From Ars Technica:
TV maker Panasonic says it has developed the world’s best weather model
For a long time, weather forecasting was largely the domain of governments, with the National Weather Service leading the way in the United States. During the last two decades, however, the private weather forecasting industry—which includes well-known companies such as AccuWeather and many hundreds of smaller businesses—has grown up and is now estimated to be worth as much as $6 billion.
Even so, these companies largely just repackage model forecasts and incorporate data from government weather agencies. The heavy lifting, in the form of sophisticated computer models and the processing power to run them, is still done by the large ECMWF center in Europe, NOAA's National Centers for Environmental Prediction in the United States, and similar forecasting services in Canada, Japan, and elsewhere.
All of these government agencies, with staff generally in the hundreds of scientists, forecasters, and coders, collect weather data from a variety of sources, including shared satellite data along with in-situ observations from aircraft, balloon soundings, surface observations, and other sources. Then, using their own software, they "assimilate" these observations into physics-based forecast models that simulate weather conditions around the globe at various levels of the atmosphere over a 10- to 16-day period. The results from these global models are then used to predict things like the high temperature on Wednesday and whether it is likely to rain on Saturday afternoon for little Joey's birthday party.
For much of the 21st century, to the chagrin of US scientists and forecasters, the American GFS model—Global Forecasting System—has lagged behind the European ECMWF output, especially in things like five- and seven-day forecasts and tropical cyclone track forecasting. This is largely due to the fact that the ECMWF has typically had more dedicated supercomputers to run their forecast models and a much more sophisticated system to get more—and more timely—data into its forecast model as initial conditions.
And the numbers:
Large, multinational electronics company Panasonic now wants to crash the party. In an exclusive interview with Ars, Neil Jacobs, the chief scientist for Panasonic Weather Solutions, said the company has been running its own global model for several years on an 11,000-core supercomputer. And that PWS model, he said, has not only been outperforming the GFS model but has become competitive with the gold-standard ECMWF model. "We started the global model development in 2008 and finally got to the point where we were outperforming ECMWF by late last year," Jacobs said.
There are various ways to measure model accuracy, but one of the most widely recognized is "anomaly correlation" at the 500mb, or mid-level of the atmosphere, over a 30-day average. Higher scores are better. Recently, Jacobs said, the PWS model has scored a .926, compared to a .923 for the ECMWF and .908 for the GFS. Essentially, then, a team of five weather modelers and five software engineers, as well as about 20 meteorologists and computer modeling experts at universities under contract, claims to have beaten the best government weather forecasting centers in the world.
A lot of very clever thinking and one spectacular purchase - they have insturmentation on commercial airplanes that gives real-time data (3,500 readings per day over the world) for their model.
Local weather guy Cliff Mass talks about the ECMWF in several posts - our failure basically comes down to poor computers available for forecasters. Because the European model runs on much larger computers, they are able to incorporate a lot more data. Note that the Panasonic system uses 11,000 cores. In one of Cliff's articles from 2012, he said at that time that our National Weather Service was using two IBM supercomputers with 4,992 cores each - a far cry from 11,000. Doubling the horizontal resolution requires eight times more computing power.
We have a long way to come but it is very cool that independent corporations are seeing the market share and responding well.
Leave a comment