The Washington PostDemocracy Dies in Darkness

Trump administration has EPIC plan to develop the world’s smartest weather forecasting model

NASA visualization of Hurricane Sandy. (NASA)

It was October 2012 when the European weather prediction model beat its American counterpart in forecasting Hurricane Sandy’s hard left turn into the U.S. coastline. What scientists had known for years — that the European forecast model was superior to the American — caught the attention of the U.S. public and Congress.

Since then, the National Oceanic and Atmospheric Administration, with funding support from Congress, has worked intensely to improve the American model. It has boosted its computing power, improved the way it brings in data, and enhanced how it simulates weather systems at small scales. Yet, more than six years later, it still trails the European model in overall accuracy.

Neil Jacobs, the acting head of NOAA and a meteorologist, is committed to closing the gap between the models. Since being appointed to the Trump administration, he has made one of his top priorities installing a process that will allow U.S. forecast modeling to reach its potential and become world-class.

As part of its 2020 budget request, to the tune of $15 million, NOAA has proposed the establishment of the Earth Prediction Innovation Center (EPIC), which it says “will advance U.S. weather modeling and reclaim international leadership in the area of numerical weather prediction.”

In an interview, Jacobs blamed recent U.S. modeling shortfalls on a lack of research investment. He said the United States now spends about the same amount on operating its flagship model, the Global Forecast System (GFS), as it does on research initiatives to improve it. By contrast, the European Center for Medium-Range Weather Forecasts spends roughly five times as much on research. Jacobs said he’d like to see NOAA “grow research five times” to keep pace.

This is where EPIC comes in.

Whereas historically it has been difficult for the outside research partners to test, evaluate and provide feedback on the American modeling system, EPIC will provide the framework for a “community modeling” effort. This, Jacobs said, will break down the barrier between NOAA’s modeling and the outside world and accelerate model development.

The code for the GFS, Jacobs said, will be open-source and available to anyone. EPIC will provide training and support so outside scientists can become comfortable running and adapting the model for their research efforts.

“My hope is that when we throw this code out there, university researchers, principal investigators, post-docs and industry scientists will download the code and start to run it,” Jacobs said. “They’ll help debug [the model], improve its physics and data assimilation, and we’ll benefit from their research.”

Jacobs admitted that this community approach will present some challenges. He said that, as the current version of the code isn’t user-friendly, it will need to be made more accessible and “hardware agnostic” to run on different systems.

But he was upbeat about its potential benefits, saying EPIC would use cloud computing so multiple users could run the model simultaneously. In the past, bottlenecks could arise because the model was operating in a single computing environment.

If users identify a model error, “we’ll have the ability to stand up parallel versions of the model, do tests and determine month in and month out how proposed updates improve the situation,” Jacobs said.

Operating the model in the cloud, he stressed, will offer both NOAA and its users tremendous flexibility considering the increasing number of vendors and decreasing costs for cloud computing.

As proposed, EPIC will be a virtual center with no physical location but operated from existing NOAA modeling hubs. Jacobs said he expects program funding to support work on the model through contracts and grants awarded to partners in academia and research labs.

Cliff Mass, a professor of atmospheric science at the University of Washington and longtime critic of how NOAA has managed its modeling efforts, wrote in a blog post that he is “really optimistic” about EPIC.

“Done correctly, EPIC could lead to a much more effective and coordinated approach to developing a new U.S. global modeling capability,” he wrote. “A modular, unified national modeling system shared between government, academia, and the private sector. Will the U.S. FINALLY organize itself properly to regain leadership in global numerical weather prediction? Time will tell. But I am more optimistic today than I have been in years.”

The Trump administration’s 2020 proposed budget for NOAA, including support for EPIC, is now in the hands of Congress.