In the world of weather forecasting, the battle for supremacy is generally being waged between the Euro and American weather models maintained by U.S. and European Union governmental agencies. But there are tons of other options to choose from, and a new one emerged on Tuesday at the Consumer Electronics Show (CES) in Las Vegas.
IBM and its subsidiary The Weather Company (TWC) have created a model that aims to offer a wildly detailed view of weather around the world. If proven accurate, the model could change forecasting, particularly in the developing world where a paucity of data makes good predictions hard. But it could also invite controversy by relying on cell phone data from the Weather Channel app, which the city of Los Angeles recently sued over for misleading users about how much data it collects.
The new model, dubbed the Global High-Resolution Atmospheric Forecasting System (or the oddly-acronymed GRAF), has a couple of things going for it, but the most impressive is its resolution. The model builds off of MPAS model framework that was created by the federally-funded National Center for Atmospheric Research. Using that framework and powerful supercomputers (more on that in a sec), GRAF spits out forecasts at 3-kilometer resolution in places like the U.S. with a plethora of data. In the developing world where data is less available, that resolution scales to 15 kilometers. GRAF also updates every hour.
In comparison, the European model offers a flat 9-kilometer resolution while the American model runs at 13 kilometers. These models are only updated a few times a day.
“One of the goals of... running at higher resolution is to bring standards for weather forecasting [in the developing world] up to other standards in the rest of the world,” Todd Hutchinson, a computational meteorological analysis and prediction lead for TWC/IBM, told Earther.
The super-high resolution allows the model to resolve intimate details in the atmosphere, showing not just a big storm coming but singular thunderstorm cells embedded in it. Having that type of information could be useful for a host of people from emergency managers preparing for a tornado outbreak to air traffic controllers worried about turbulence.
But higher resolution doesn’t immediately guarantee a better forecast. The model has to, you know, actually get the weather right. Hutchinson said GRAF has run relatively smoothly on the whole, but that there are still challenges in forecasting storms where the model segues from 3-kilometer to 15-kilometer resolution. That happens gradually, and the shifting algorithms from one area to the next are complex conga line that will hopefully go smoother as Hutchinson and his team gather more model runs to compare to the actual weather happening in the real world.
None of this would be possible without massive computing power. The system runs on a supercomputer setup similar to Department of Energy’s Summit and Sierra supercomputers, which are among the beefiest in the world (although it’s not necessarily as fast). To process all the data, the system has 3.5 petabytes of elastic storage, equivalent to roughly 55,000 64-gigabyte iPhones.
The reason all this computational power is needed is that in addition to traditional forms of weather data from balloons and weather station observations around the world, GRAF also draws on atmospheric data collected by planes and ground pressure data from millions of cell phones that have the Weather Channel app.
Most smartphones have a barometer inside, which is used to help your phone’s fitness tracker figure out how many stairs you climbed. It’s been a holy grail for meteorologists for years, because pressure provides clues about storms. But while accessing a network of tiny pressure sensors opens up a world of forecast improvements, it also raises privacy concerns, especially in this instance. Just last week, the Weather Channel app came under fire for allegedly deceiving users and collecting a ton of their data for TWC to profit off of.
At CES, IBM said users can opt-in to provide this new type of data collection once the model gets ramped up later in 2019. But it’s yet another reminder of the tradeoffs users increasingly face as they weigh privacy versus whether they should grab that umbrella before running out the door.