Knowing Which Way the Wind Blows


Knowing Which Way the Wind Blows

Hurricanes and High Performance Computing

Hurricanes are my passion, but I am fascinated by all weather. When I was growing up, a family vacation to Myrtle Beach was threatened by a hurricane, and I spent the week watching the Weather Channel instead of lying on the rain-soaked sand. That event sparked my love of weather. I particularly enjoy using high-tech observations to reveal what can’t be seen with our eyes. Now I work as a weather modeling research scientist at NOAA’s Geophysical Fluid Dynamics Lab (GFDL). I support the Next Generation Global Prediction System (NGGPS) project which is an effort, largely motivated by Hurricane Sandy, to improve lead time for hurricanes and severe storms so that communities will be better prepared. I get to work with GFDL’s global weather model and run it on supercomputers. Every day, I get to tackle HPC and numerical weather prediction challenges. For example, right now I am working on HPC codes that won’t crash under the crush of bigger data sets from higher-resolution forecasts.

Hurricanes are my passion, but I am fascinated by all weather.

A Lot Can Happen in 13 Square Kilometers

NOAA is taking a big step forward and bringing groups of scientists together to develop the NGGPS (government, academia and industry). The new system will provide weather forecasts for the next few decades, and it requires moving into the challenging realm of cloud-resolving resolution. Today’s weather models are based on grids that are 13 square km. The NGGPS will be able to resolve the detail of regional models, while maintaining skill with the large-scale flow. Who knows? In a matter of years we may have global forecasts of 1-5 square km! Imagine the applications for pinpointing severe storms, anticipating the paths of hurricanes and staging resources well in advance of major weather events.

This means equations describing the motion of the atmosphere must be solved in smaller grid boxes. And some variables that were previously parameterized, like vertical motion, must now be solved for explicitly. This presents an increasing demand on HPC, and the NGGPS must be able to scale to very high processor counts.

Imagine the applications for pinpointing severe storms, anticipating the paths of hurricanes and staging resources well in advance of major weather events.
When we start talking about that kind of resolution, the vertical motion in the atmosphere is a lot more critical. Gaining granularity relies on HPC resources and the dynamical core (dycore), among several other components of the model. A dycore describes air movement and atmospheric behavior. NOAA and the National Weather Service had six candidate dycores, and GFDL’s FV3 Dycore was selected for its skill, ability to scale and its flexibility.

With the FV3 dycore comes the ability to create a more-refined look at weather systems using grid-stretching and two-way nesting capabilities. A nest, which is a domain of higher resolution embedded within the global model, allows for simultaneous, coupled regional and global solutions. FV3’s nested grid can use different time steps and physics parameterizations than the coarser grid. There is two-way interaction with wind and temperature fields. FV3’s nesting efficiency is 96.9 percent running with 3072 processor cores, while other non-uniform domain competition is at 64 percent. Engility Weather Protection Blog Caption: Grid nesting allows us to more than triple the resolution of the Hurricane Sandy forecast without tripling the cost. Notice the enhanced fine-scale structure of the clouds and the better placement of the low-pressure system in the North Atlantic - an important feature for steering Sandy back towards the U.S. The intensity forecast is also improved with the nest.

Grid nesting allows us to more than triple the resolution of a hurricane forecast without tripling the cost. That allows for better forecasting of where a hurricane will make landfall. The intensity forecast is also often improved with the higher resolution nest.

Modeling at cloud-resolving resolutions is even more important for thunderstorm forecasting. The timing and location (within a few hundred km) of a 2013 outbreak of tornadoes in Oklahoma showed up in the global forecast three days in advance.

By efficiently using HPC resources and taking advantage of the research and experience of the academic, government and private sector communities, we are partnering with NOAA to plan ahead for a unified weather-climate modeling system that will provide weather forecasts for the next generation—helping to protect lives and property from threatening weather events.

Recent Publications

Share this Post:

Posted by Shannon Rees

I received my B.S. in Environmental Science from the University of Delaware and my M.S. in Meteorology from the University of Hawaii at Manoa. My favorite “work” related memory was driving around Oahu in a Doppler on Wheels studying the trade wind showers and thunderstorms (if we were lucky) with radar observations.

I am also a member of SigHPC, AMS (American Meteorological Society), and AGU (American Geophysical Union).