During its three weeks of operation, the new GFS remained outclassed. On a standard measure—predicting the altitude at which the atmospheric pressure is half as great as at sea level—it still trails the ECMWF model.
Nonetheless, the GFS's strong showing during January's nor'easter offers solace to critics who feared America would never catch up with Europe in matters meteorological. Weather forecasting is fiendishly complex, and improvements tend to arise not from great leaps forward but rather an accumulation of incremental advances.
The ECMWF's most obvious advantage has been in raw computing power. Its Cray XC30 supercomputer can perform up to 2 quadrillion calculations a second, about ten times more than the GFS hardware before the recent upgrade. As a result, it carves up the Earth's atmosphere into svelte cells 16km square and 137 layers deep, compared with a bulky 27km and a mere 64 layers for the old GFS. The ECMWF's computing muscle also lets it start its projections with a replay of the past 12 hours of weather, using 40m data points derived from observations collected by ground stations, aeroplanes, balloons and satellites. In contrast, the GFS begins with a snapshot of a single moment.
The ECMWF also deserves credit for deploying its computational force wisely. The centre was a pioneer in using satellites to fill gaps in the data over the oceans, and in developing “ensemble forecasts” that generate a range of outcomes by employing slightly different starting conditions to produce multiple predictions. Its current model runs 52 such forecasts in parallel, each with a probability assigned to it.
Weather forecasters in America have full access to the ECMWF's model. However, the United States still has good reason not to free-ride on the Europeans' work. Private American firms have to pay for it, and the ECMWF is unlikely to develop regional or local models focused specifically on America. Moreover, giving the ECMWF a worthy competitor would probably lead to better forecasts overall.
The new GFS has certainly narrowed the gap. Its resolution is now 13km, though it still has only 64 layers. By November it is expected to run on a faster computer than the ECMWF's. It could be in line for further upgrades if the new, Republican Congress reintroduces the Weather Forecasting Improvement Act proposed last year—though the party's global-warming sceptics are likely to demand that much of the additional $120m a year the bill offered be taken away from research on climate change.
According to Cliff Mass, a professor of meteorology at the University of Washington, more money will not be enough to catch up with the Europeans. America, he says, must integrate its separate research and forecasting divisions, and include more contributions from non-government experts. Compared with pushing through cultural change in large public bureaucracies, predicting the weather is easy.