient, “using less than 5 percent of the theoretical power available.”
“Even small increases in that efficiency enables essentially ‘free’ extra computing power to be used for improving the weather forecast accuracy,” Neilley said.
“Collaboration across the private, public and research sectors that leverages collective skill and assets to advance meteorology is something we can all get behind,” the National Weather Service said in a statement.
This story has been updated to clarify that NCAR pulled the MPAS model out of the running for the GFS replacement before the committee selected the FV3.